Greptile vs cubic

Agent access

1. ai-native userPoint an agent at llms.txt or agent-oriented docs

   weight 2 · round to cubic
   Greptile✓fullprobed8/10

   Greptile has a confirmed llms.txt at the root (HTTP 200) plus per-page .md variants (e.g. docs/introduction.md) and a documentation index at docs/llms.txt, explicitly designed for agents to fetch; docs also explicitly invite handing pages to a coding agent to run setup. Missing for 10: independent/community confirmation that an external agent successfully consumed llms.txt end-to-end, and no evidence of an OpenAPI spec to complement it.

   • [probe] “PROBE llms.txt: HTTP 200 at https://www.greptile.com/llms.txt # Greptile > Greptile is an independent AI code validation platform for GitHu…”
   • [probe] “PROBE docs-md: HTTP 200 at https://www.greptile.com/docs/introduction.md > ## Documentation Index > Fetch the complete documentation index a…”
   • [claimed-docs] “Set up Greptile from your terminal with `greptile onboard` — or hand this page to your coding agent and have it run the setup for you.”

   proof ↗⚑ flag

   cubic✓fullprobed9/10

   cubic hosts a live llms.txt (HTTP 200) and .md-suffixed agent-friendly docs pages that link back to the llms.txt index, confirming an agent can be pointed at these docs and navigate them. missing for 10: no independent/community confirmation of an agent actually consuming llms.txt successfully in practice.

   • [probe] “PROBE llms.txt: HTTP 200 at https://docs.cubic.dev/llms.txt # cubic documentation > cubic reviews code on GitHub and in local coding workfl…”
   • [probe] “PROBE docs-md: HTTP 200 at https://docs.cubic.dev/ai-review/introduction.md > ## Documentation Index > Fetch the complete documentation inde…”

   proof ↗⚑ flag

2. ai-native userRun the product headlessly / in CI for automation

   weight 2 · round to Greptile
   Greptile✓fullprobed8/10

   Greptile ships an official CLI (`greptile review`, `greptile onboard`) that runs headlessly against a diff in ~60 seconds and is explicitly designed for coding agents to invoke on their own code, plus automated PR-triggered reviews on GitHub/GitLab that require no human interaction — both classic CI/automation patterns. missing for 10: explicit CI pipeline example (e.g., GitHub Actions/GitLab CI YAML snippet) and independent third-party confirmation of headless CI usage

   • [claimed-docs] “Run `greptile review`, get a review of the feature branch diff in 60 seconds.”
   • [claimed-docs] “Coding agents can use the greptile cli to review their own code”
   • [claimed-docs] “Set up Greptile AI code reviews in 5 minutes. Connect GitHub or GitLab, configure review triggers, and get automated feedback on your first …”
   • [claimed-docs] “`greptile onboard` creates your organization, connects GitHub or GitLab, enables repositories, and imports your existing AI rules files — th…”
   • [probe] “official CLI documented at https://www.greptile.com/docs/code-review/greptile-cli”

   proof ↗⚑ flag

   cubic~partialprobed6/10

   cubic's core review flow is already headless — it auto-reviews PRs on open with no manual step (cubic-docs-1, cubic-docs-47), and the cubic CLI can review local changes before a push (cubic-docs-7, cubic-probe-5), which is plausible to script into CI. However there is no explicit documentation of a CI/pipeline integration mode, exit codes for pass/fail gating, or GitHub Actions/other CI examples. Missing for 10: explicit CI pipeline integration docs, non-interactive/exit-code behavior for automation gating, and independent confirmation of CLI use in CI.

   • [claimed-docs] “Once installed, cubic automatically reviews new pull requests.”
   • [claimed-docs] “The **cubic CLI** reviews local changes before you push. It finds bugs and generates a prompt that your coding agent can use to fix them.”
   • [probe] “official CLI documented at https://docs.cubic.dev/ide/cli-review”
   • [claimed-docs] “The AI automatically scans every pull request for a wide range of issues, including logic errors, style inconsistencies, and potential secur…”
   • [claimed-docs] “Codebase scans deploy thousands of AI agents to find bugs and vulnerabilities across your repository.”

   proof ↗⚑ flag

3. ai-native userPlug MCP servers into this product so it can use their tools

   weight 3 · round drawn
   Greptile—none0/10

   All MCP-related evidence describes Greptile exposing its own MCP server so other tools (Cursor, Claude Code, VS Code, Codex) can pull Greptile's review data and fixes (docs-23, docs-24, probe-4) — this is Greptile acting as the MCP server, not as a client that consumes external MCP servers' tools. There is no evidence Greptile itself can be configured to plug in and use third-party MCP servers/tools.

   • [claimed-docs] “Fetch PR comments - Get unaddressed Greptile feedback for any PR”
   • [claimed-docs] “Connect Cursor, Claude Code, VS Code, or Codex to Greptile MCP with OAuth.”
   • [probe] “official MCP server documented at https://www.greptile.com/docs/mcp-v2/overview”

   proof ↗⚑ flag

   cubic—none0/10

   Cubic's MCP-related evidence is all about cubic exposing its own MCP server for other coding agents to consume (cubic-docs-8, cubic-probe-4) — the reverse of this story, which asks whether the AI-native user can plug external MCP servers into cubic so cubic can use their tools. No evidence shows cubic acting as an MCP client consuming third-party MCP servers; integrations like Jira/Linear/Notion (cubic-docs-37, cubic-docs-61) appear to be native integrations, not MCP-based tool consumption.

   • [claimed-docs] “Connect cubic's MCP server to your coding agent to read review findings and codebase context, request PR reviews, and triage PR or codebase …”
   • [probe] “official MCP server documented at https://docs.cubic.dev/ide/mcp-server”
   • [claimed-docs] “Connect to Jira, Linear, Asana, and Notion”
   • [claimed-docs] “Add page context from Notion to make AI reviews understand your requirements and generate accurate PR descriptions.”

   proof ↗⚑ flag

4. ai-native userConnect an agent via an official MCP server

   weight 3 · round drawn
   Greptile✓fullprobed8/10

   Greptile documents an official MCP server (mcp-v2) with OAuth-based connection for Cursor, Claude Code, VS Code, and Codex, exposing tools to fetch PR comments, apply fixes, and manage patterns from the agent's IDE. This is well-documented first-party functionality with multiple docs pages (overview, setup, auto-fix, custom-context) and a probe confirming the endpoint exists. Missing for 10: independent/hands-on community verification that the MCP server works as described, and a fuller list of exposed tools/capabilities beyond the docs summaries.

   • [claimed-docs] “Fetch PR comments - Get unaddressed Greptile feedback for any PR”
   • [claimed-docs] “Connect Cursor, Claude Code, VS Code, or Codex to Greptile MCP with OAuth.”
   • [claimed-docs] “List unaddressed Greptile comments for PR #5 in owner/repo”
   • [claimed-docs] “With MCP, you can view, search, and create patterns from your IDE.”
   • [probe] “official MCP server documented at https://www.greptile.com/docs/mcp-v2/overview”

   proof ↗⚑ flag

   cubic✓fullprobed8/10

   cubic documents an official MCP server that lets coding agents read review findings/codebase context, request PR reviews, triage issues, and even manage subscription/seats without leaving the MCP client, with a dedicated docs page (cubic-probe-4) and integration guidance across Cursor, Claude Code, VS Code, Codex, Gemini CLI, etc. missing for 10: independent/hands-on verification that the MCP server works reliably in practice beyond first-party docs.

   • [claimed-docs] “Connect cubic's MCP server to your coding agent to read review findings and codebase context, request PR reviews, and triage PR or codebase …”
   • [claimed-docs] “You can now ask your coding agent to check your cubic subscription, manage team seats and roles, and purchase more seats without leaving you…”
   • [claimed-docs] “Install and connect cubic in Cursor, Claude Code, VS Code, Codex, Gemini CLI, or another coding tool.”
   • [probe] “official MCP server documented at https://docs.cubic.dev/ide/mcp-server”

   proof ↗⚑ flag

5. ai-native userUse an official CLI

   weight 2 · round drawn
   Greptile✓fullprobed8/10

   Greptile documents an official CLI (`greptile onboard`, `greptile review`) that sets up the org and runs full PR reviews with confidence scores, diagrams, and test generation in ~60 seconds, and explicitly positions it for agentic use ('Coding agents can use the greptile cli to review their own code'). This is well-documented first-party functionality with a dedicated docs page and probe confirmation. Missing for 10: independent/hands-on community verification specifically of the CLI (community evidence covers the web product, not CLI usage) and no public API/OpenAPI spec surfaced for deeper CLI-adjacent automation.

   • [claimed-docs] “Set up Greptile from your terminal with `greptile onboard` — or hand this page to your coding agent and have it run the setup for you.”
   • [claimed-docs] “Run `greptile review`, get a review of the feature branch diff in 60 seconds.”
   • [claimed-docs] “Coding agents can use the greptile cli to review their own code”
   • [claimed-docs] “`greptile onboard` creates your organization, connects GitHub or GitLab, enables repositories, and imports your existing AI rules files — th…”
   • [probe] “official CLI documented at https://www.greptile.com/docs/code-review/greptile-cli”

   proof ↗⚑ flag

   cubic✓fullprobed8/10

   cubic ships an official CLI (documented at /ide/cli-review) that reviews local changes before push, finds bugs, and generates prompts for coding agents to fix them, fitting the ai-native/agentic workflow theme. Missing for 10: independent hands-on verification of the CLI's usage/UX beyond first-party docs, and more detail on CLI command surface/options.

   • [claimed-docs] “The **cubic CLI** reviews local changes before you push. It finds bugs and generates a prompt that your coding agent can use to fix them.”
   • [probe] “official CLI documented at https://docs.cubic.dev/ide/cli-review”

   proof ↗⚑ flag

6. ai-native userDrive the product through a documented public API

   weight 3 · round drawn
   Greptile~partialprobed6/10

   Greptile exposes a documented MCP v2 server (fetch PR comments, apply fixes, manage patterns) and a CLI (`greptile review`, `greptile onboard`) that let an AI agent drive the product programmatically, which functions as a de-facto public API for agentic use, but explicit REST/OpenAPI documentation was probed and returned 404s, meaning no conventional public API spec is confirmed. missing for 10: a documented REST/OpenAPI-style public API, independent confirmation that the MCP/CLI interfaces are being used as a full 'public API' by third parties.

   • [claimed-docs] “Fetch PR comments - Get unaddressed Greptile feedback for any PR”
   • [claimed-docs] “Connect Cursor, Claude Code, VS Code, or Codex to Greptile MCP with OAuth.”
   • [claimed-docs] “List unaddressed Greptile comments for PR #5 in owner/repo”
   • [claimed-docs] “With MCP, you can view, search, and create patterns from your IDE.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.greptile.com/openapi.json, https://www.greptile.com/swagger.json, https://www.greptile.c…”
   • [probe] “official MCP server documented at https://www.greptile.com/docs/mcp-v2/overview”
   • [probe] “official CLI documented at https://www.greptile.com/docs/code-review/greptile-cli”
   • [claimed-docs] “Run `greptile review`, get a review of the feature branch diff in 60 seconds.”
   • [claimed-docs] “Coding agents can use the greptile cli to review their own code”

   proof ↗⚑ flag

   cubic~partialprobed6/10

   cubic exposes some programmatic surfaces — a documented Analytics API for PR-level metrics (cubic-docs-16), an official MCP server for agents to request reviews/triage issues (cubic-docs-8, cubic-probe-4), and a CLI for local review (cubic-probe-5) — which together let an AI-native user drive parts of the product programmatically. However, a probe for a general public REST/OpenAPI spec returned 404 on all standard paths, indicating no comprehensive documented public API beyond these narrower interfaces. Missing for 10: a full OpenAPI/REST API spec covering core product actions (not just analytics), independent hands-on verification of the Analytics API and MCP server working as documented.

   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”
   • [claimed-docs] “Connect cubic's MCP server to your coding agent to read review findings and codebase context, request PR reviews, and triage PR or codebase …”
   • [probe] “official MCP server documented at https://docs.cubic.dev/ide/mcp-server”
   • [probe] “official CLI documented at https://docs.cubic.dev/ide/cli-review”
   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.cubic.dev/openapi.json, https://docs.cubic.dev/swagger.json, https://docs.cubic.dev/api…”

   proof ↗⚑ flag

7. ai-native userIssue scoped/least-privilege API credentials for an agent

   weight 2 · round drawn
   Greptile—none0/10

   No evidence of scoped or least-privilege API credential/token issuance for agents; MCP setup mentions OAuth connection but nothing about scoping permissions or credential granularity. Missing for 10: any documentation of API key scopes, permission levels, or least-privilege token issuance for agent access.

   ⚑ flag

   cubic—none0/10

   cubic documents role-based access control for human team members managing subscriptions/settings (cubic-docs-56) and offers an MCP server/Analytics API, but there is no evidence of any mechanism to issue scoped or least-privilege API credentials/tokens specifically for an agent's use.

   • [claimed-docs] “cubic uses a role-based access control system to manage who can make changes to your team's subscription and settings. Every team member has…”
   • [claimed-docs] “Connect cubic's MCP server to your coding agent to read review findings and codebase context, request PR reviews, and triage PR or codebase …”
   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”

   proof ↗⚑ flag

8. ai-native userBuild against official SDKs

   weight 2 · round drawn
   Greptile—none0/10

   Evidence shows Greptile offers a CLI and an MCP server for agent integration, but no official SDKs (client libraries) are documented anywhere, and the openapi probe returned 404 for all candidate API-spec paths, indicating no public API/SDK surface to build against.

   • [probe] “PROBE openapi: all candidate paths 404 (https://www.greptile.com/openapi.json, https://www.greptile.com/swagger.json, https://www.greptile.c…”
   • [probe] “official MCP server documented at https://www.greptile.com/docs/mcp-v2/overview”
   • [probe] “official CLI documented at https://www.greptile.com/docs/code-review/greptile-cli”
   • [claimed-docs] “Run `greptile review`, get a review of the feature branch diff in 60 seconds.”
   • [claimed-docs] “Connect Cursor, Claude Code, VS Code, or Codex to Greptile MCP with OAuth.”

   proof ↗⚑ flag

   cubic—none0/10

   Cubic documents an MCP server, CLI, and Analytics API, but the evidence pack shows no official client SDKs (e.g., Python/JS libraries) for building against cubic, and explicit probes for an OpenAPI/swagger spec returned 404s, indicating no formal SDK surface is published.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.cubic.dev/openapi.json, https://docs.cubic.dev/swagger.json, https://docs.cubic.dev/api…”
   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”
   • [probe] “official MCP server documented at https://docs.cubic.dev/ide/mcp-server”
   • [probe] “official CLI documented at https://docs.cubic.dev/ide/cli-review”

   proof ↗⚑ flag

9. ai-native userSubscribe to events via webhooks

   weight 2 · round drawn
   Greptile—none0/10

   No evidence of any webhook subscription mechanism for events; Greptile's integrations documented are MCP, CLI, and agent 'Fix with your Agent' flows, but no docs mention webhooks for subscribing to review or event notifications.

   ⚑ flag

   cubic—none0/10

   No evidence of a webhook subscription mechanism; cubic offers an Analytics API, MCP server, and GitHub comment interactions but nothing about outbound event webhooks. OpenAPI/webhook probes returned 404s, and no docs mention webhook subscriptions.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.cubic.dev/openapi.json, https://docs.cubic.dev/swagger.json, https://docs.cubic.dev/api…”
   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”

   proof ↗⚑ flag

Agentic features

1. ai-native userGet AI-generated insights and suggestions from my data inside the product

   weight 2 · round to cubic
   Greptile!disputedcontradicted6/10

   Greptile's core product surfaces AI-generated insights from codebase data — review comments, 0-5 confidence scores, sequence diagrams, auto-generated tests, security findings, and a learning system that improves suggestions from feedback (greptile-docs-16/17/18/25/33, comm-1, comm-8). However, hands-on community reports document concrete failures of this insight-generation capability: repos stuck processing or erroring out, incomplete/incorrect answers on template-heavy code, and search bugs on multi-branch repos (greptile-comm-3/4/5/6/7), directly contradicting the polished vendor narrative for at least some users. Missing for 10: consistent independent verification that insights are reliably accurate across codebases, resolution of the documented processing/error failures, and more recent hands-on corroboration beyond mixed 2024 anecdotes.

   • [claimed-docs] “Greptile is an AI code review agent that automatically reviews every pull request with complete understanding of your codebase.”
   • [claimed-docs] “Know how safe a PR is to merge at a glance with a 0-5 score”
   • [claimed-docs] “Parse complex changes with sequence diagrams and flowcharts generated for every PR”
   • [claimed-docs] “Automatically generate unit tests for new and changed code in every PR”
   • [claimed-docs] “Greptile pairs static scanning with an AI security agent to catch vulnerabilities in every pull request.”
   • [claimed-docs] “Greptile learns from your team's feedback to provide increasingly relevant suggestions. The primary training methods are emoji reactions and…”
   • [community] “Asked Greptile about how BigDecimal is encoded in as_json in Rails; got an accurate answer matching what user learned from PRs/issues/code r…”
   • [community] “My employer uses greptile and I'm pretty happy with it. Sometimes it can be a bit overzealous but more often than not it catches real issues…”
   • [community] “Tried it on my own C++ codebase - impressed it found the C++ standard used, but it's really bad at analyzing templates; answers always incom…”
   • [community] “Not a single repo I've tried works. A lot of them seem not to have finished processing, but even the ones that have finished don't work.”
   • [community] “Ran it on my OSS project - stuck at 99% loading for ~30 minutes, then got 'Error: Internal error while locating sources' on every query. 'Th…”
   • [community] “Loading a repo with many branches (cassandra), searching for 'vsearch' surfaced branch-named results (CNDB-8708-vsearch, DSP-23946-vsearch) …”
   • [community] “Getting 'Error: Internal error while processing request' while trying on my personal public github repo.”

   proof ↗⚑ flag

   cubic✓fullcommunity9/10

   cubic delivers extensive AI-generated insights: automated PR review comments flagging bugs/security issues, AI-generated PR summaries, an AI wiki that indexes the codebase into diagrams/docs, analytics dashboards (AI coding, AI review, delivery) with PR-level insight data, and issue-analysis that checks PRs against linked Jira/Linear tickets. A community reviewer corroborates the quality of these AI-generated technical explanations as more useful than generic LLM output. Missing for 10: independent large-scale benchmarking of insight accuracy beyond one HN anecdote.

   • [claimed-docs] “Spots bugs and improvements: Comments on bugs and improvements in pull requests”
   • [claimed-docs] “PR summaries: Generates PR descriptions based on code changes”
   • [claimed-docs] “cubic's AI wiki automatically indexes your codebase and produces searchable wikis, complete with links to source code, architecture diagrams…”
   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”
   • [claimed-docs] “cubic can automatically analyze your pull requests to see if they meet the requirements from your linked Linear or Jira issues.”
   • [claimed-docs] “The analytics dashboard shows how your team ships code across three lenses: AI coding usage, AI review impact, and delivery speed.”
   • [community] “I really liked it - it hit the mark. The current balance works very well, and it genuinely surprised me. It provides more technical explanat…”

   proof ↗⚑ flag

2. ai-native userSet up automations that run autonomously in the background

   weight 2 · round to cubic
   Greptile✓fullclaimed7/10

   Greptile's core product is background automation: it automatically reviews every PR once triggers are configured (greptile-docs-1, greptile-docs-5), runs an autonomous sandboxed runtime-testing agent (TREX, greptile-docs-19), pairs static+security agents run automatically per PR (greptile-docs-25), and routes reviews to different models autonomously based on PR author (greptile-docs-27) - all without manual per-run invocation. Custom rules, strictness controls, and greptile.json let users configure how these background automations behave (greptile-docs-20, greptile-docs-21, greptile-docs-22). missing for 10: independent/hands-on confirmation that the automation runs reliably unattended over time (community evidence only covers review quality/accuracy, not background reliability), and no evidence of general-purpose scheduled/triggered automations beyond PR-review-centric events

   • [claimed-docs] “Greptile is an AI code review agent that automatically reviews every pull request with complete understanding of your codebase.”
   • [claimed-docs] “Set up Greptile AI code reviews in 5 minutes. Connect GitHub or GitLab, configure review triggers, and get automated feedback on your first …”
   • [claimed-docs] “TREX runs your PR branch in a sandbox to find the bugs that only show up at runtime. It starts services, mocks inputs, clicks through UI flo…”
   • [claimed-docs] “Greptile pairs static scanning with an AI security agent to catch vulnerabilities in every pull request.”
   • [claimed-docs] “v5 is faster, catches more bugs, and flags fewer false positives by running a swarm of narrowly scoped agents in parallel.”
   • [claimed-docs] “Greptile can now detect when a PR was authored by a coding agent and route the review to a different model. If Claude wrote it, GPT reviews …”
   • [claimed-docs] “Control how strict Greptile is about leaving comments with the strictness setting (1–3).”
   • [claimed-docs] “Place `greptile.json` in your repository root. Settings are read from the source branch of the PR and override dashboard settings.”

   proof ↗⚑ flag

   cubic✓fullclaimed8/10

   cubic automatically reviews every new PR without manual trigger, runs codebase-wide scans deploying many agents, can auto-fix issues and auto-approve PRs, and continuously learns from feedback—all happening in the background once installed, per docs-1, docs-4, docs-6, docs-12, docs-29, docs-52. This matches the story of autonomous background automation for code review workflows. Missing for 10: independent/hands-on verification that these automations run reliably unattended over time, and clearer detail on scheduling/triggers beyond PR events.

   • [claimed-docs] “Once installed, cubic automatically reviews new pull requests.”
   • [claimed-docs] “cubic generates the fix and pushes it to your PR branch.”
   • [claimed-docs] “Enable [auto-approval](/ai-review/auto-approval) to let cubic approve PRs that meet your review policy.”
   • [claimed-docs] “Codebase scans deploy thousands of AI agents to find bugs and vulnerabilities across your repository.”
   • [claimed-docs] “Enable auto-approval to let cubic approve PRs that meet your review policy.”
   • [claimed-docs] “cubic can automatically fix issues in the background. When the AI spots a problem, you can click the **Fix with cubic** button”
   • [claimed-docs] “Auto-approval lets you skip human review for pull requests that cubic determines are low risk and issue-free.”
   • [claimed-docs] “cubic automatically learns when you: Reply to its comments... React to suggestions: Thumbs up/down help cubic calibrate what's useful vs noi…”

   proof ↗⚑ flag

3. ai-native userDelegate tasks to a built-in AI assistant inside the product

   weight 3 · round to cubic
   Greptile!disputedcontradicted5/10

   Greptile's docs strongly position it as a built-in AI agent you delegate to — it reviews PRs, generates unit tests, runs sandboxed dynamic tests (TREX), answers codebase questions, and iterates fixes (greptile-docs-1, 18, 19, 33). Community reports include successful task delegation (accurate Q&A, catching real issues in production use) but also concrete hands-on failures — repos stuck processing, 'Internal error' responses, incomplete/incorrect analysis on complex code like C++ templates — directly contradicting reliability of the delegated-task experience. Missing for 10: consistent reliability across codebases/languages, resolution of reported internal errors, and independent benchmarks confirming task success rates.

   • [claimed-docs] “Greptile is an AI code review agent that automatically reviews every pull request with complete understanding of your codebase.”
   • [claimed-docs] “Automatically generate unit tests for new and changed code in every PR”
   • [claimed-docs] “TREX runs your PR branch in a sandbox to find the bugs that only show up at runtime. It starts services, mocks inputs, clicks through UI flo…”
   • [claimed-docs] “Greptile learns from your team's feedback to provide increasingly relevant suggestions. The primary training methods are emoji reactions and…”
   • [community] “Asked Greptile about how BigDecimal is encoded in as_json in Rails; got an accurate answer matching what user learned from PRs/issues/code r…”
   • [community] “Tried it on my own C++ codebase - impressed it found the C++ standard used, but it's really bad at analyzing templates; answers always incom…”
   • [community] “Not a single repo I've tried works. A lot of them seem not to have finished processing, but even the ones that have finished don't work.”
   • [community] “Ran it on my OSS project - stuck at 99% loading for ~30 minutes, then got 'Error: Internal error while locating sources' on every query. 'Th…”
   • [community] “Getting 'Error: Internal error while processing request' while trying on my personal public github repo.”
   • [community] “My employer uses greptile and I'm pretty happy with it. Sometimes it can be a bit overzealous but more often than not it catches real issues…”

   proof ↗⚑ flag

   cubic✓fullcommunity7/10

   cubic acts as a built-in AI teammate that users can direct with explicit commands (e.g. '@cubic-dev-ai review this PR', triggering ultrareview, asking it to generate and push a fix, replying for clarification, enabling auto-approval, or running codebase scans that deploy AI agents), and a community reviewer corroborates that it does more than basic checks. Missing for 10: independent/hands-on verification of these delegated actions actually completing end-to-end, and evidence of broader (non-review) task delegation beyond the code-review domain.

   • [claimed-docs] “To review a PR that was opened _before_ you installed the app, comment: `@cubic-dev-ai review this PR`.”
   • [claimed-docs] “Ultrareview runs a longer review using cubic's most capable review models, which is useful for risky migrations, security-sensitive changes,…”
   • [claimed-docs] “cubic generates the fix and pushes it to your PR branch.”
   • [claimed-docs] “Reply to a review comment to ask for clarification:”
   • [claimed-docs] “Enable [auto-approval](/ai-review/auto-approval) to let cubic approve PRs that meet your review policy.”
   • [claimed-docs] “Codebase scans deploy thousands of AI agents to find bugs and vulnerabilities across your repository.”
   • [community] “I really liked it - it hit the mark. The current balance works very well, and it genuinely surprised me. It provides more technical explanat…”

   proof ↗⚑ flag

4. ai-native userOperate the product with natural-language commands

   weight 2 · round to cubic
   Greptile~partialcommunity5/10

   Greptile supports natural-language interaction for codebase Q&A (comm-1 shows a user asking an English question and getting an accurate answer) and lets teams define custom review rules in plain language (greptile-docs-21), but its primary interfaces (CLI commands like `greptile review`, MCP tool calls, PR-trigger automation) are structured, not conversational commands. missing for 10: a documented chat/command interface for issuing arbitrary natural-language instructions to control reviews or agent actions, and independent verification that NL querying works reliably at scale.

   • [community] “Asked Greptile about how BigDecimal is encoded in as_json in Rails; got an accurate answer matching what user learned from PRs/issues/code r…”
   • [claimed-docs] “Create custom rules that catch team-specific issues”
   • [claimed-docs] “Your 👍/👎 reactions and replies teach Greptile what matters. After 2-3 weeks, it stops commenting on things you don't care about.”
   • [claimed-docs] “Run `greptile review`, get a review of the feature branch diff in 60 seconds.”

   proof ↗⚑ flag

   cubic✓fullclaimed7/10

   cubic supports natural-language interaction: commenting `@cubic-dev-ai review this PR`, replying to review comments to ask clarification or trigger fixes, and requesting an Ultrareview, plus MCP server integration letting a coding agent issue natural-language requests (review, triage, manage subscription) on cubic's behalf. This covers command-like NL operation across PR comments and agent tooling. missing for 10: independent/hands-on verification of NL command robustness beyond documented examples, and no evidence of a broader conversational/chat interface beyond structured comment triggers.

   • [claimed-docs] “To review a PR that was opened _before_ you installed the app, comment: `@cubic-dev-ai review this PR`.”
   • [claimed-docs] “Reply to a review comment to ask for clarification:”
   • [claimed-docs] “@cubic-dev-ai review this PR”
   • [claimed-docs] “Interact with cubic in PR comments to ask questions, trigger reviews, and fix issues. You can tag `@cubic-dev-ai` directly, or reply to cubi…”
   • [claimed-docs] “Connect cubic's MCP server to your coding agent to read review findings and codebase context, request PR reviews, and triage PR or codebase …”
   • [claimed-docs] “You can now ask your coding agent to check your cubic subscription, manage team seats and roles, and purchase more seats without leaving you…”
   • [claimed-docs] “When a pull request needs extra scrutiny, you can ask cubic for an Ultrareview.”

   proof ↗⚑ flag

Api quality

1. ai-native userExplore an interactive API reference with runnable examples

   weight 2 · round drawn
   Greptile—none0/10

   No evidence of an interactive API reference or runnable examples; the openapi.json/swagger.json probe explicitly returned 404s across all candidate paths, and docs only describe CLI/MCP/dashboard workflows, not an API explorer.

   • [probe] “PROBE openapi: all candidate paths 404 (https://www.greptile.com/openapi.json, https://www.greptile.com/swagger.json, https://www.greptile.c…”

   proof ↗⚑ flag

   cubic—none0/10

   There is an Analytics API mentioned but no evidence of an interactive API reference or runnable examples; probes explicitly show no OpenAPI/swagger spec found (all candidate paths 404), indicating no interactive API documentation exists.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.cubic.dev/openapi.json, https://docs.cubic.dev/swagger.json, https://docs.cubic.dev/api…”
   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”

   proof ↗⚑ flag

2. ai-native userDownload a machine-readable API spec (OpenAPI or equivalent)

   weight 2 · round drawn
   Greptile—none0/10

   A direct probe found no OpenAPI/Swagger spec at any standard location (all 404s), and no other evidence mentions a machine-readable API spec being available for download.

   • [probe] “PROBE openapi: all candidate paths 404 (https://www.greptile.com/openapi.json, https://www.greptile.com/swagger.json, https://www.greptile.c…”

   proof ↗⚑ flag

   cubic—none0/10

   cubic has an Analytics API but there is no documented machine-readable OpenAPI spec or equivalent; a direct probe for openapi.json/swagger.json at standard paths returned 404 across all candidates.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.cubic.dev/openapi.json, https://docs.cubic.dev/swagger.json, https://docs.cubic.dev/api…”
   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”

   proof ↗⚑ flag

3. ai-native userRely on versioned APIs with a documented deprecation policy

   weight 2 · round drawn
   Greptile—none0/10

   No evidence of API versioning scheme or a documented deprecation policy; the openapi probe returned 404s and no changelog entries mention API versioning/deprecation. Missing for 10: versioned API docs, explicit deprecation policy, changelog entries about API version sunsetting.

   • [probe] “PROBE openapi: all candidate paths 404 (https://www.greptile.com/openapi.json, https://www.greptile.com/swagger.json, https://www.greptile.c…”

   proof ↗⚑ flag

   cubic—none0/10

   There is an Analytics API mentioned, but no evidence of versioning scheme or a documented deprecation policy; the openapi probe returned 404s across all candidate paths, indicating no discoverable API spec/versioning docs.

   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.cubic.dev/openapi.json, https://docs.cubic.dev/swagger.json, https://docs.cubic.dev/api…”

   proof ↗⚑ flag

Ai authored

1. ai-native userThe reviewer holds the line on AI-generated PRs — it verifies agent-authored code at a volume no human team could review

   weight 2 · round drawn
   Greptile✓fullcommunity8/10

   Greptile explicitly documents detecting when a PR was AI-authored and routing it to a different model for review (docs-27), reviews every PR automatically at scale via parallel agent swarms (docs-26), runs runtime sandbox testing (TREX, docs-19), and offers an iterative 'Greplooping' fix cycle until a PR hits 5/5 confidence with zero unresolved comments (docs-13/28) — directly matching the 'holds the line on agent PRs at volume' story. Community evidence corroborates real-world catch-rate value (comm-8) though some older reports cite reliability/error issues unrelated specifically to this agent-detection feature. Missing for 10: independent hands-on verification of the specific agent-authorship detection/routing feature and concrete volume/throughput metrics.

   • [claimed-docs] “Greptile can now detect when a PR was authored by a coding agent and route the review to a different model. If Claude wrote it, GPT reviews …”
   • [claimed-docs] “v5 is faster, catches more bugs, and flags fewer false positives by running a swarm of narrowly scoped agents in parallel.”
   • [claimed-docs] “TREX runs your PR branch in a sandbox to find the bugs that only show up at runtime. It starts services, mocks inputs, clicks through UI flo…”
   • [claimed-docs] “Iteratively fix a PR until Greptile gives a perfect review: 5/5 confidence, zero unresolved comments.”
   • [claimed-docs] “Greplooping Iteratively fix a PR until Greptile gives a perfect review: 5/5 confidence, zero unresolved comments.”
   • [claimed-docs] “Greptile pairs static scanning with an AI security agent to catch vulnerabilities in every pull request.”
   • [community] “My employer uses greptile and I'm pretty happy with it. Sometimes it can be a bit overzealous but more often than not it catches real issues…”

   proof ↗⚑ flag

   cubic✓fullcommunity8/10

   cubic is explicitly built to auto-review every PR, run deep Ultrareview/codebase scans deploying 'thousands of AI agents,' auto-fix and auto-approve, and surface analytics specifically tracking AI-authored code volume — directly matching the story of scaling review beyond human capacity. Community feedback (cubic-comm-1) corroborates real-world value, though cubic-comm-2 raises a substantive critique that YAML-based rules are less robust than code-based checks, a minor caveat rather than a contradiction of the core capability. Missing for 10: independent large-scale benchmark/study proving review quality holds at massive AI-PR volume, and more than two community data points.

   • [claimed-docs] “Once installed, cubic automatically reviews new pull requests.”
   • [claimed-docs] “Codebase scans deploy thousands of AI agents to find bugs and vulnerabilities across your repository.”
   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”
   • [claimed-docs] “Flex capacity keeps GitHub PR AI reviews running after your workspace uses its included reviewed-line capacity.”
   • [claimed-docs] “The AI coding tab shows how AI assistants contribute to merged pull requests in your repositories.”
   • [claimed-docs] “The AI automatically scans every pull request for a wide range of issues, including logic errors, style inconsistencies, and potential secur…”
   • [claimed-docs] “Auto-approval lets you skip human review for pull requests that cubic determines are low risk and issue-free.”
   • [community] “I really liked it - it hit the mark. The current balance works very well, and it genuinely surprised me. It provides more technical explanat…”
   • [community] “The 'in-codebase' approach is the right one, but a YAML file with plain text is a half-measure. The most reliable rule that 'lives next to t…”

   proof ↗⚑ flag

Checks

1. ai-native userI define custom agentic pre-merge checks in plain language — 'docs updated', 'tests cover new paths' — that run on every PR

   weight 2 · round to cubic
   Greptile~partialclaimed6/10

   Greptile supports custom, team-defined rules in plain language ('Create custom rules that catch team-specific issues') configured via greptile.json, and these run automatically on every PR alongside a confidence score/unresolved-comment gate that can act as a de facto merge check (docs-21, docs-22, docs-13). However, there's no explicit evidence of a dedicated 'pre-merge check' construct with pass/fail CI status gating, nor examples matching the story's specific phrasing (e.g., 'docs updated', 'tests cover new paths') as discrete named checks. Missing for 10: explicit merge-blocking/CI-status-check integration tied to named custom checks, and concrete examples/documentation of plain-language check definitions beyond generic 'custom rules'.

   • [claimed-docs] “Create custom rules that catch team-specific issues”
   • [claimed-docs] “Place `greptile.json` in your repository root. Settings are read from the source branch of the PR and override dashboard settings.”
   • [claimed-docs] “Iteratively fix a PR until Greptile gives a perfect review: 5/5 confidence, zero unresolved comments.”
   • [claimed-docs] “Control how strict Greptile is about leaving comments with the strictness setting (1–3).”
   • [claimed-docs] “With noise control, Greptile limits reviews to high-signal insights, skipping low impact or repetitive feedback.”

   proof ↗⚑ flag

   cubic✓fullcommunity7/10

   cubic's 'Custom agents' feature lets teams define plain-language review rules enforcing org-specific best practices, stored in cubic.yaml, applied automatically across every PR, with a community library for adoption — directly matching the story of custom agentic pre-merge checks defined in plain language. Missing for 10: no first-party example matching the exact 'docs updated' / 'tests cover new paths' phrasing, and no independent hands-on validation of how reliably these plain-text rules enforce (one community comment raises skepticism about plain-text rule reliability vs. hard-coded checks, though this is not a concrete failure report).

   • [claimed-docs] “Custom agents are review rules that enforce your organization's specific best practices across pull requests.”
   • [claimed-docs] “Custom agents: Enforce your team's coding standards”
   • [claimed-docs] “You can also adopt existing agents from the community library with a single click.”
   • [claimed-docs] “cubic.yaml lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, an…”
   • [claimed-docs] “`cubic.yaml` lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, …”
   • [community] “The 'in-codebase' approach is the right one, but a YAML file with plain text is a half-measure. The most reliable rule that 'lives next to t…”

   proof ↗⚑ flag

Fixes

1. developerI turn a review finding into an applied fix — a committed patch or an agent-generated follow-up — without leaving the PR

   weight 3 · round drawn
   Greptile✓fullclaimed8/10

   Docs describe a complete in-PR fix loop: each review comment has a 'Fix with your Agent' button that sends file paths/lines/suggested code to Claude Code, Codex, Cursor, or Devin, a 'Fix All' button for batch fixes, MCP integration to fetch/resolve comments from the IDE, and 'Greplooping' to iteratively fix until a perfect review score—all without leaving the PR context. Missing for 10: independent/hands-on evidence confirming the fix-apply workflow works reliably in practice (community evidence in the pack addresses general review quality/accuracy, not this specific fix-application flow).

   • [claimed-docs] “Every review comment includes a Fix with your Agent button that sends the issue — with file paths, line numbers, and suggested code — straig…”
   • [claimed-docs] “A Fix All button in the review summary sends every issue at once.”
   • [claimed-docs] “Launch any agent in one click with a Greptile comment, full context, and suggested fix.”
   • [claimed-docs] “Access Greptile review comments, apply fixes, and manage coding patterns directly from your agent of choice.”
   • [claimed-docs] “Iteratively fix a PR until Greptile gives a perfect review: 5/5 confidence, zero unresolved comments.”
   • [claimed-docs] “Fetch PR comments - Get unaddressed Greptile feedback for any PR”
   • [claimed-docs] “Greplooping Iteratively fix a PR until Greptile gives a perfect review: 5/5 confidence, zero unresolved comments.”
   • [claimed-docs] “Org admins can go to **Settings → Code Review → Default Coding Agents** to turn on Fix with your Agent for the org.”
   • [claimed-docs] “Every Greptile review comment includes a **Fix with your Agent** button. Click it, and the issue gets sent straight to your coding agent wit…”

   proof ↗⚑ flag

   cubic✓fullclaimed8/10

   cubic lets developers turn review findings into applied fixes directly in the PR: it can generate a fix and push it to the PR branch, offers a one-click 'Fix with cubic' button, and its coding-agent/MCP integration lets an agent read findings and generate follow-up fixes without leaving the PR workflow. Auto-resolve threads and interactive PR comments reinforce that the fix loop stays within the PR. Missing for 10: independent/hands-on verification of the auto-fix push actually working in practice beyond vendor docs, and more detail on failure/edge cases of the auto-fix flow.

   • [claimed-docs] “cubic generates the fix and pushes it to your PR branch.”
   • [claimed-docs] “cubic can automatically fix issues in the background. When the AI spots a problem, you can click the **Fix with cubic** button”
   • [claimed-docs] “cubic can automatically fix issues identified during code review. Request a targeted fix with one click.”
   • [claimed-docs] “Auto-resolve threads: Enable automatic thread resolution to close findings when the issue is fixed”
   • [claimed-docs] “Interact with cubic in PR comments to ask questions, trigger reviews, and fix issues.”
   • [claimed-docs] “Connect cubic's MCP server to your coding agent to read review findings and codebase context, request PR reviews, and triage PR or codebase …”
   • [claimed-docs] “Coding agents: Generate fixes when requested, using your configured provider”

   proof ↗⚑ flag

Handoff

1. ai-native userReview findings hand off cleanly to my coding agent — copyable fix prompts or direct integration with Claude Code, Cursor, or Codex

   weight 2 · round to Greptile
   Greptile✓fullclaimed9/10

   Greptile documents a 'Fix with your Agent' button on every review comment that sends file paths, line numbers, and suggested code directly to Claude Code, Cursor, Codex, Devin, or Conductor, plus a 'Fix All' button for batch handoff, and an MCP v2 server letting agents fetch/resolve unaddressed comments with OAuth support for Cursor/Claude Code/VS Code/Codex. This directly matches the story's clean hand-off requirement with both one-click integration and MCP-based programmatic access. Missing for 10: independent/hands-on community verification that the 'Fix with your Agent' handoff works reliably in practice (only vendor docs cited).

   • [claimed-docs] “Every review comment includes a Fix with your Agent button that sends the issue — with file paths, line numbers, and suggested code — straig…”
   • [claimed-docs] “A Fix All button in the review summary sends every issue at once.”
   • [claimed-docs] “Launch any agent in one click with a Greptile comment, full context, and suggested fix.”
   • [claimed-docs] “Access Greptile review comments, apply fixes, and manage coding patterns directly from your agent of choice.”
   • [claimed-docs] “View and resolve review comments directly from Claude Code.”
   • [claimed-docs] “Fetch PR comments - Get unaddressed Greptile feedback for any PR”
   • [claimed-docs] “Connect Cursor, Claude Code, VS Code, or Codex to Greptile MCP with OAuth.”
   • [claimed-docs] “Org admins can go to **Settings → Code Review → Default Coding Agents** to turn on Fix with your Agent for the org.”
   • [claimed-docs] “Every Greptile review comment includes a **Fix with your Agent** button. Click it, and the issue gets sent straight to your coding agent wit…”

   proof ↗⚑ flag

   cubic✓fullprobed8/10

   cubic explicitly supports both hand-off paths: the CLI reviews local changes and 'generates a prompt that your coding agent can use to fix them' (cubic-docs-7), and cubic's MCP server lets you 'connect cubic's MCP server to your coding agent' with named support for Cursor, Claude Code, VS Code, Codex, and Gemini CLI (cubic-docs-8, cubic-docs-35). PR-level auto-fix ('Fix with cubic') that pushes fixes directly to the branch (cubic-docs-4, cubic-docs-29, cubic-docs-53) and coding-agent skills (cubic-docs-17) further reinforce clean hand-off to agents. Missing for 10: independent/hands-on confirmation that the generated prompts or MCP flow work smoothly in practice, and no community evidence specifically validating the agent hand-off UX.

   • [claimed-docs] “The **cubic CLI** reviews local changes before you push. It finds bugs and generates a prompt that your coding agent can use to fix them.”
   • [claimed-docs] “Connect cubic's MCP server to your coding agent to read review findings and codebase context, request PR reviews, and triage PR or codebase …”
   • [claimed-docs] “Install and connect cubic in Cursor, Claude Code, VS Code, Codex, Gemini CLI, or another coding tool.”
   • [claimed-docs] “cubic generates the fix and pushes it to your PR branch.”
   • [claimed-docs] “cubic can automatically fix issues in the background. When the AI spots a problem, you can click the **Fix with cubic** button”
   • [claimed-docs] “cubic can automatically fix issues identified during code review. Request a targeted fix with one click.”
   • [claimed-docs] “cubic skills give your coding agent instructions for reviewing code, handling PR comments, and using your team's codebase context.”
   • [probe] “official MCP server documented at https://docs.cubic.dev/ide/mcp-server”
   • [probe] “official CLI documented at https://docs.cubic.dev/ide/cli-review”

   proof ↗⚑ flag

1. ai-native userPerform bulk operations across many items at once

   weight 2 · round to Greptile
   Greptile~partialclaimed6/10

   Greptile documents a 'Fix All' button that sends every issue in a review at once to a coding agent, and `greptile onboard` can enable multiple repositories and import existing AI rules files in one command — both genuine bulk operations. However, evidence doesn't show bulk actions across many PRs, repos, or reviews simultaneously (e.g., batch-reviewing multiple PRs, mass-applying custom rules across repos), and there's no independent/hands-on confirmation of these bulk features working at scale. missing for 10: cross-PR/cross-repo bulk review or fix operations, evidence of bulk custom-rule application across an org, independent verification of Fix All/onboard at scale.

   • [claimed-docs] “A Fix All button in the review summary sends every issue at once.”
   • [claimed-docs] “Iteratively fix a PR until Greptile gives a perfect review: 5/5 confidence, zero unresolved comments.”
   • [claimed-docs] “Greplooping Iteratively fix a PR until Greptile gives a perfect review: 5/5 confidence, zero unresolved comments.”
   • [claimed-docs] “`greptile onboard` creates your organization, connects GitHub or GitLab, enables repositories, and imports your existing AI rules files — th…”
   • [claimed-docs] “Org admins can go to **Settings → Code Review → Default Coding Agents** to turn on Fix with your Agent for the org.”

   proof ↗⚑ flag

   cubic~partialclaimed5/10

   cubic supports some bulk-scale automation — codebase scans that 'deploy thousands of AI agents to find bugs and vulnerabilities across your repository' (cubic-docs-12) and an Analytics API that aggregates PR-level data across many PRs (cubic-docs-16, cubic-docs-59) — but there is no evidence of user-triggered bulk operations like batch-fixing, batch-approving, or batch-triaging many PRs/items at once beyond per-PR auto-approval and auto-fix. Missing for 10: explicit bulk/batch action UI or API (e.g., select-many-and-fix, bulk approve across PRs), evidence of scale/performance at bulk volume, and independent confirmation of the codebase-scan bulk claim.

   • [claimed-docs] “Codebase scans deploy thousands of AI agents to find bugs and vulnerabilities across your repository.”
   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”
   • [claimed-docs] “The analytics dashboard shows how your team ships code across three lenses: AI coding usage, AI review impact, and delivery speed.”
   • [claimed-docs] “Cross-repo reviews help cubic catch changes that need a matching update in another repository. Link related repositories so reviews can chec…”
   • [claimed-docs] “Enable [auto-approval](/ai-review/auto-approval) to let cubic approve PRs that meet your review policy.”

   proof ↗⚑ flag

2. ai-native userDefine rules that trigger actions automatically on events

   weight 3 · round drawn
   Greptile~partialclaimed5/10

   Greptile lets users configure review triggers, custom standards/rules (greptile.json, custom-standards docs), strictness/noise control, and automated actions like Fix with your Agent triggered on review events, which is a form of rule-based automation tied to PR events. However, this is scoped to code review workflows only (triggering on PR open/comment events) rather than a general-purpose event/rule engine for arbitrary automation actions. missing for 10: evidence of a broader/generic rules engine spanning multiple event types beyond PR review, user-defined conditional logic beyond review strictness/custom standards, and independent confirmation these triggers work reliably as described.

   • [claimed-docs] “Set up Greptile AI code reviews in 5 minutes. Connect GitHub or GitLab, configure review triggers, and get automated feedback on your first …”
   • [claimed-docs] “Create custom rules that catch team-specific issues”
   • [claimed-docs] “Place `greptile.json` in your repository root. Settings are read from the source branch of the PR and override dashboard settings.”
   • [claimed-docs] “Control how strict Greptile is about leaving comments with the strictness setting (1–3).”
   • [claimed-docs] “With noise control, Greptile limits reviews to high-signal insights, skipping low impact or repetitive feedback.”
   • [claimed-docs] “Every Greptile review comment includes a **Fix with your Agent** button. Click it, and the issue gets sent straight to your coding agent wit…”
   • [claimed-docs] “Org admins can go to **Settings → Code Review → Default Coding Agents** to turn on Fix with your Agent for the org.”

   proof ↗⚑ flag

   cubic~partialcommunity5/10

   cubic supports event-triggered automation within its code-review domain: PRs auto-trigger reviews, custom agents enforce rules on PRs, auto-approval and auto-fix run on policy conditions, and cubic.yaml configures behavior — but this is domain-specific (PR/code-review events) rather than a general-purpose rules engine for arbitrary events/actions that an AI-native user could define. Missing for 10: evidence of a generalized rule/trigger builder spanning arbitrary event sources and actions beyond PR review workflows, and independent hands-on validation of custom-agent rule reliability (community feedback questions robustness of YAML-based rules).

   • [claimed-docs] “Once installed, cubic automatically reviews new pull requests.”
   • [claimed-docs] “Custom agents are review rules that enforce your organization's specific best practices across pull requests.”
   • [claimed-docs] “Enable [auto-approval](/ai-review/auto-approval) to let cubic approve PRs that meet your review policy.”
   • [claimed-docs] “Enable auto-approval to let cubic approve PRs that meet your review policy.”
   • [claimed-docs] “cubic can automatically fix issues in the background. When the AI spots a problem, you can click the **Fix with cubic** button”
   • [claimed-docs] “cubic.yaml lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, an…”
   • [claimed-docs] “Auto-approval lets you skip human review for pull requests that cubic determines are low risk and issue-free.”
   • [community] “The 'in-codebase' approach is the right one, but a YAML file with plain text is a half-measure. The most reliable rule that 'lives next to t…”

   proof ↗⚑ flag

3. ai-native userVersion, review, and roll back my automations

   weight 1 · round to Greptile
   Greptile~partialclaimed4/10

   Greptile's automation config (`greptile.json`) lives in the repo root and is read per-branch, which implicitly gives git-based versioning and revertibility, and dashboard settings like strictness/custom rules can presumably be changed — but there is no explicit dashboard version history, audit log, or documented rollback feature for automation settings themselves. Missing for 10: explicit versioning/audit-trail UI for automation configs, a documented rollback/undo mechanism, and any review workflow for changes to the automation rules beyond relying on git for the JSON file.

   • [claimed-docs] “Place `greptile.json` in your repository root. Settings are read from the source branch of the PR and override dashboard settings.”
   • [claimed-docs] “Control how strict Greptile is about leaving comments with the strictness setting (1–3).”
   • [claimed-docs] “Create custom rules that catch team-specific issues”
   • [claimed-docs] “Greptile learns from your team's feedback to provide increasingly relevant suggestions. The primary training methods are emoji reactions and…”

   proof ↗⚑ flag

   cubic~partialclaimed3/10

   cubic's configuration (custom agents, review policies) lives in a `cubic.yaml` file at the repo root, which implies it inherits standard git version control (commit history, diff, revert) for review and rollback, but there is no dedicated in-product versioning, audit trail, or rollback UI/feature described for cubic's automations (custom agents, auto-approval rules, etc.). Missing for 10: explicit in-product version history/changelog for custom agents or review policies, a documented rollback mechanism beyond generic git, and any review workflow specifically for automation configuration changes.

   • [claimed-docs] “`cubic.yaml` lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, …”
   • [claimed-docs] “cubic.yaml lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, an…”
   • [claimed-docs] “Custom agents are review rules that enforce your organization's specific best practices across pull requests.”
   • [claimed-docs] “Custom agents: Enforce your team's coding standards”

   proof ↗⚑ flag

Context

1. engineering leadThe reviewer understands changes that span multiple repositories or a large monorepo and reviews them coherently

   weight 2 · round to cubic
   Greptile—none0/10

   Greptile's evidence covers per-PR review, codebase-wide Q&A, and self-hosted deployment for large dev teams, but there is no documentation or feature description addressing coherent review of changes spanning multiple repositories or explicit monorepo-scale review orchestration. The only tangentially relevant community evidence (greptile-comm-6) reports a bug where searching a large, many-branch repo (Cassandra) failed to surface correct results, which does not support (and mildly undercuts) coherent large-scale/monorepo understanding.

   • [claimed-docs] “Greptile is an AI code review agent that automatically reviews every pull request with complete understanding of your codebase.”
   • [community] “Loading a repo with many branches (cassandra), searching for 'vsearch' surfaced branch-named results (CNDB-8708-vsearch, DSP-23946-vsearch) …”

   proof ↗⚑ flag

   cubic~partialclaimed7/10

   cubic explicitly ships cross-repo reviews (link related repositories so reviews can check shared APIs/schemas/docs) and codebase-wide indexing via AI wiki and codebase scans, which support coherent understanding of changes spanning repos or large codebases. However, evidence is entirely vendor docs with no independent/hands-on validation of monorepo-scale coherence, and the cross-repo feature description is thin (only checking shared APIs/schemas, not full monorepo-scale reasoning). Missing for 10: independent/hands-on evidence of multi-repo review coherence at scale, explicit monorepo support details, and any case study of large/complex spanning changes reviewed successfully.

   • [claimed-docs] “Link related repositories so reviews can check shared APIs, schemas, docs, or release work.”
   • [claimed-docs] “Cross-repo reviews help cubic catch changes that need a matching update in another repository. Link related repositories so reviews can chec…”
   • [claimed-docs] “Codebase scans deploy thousands of AI agents to find bugs and vulnerabilities across your repository.”
   • [claimed-docs] “cubic's AI wiki automatically indexes your codebase and produces searchable wikis, complete with links to source code, architecture diagrams…”

   proof ↗⚑ flag

2. developerReview comments reflect the whole repository — call sites, related modules, existing conventions — not just the changed hunks

   weight 3 · round to cubic
   Greptile~partialcommunity6/10

   Greptile's docs explicitly claim reviews use 'complete understanding of your codebase' (greptile-docs-1), index existing rule/convention files like AGENTS.md/cursor.rules (greptile-docs-35), and support custom team-specific rules (greptile-docs-21) that presumably span beyond the diff. A community hands-on report corroborates deep repo-level Q&A accuracy (greptile-comm-1), but other hands-on reports describe incomplete/incorrect answers on complex code (templates) and reliability failures on some repos (greptile-comm-3, greptile-comm-4, greptile-comm-5), showing the whole-repo understanding is inconsistent in practice. Missing for 10: independent verification specifically that PR review COMMENTS (not just Q&A) cite call sites/related modules beyond the diff, and resolution of the negative reliability reports.

   • [claimed-docs] “Greptile is an AI code review agent that automatically reviews every pull request with complete understanding of your codebase.”
   • [claimed-docs] “Automatically index existing rule files like Claude.md, AGENTS.md, and cursor.rules for richer review context.”
   • [claimed-docs] “Create custom rules that catch team-specific issues”
   • [community] “Asked Greptile about how BigDecimal is encoded in as_json in Rails; got an accurate answer matching what user learned from PRs/issues/code r…”
   • [community] “Tried it on my own C++ codebase - impressed it found the C++ standard used, but it's really bad at analyzing templates; answers always incom…”
   • [community] “Not a single repo I've tried works. A lot of them seem not to have finished processing, but even the ones that have finished don't work.”

   proof ↗⚑ flag

   cubic~partialclaimed7/10

   cubic's docs describe repo-wide context mechanisms beyond the diff: it auto-detects context files (README, AGENTS.md, .cursorrules), links related repositories for cross-repo reviews of shared APIs/schemas, runs codebase scans across the whole repo, builds an AI wiki index of the codebase, and learns 'unwritten rules' from senior engineers' historical reviews and team feedback — all suggesting reviews draw on broader repository knowledge, not just changed hunks. However, no evidence explicitly confirms it traces call sites or cross-references unrelated modules within a single-repo review, and there's no independent/hands-on corroboration of this specific behavior. Missing for 10: explicit call-site/dependency-tracing evidence, independent verification that review comments actually cite non-hunk code.

   • [claimed-docs] “cubic automatically detects and uses context files in your repository, including `README.md`, `context.md`, `AGENTS.md`, `.cursorrules`”
   • [claimed-docs] “cubic automatically detects and uses context files in your repository, including `README.md`, `context.md`, `AGENTS.md`, `.cursorrules`, age…”
   • [claimed-docs] “Link related repositories so reviews can check shared APIs, schemas, docs, or release work.”
   • [claimed-docs] “Cross-repo reviews help cubic catch changes that need a matching update in another repository. Link related repositories so reviews can chec…”
   • [claimed-docs] “Codebase scans deploy thousands of AI agents to find bugs and vulnerabilities across your repository.”
   • [claimed-docs] “cubic's AI wiki automatically indexes your codebase and produces searchable wikis, complete with links to source code, architecture diagrams…”
   • [claimed-docs] “cubic learns your team's unwritten rules from two sources: direct feedback on its comments, and patterns extracted from your senior engineer…”

   proof ↗⚑ flag

Memory

1. ai-native userThe reviewer builds a persistent memory of my team's conventions and past review decisions and applies it to future PRs

   weight 2 · round to cubic
   Greptile✓fullclaimed7/10

   Greptile documents a persistent-learning system: thumbs up/down reactions and comments train the model over weeks (docs-4, docs-33), custom rules/patterns can be created and searched via MCP (docs-21, docs-30), team conventions are codified in a `greptile.json` config that overrides defaults per-repo (docs-22), and existing rule files like CLAUDE.md/AGENTS.md/cursor.rules are auto-indexed for context (docs-35, docs-31). This directly matches the story of building and applying a persistent memory of conventions/past decisions to future PRs. Missing for 10: independent/hands-on confirmation that the learning system measurably reduces noise over time as claimed, and community evidence is mixed/dated rather than specifically validating the memory feature.

   • [claimed-docs] “Your 👍/👎 reactions and replies teach Greptile what matters. After 2-3 weeks, it stops commenting on things you don't care about.”
   • [claimed-docs] “Greptile learns from your team's feedback to provide increasingly relevant suggestions. The primary training methods are emoji reactions and…”
   • [claimed-docs] “Create custom rules that catch team-specific issues”
   • [claimed-docs] “Place `greptile.json` in your repository root. Settings are read from the source branch of the PR and override dashboard settings.”
   • [claimed-docs] “With MCP, you can view, search, and create patterns from your IDE.”
   • [claimed-docs] “Automatically index existing rule files like Claude.md, AGENTS.md, and cursor.rules for richer review context.”
   • [claimed-docs] “`greptile onboard` creates your organization, connects GitHub or GitLab, enables repositories, and imports your existing AI rules files — th…”
   • [claimed-docs] “With noise control, Greptile limits reviews to high-signal insights, skipping low impact or repetitive feedback.”

   proof ↗⚑ flag

   cubic✓fullcommunity8/10

   cubic explicitly documents persistent memory: it learns from replies/reactions to its comments and from patterns in senior engineers' historical reviews, storing conventions in cubic.yaml and custom agents that persist across PRs and repos (cubic-docs-10,32,33,54,9,11,34). This directly matches the ai-native persona's desire for a reviewer that builds team-convention memory over time. Missing for 10: independent/hands-on verification that learned rules actually persist and improve accuracy over many PRs (one community comment questions whether YAML-based rules are robust vs. hard failures, cubic-comm-2), and no quantitative data on learning effectiveness.

   • [claimed-docs] “cubic automatically learns when you: Reply to its comments... React to suggestions: Thumbs up/down help cubic calibrate what's useful vs noi…”
   • [claimed-docs] “Reply to its comments: Type a response explaining what cubic should do differently, and it remembers for next time”
   • [claimed-docs] “React to suggestions: Thumbs up/down help cubic calibrate what's useful vs noise for your specific codebase”
   • [claimed-docs] “cubic learns your team's unwritten rules from two sources: direct feedback on its comments, and patterns extracted from your senior engineer…”
   • [claimed-docs] “Custom agents are review rules that enforce your organization's specific best practices across pull requests.”
   • [claimed-docs] “`cubic.yaml` lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, …”
   • [claimed-docs] “cubic.yaml lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, an…”
   • [community] “The 'in-codebase' approach is the right one, but a YAML file with plain text is a half-measure. The most reliable rule that 'lives next to t…”

   proof ↗⚑ flag

Chat

1. developerI reply to the reviewer in the PR thread to ask questions, get explanations, or issue commands — and it answers in context

   weight 2 · round to cubic
   Greptile~partialcommunity5/10

   Docs confirm the PR-thread reply channel exists and is used to tune reviews (👍/👎 and replies 'teach' Greptile, greptile-docs-4/33), and one community report shows Greptile answering a direct codebase question accurately (greptile-comm-1), while another notes it gives authors 'a chance to reply' during review (greptile-comm-8). However, there's no explicit documentation of replying with a question/command in-thread and receiving a contextual answer (as opposed to reaction-based learning), and other community reports show inconsistent/failed query handling (greptile-comm-3,4,5,7). Missing for 10: first-party docs explicitly describing PR-thread Q&A/command handling and consistent independent confirmation of in-context answers.

   • [claimed-docs] “Your 👍/👎 reactions and replies teach Greptile what matters. After 2-3 weeks, it stops commenting on things you don't care about.”
   • [claimed-docs] “Greptile learns from your team's feedback to provide increasingly relevant suggestions. The primary training methods are emoji reactions and…”
   • [community] “Asked Greptile about how BigDecimal is encoded in as_json in Rails; got an accurate answer matching what user learned from PRs/issues/code r…”
   • [community] “My employer uses greptile and I'm pretty happy with it. Sometimes it can be a bit overzealous but more often than not it catches real issues…”
   • [community] “Tried it on my own C++ codebase - impressed it found the C++ standard used, but it's really bad at analyzing templates; answers always incom…”
   • [community] “Not a single repo I've tried works. A lot of them seem not to have finished processing, but even the ones that have finished don't work.”

   proof ↗⚑ flag

   cubic✓fullcommunity8/10

   Docs explicitly describe replying to cubic's PR comments to ask questions, get clarifications, issue fix commands, or trigger reviews/Ultrareview, with cubic responding in-thread and even learning from replies. This is well documented across multiple pages (interactive-comments, memory-and-learning, quickstart) and corroborated by a community user noting cubic gives deeper technical explanations than plain LLMs. missing for 10: independent hands-on verification of multi-turn contextual follow-up quality/accuracy in real threads.

   • [claimed-docs] “Reply to a review comment to ask for clarification:”
   • [claimed-docs] “Reply to a review comment to ask for clarification”
   • [claimed-docs] “Interact with cubic in PR comments to ask questions, trigger reviews, and fix issues.”
   • [claimed-docs] “Interact with cubic in PR comments to ask questions, trigger reviews, and fix issues. You can tag `@cubic-dev-ai` directly, or reply to cubi…”
   • [claimed-docs] “Reply to its comments: Type a response explaining what cubic should do differently, and it remembers for next time”
   • [claimed-docs] “When a pull request needs extra scrutiny, you can ask cubic for an Ultrareview.”
   • [community] “I really liked it - it hit the mark. The current balance works very well, and it genuinely surprised me. It provides more technical explanat…”

   proof ↗⚑ flag

Control

1. developerI control when reviews run — skip drafts, trigger on demand, filter by branch or label — so the bot shows up only when wanted

   weight 1 · round to Greptile
   Greptile~partialclaimed5/10

   Greptile supports on-demand review via CLI (`greptile review`), configurable strictness/noise control, custom rules, and per-branch settings via greptile.json read from the source branch, plus generic 'configure review triggers' in quickstart — giving developers some control over when/how reviews run. However there is no explicit evidence of skipping draft PRs or filtering by label, which the story specifically calls out. Missing for 10: explicit draft-PR skip toggle, label-based filtering, and independent confirmation these controls work as documented.

   • [claimed-docs] “Set up Greptile AI code reviews in 5 minutes. Connect GitHub or GitLab, configure review triggers, and get automated feedback on your first …”
   • [claimed-docs] “Run `greptile review`, get a review of the feature branch diff in 60 seconds.”
   • [claimed-docs] “Control how strict Greptile is about leaving comments with the strictness setting (1–3).”
   • [claimed-docs] “Create custom rules that catch team-specific issues”
   • [claimed-docs] “Place `greptile.json` in your repository root. Settings are read from the source branch of the PR and override dashboard settings.”
   • [claimed-docs] “With noise control, Greptile limits reviews to high-signal insights, skipping low impact or repetitive feedback.”

   proof ↗⚑ flag

   cubic~partialclaimed4/10

   cubic clearly supports on-demand triggering (@cubic-dev-ai review this PR, Ultrareview on demand) and configurable behavior via cubic.yaml (ignore patterns, review behavior), giving developers some control over when reviews run. However there is no explicit evidence of skipping draft PRs or filtering by branch/label — cubic.yaml's 'ignore patterns' are documented generically, not tied to draft/branch/label filters. Missing for 10: explicit draft-PR skip setting, branch-based trigger rules, label-based filtering documentation.

   • [claimed-docs] “Once installed, cubic automatically reviews new pull requests.”
   • [claimed-docs] “To review a PR that was opened _before_ you installed the app, comment: `@cubic-dev-ai review this PR`.”
   • [claimed-docs] “When a pull request needs extra scrutiny, you can ask cubic for an Ultrareview.”
   • [claimed-docs] “Ultrareview is cubic's deepest review. It runs a longer, multi-pass analysis using cubic's most capable review models, and typically takes a…”
   • [claimed-docs] “`cubic.yaml` lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, …”
   • [claimed-docs] “cubic.yaml lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, an…”

   proof ↗⚑ flag

1. ai-native userDo everything through the API that I can do in the UI

   weight 2 · round drawn
   Greptile~partialprobed5/10

   Greptile exposes significant non-UI surfaces — a CLI (`greptile onboard`, `greptile review`) and an official MCP server (fetch PR comments, apply fixes, manage patterns) — that mirror parts of the UI experience (docs-14, docs-23, docs-24, docs-31, probe-4, probe-5). However, no OpenAPI/REST API spec was found (all candidate paths 404'd, greptile-probe-3), and several admin-level settings (e.g., 'Default Coding Agents' toggle, org onboarding via dashboard) are documented as UI-only (Settings → Code Review, docs-34), suggesting incomplete API/CLI parity for configuration and org management. missing for 10: a documented general-purpose REST/GraphQL API or OpenAPI spec, and evidence that all dashboard admin settings (strictness, default agents, org config) can be set via CLI/MCP rather than only the web UI.

   • [claimed-docs] “Run `greptile review`, get a review of the feature branch diff in 60 seconds.”
   • [claimed-docs] “Fetch PR comments - Get unaddressed Greptile feedback for any PR”
   • [claimed-docs] “Connect Cursor, Claude Code, VS Code, or Codex to Greptile MCP with OAuth.”
   • [claimed-docs] “`greptile onboard` creates your organization, connects GitHub or GitLab, enables repositories, and imports your existing AI rules files — th…”
   • [claimed-docs] “Org admins can go to **Settings → Code Review → Default Coding Agents** to turn on Fix with your Agent for the org.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.greptile.com/openapi.json, https://www.greptile.com/swagger.json, https://www.greptile.c…”
   • [probe] “official MCP server documented at https://www.greptile.com/docs/mcp-v2/overview”
   • [probe] “official CLI documented at https://www.greptile.com/docs/code-review/greptile-cli”

   proof ↗⚑ flag

   cubic~partialprobed5/10

   cubic exposes some programmatic surfaces beyond the UI — an Analytics API for PR-level metrics, an official MCP server for review findings/PR reviews/triage/subscription management, and a CLI for local review — but there is no general public REST/OpenAPI covering full platform functionality (probe found openapi.json/swagger.json all 404), so many UI features (wiki, custom agents config, integrations, analytics dashboards) lack documented API parity. missing for 10: a comprehensive public API/OpenAPI spec covering all UI capabilities, evidence that wiki/integrations/custom-agent configuration can be done via API, and confirmation that Analytics API + MCP together achieve full UI parity.

   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”
   • [claimed-docs] “Connect cubic's MCP server to your coding agent to read review findings and codebase context, request PR reviews, and triage PR or codebase …”
   • [claimed-docs] “You can now ask your coding agent to check your cubic subscription, manage team seats and roles, and purchase more seats without leaving you…”
   • [claimed-docs] “The **cubic CLI** reviews local changes before you push. It finds bugs and generates a prompt that your coding agent can use to fix them.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.cubic.dev/openapi.json, https://docs.cubic.dev/swagger.json, https://docs.cubic.dev/api…”
   • [probe] “official MCP server documented at https://docs.cubic.dev/ide/mcp-server”
   • [probe] “official CLI documented at https://docs.cubic.dev/ide/cli-review”

   proof ↗⚑ flag

2. ai-native userExport all of my data in open formats and leave

   weight 3 · round drawn
   Greptile—none0/10

   No evidence of any data export feature or open-format data portability for user data (reviews, patterns, learned rules, etc.); self-hosting only affects where data lives, not exportability. Self-hosted deployment keeps logs on customer servers, but there's no documented export mechanism or open-format dump users can take with them if they leave.

   • [claimed-docs] “Security page, self-hosting and logging: "For self-hosted Greptile services, Custom Apps are hosted using your own infrastructure - such as …”

   proof ↗⚑ flag

   cubic—none0/10

   Cubic is a code review/analysis platform whose data (review findings, PR comments, analytics, wiki) lives in GitHub, Jira, Notion, etc.; there is no evidence of a data export feature for users to download their cubic-generated data in open formats. The Analytics API (cubic-docs-16) offers programmatic access to some metrics but this is not documented as a full data export/portability mechanism, and no evidence addresses exporting AI wiki content, review history, or memory/learning data.

   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”
   • [claimed-docs] “cubic's AI wiki automatically indexes your codebase and produces searchable wikis, complete with links to source code, architecture diagrams…”

   proof ↗⚑ flag

3. ai-native userRead the product's source under an open license

   weight 2 · round drawn
   Greptile—none0/10

   Greptile is a closed, commercial SaaS/self-hosted product (deployed via Docker Compose/Kubernetes) with no evidence of an open-source license or publicly readable source code; self-hosting means running their binaries, not accessing source.

   • [claimed-docs] “Deploy Greptile in your own infrastructure with Docker Compose. Supports AWS, GCP, Azure, air-gapped environments, and custom LLM configurat…”
   • [claimed-docs] “Docker Compose: Up to 100 developers. Single VM, simpler operations. Kubernetes: 100+ developers. Horizontal scaling, high availability.”
   • [claimed-docs] “Security page, self-hosting and logging: "For self-hosted Greptile services, Custom Apps are hosted using your own infrastructure - such as …”

   proof ↗⚑ flag

   cubic—none0/10

   No evidence indicates cubic's source code is open or under an open license; cubic appears to be a closed, commercial SaaS product with SOC2 compliance and privacy policies but no public repository or license mentioned.

   ⚑ flag

4. ai-native userSelf-host the core product

   weight 3 · round to Greptile
   Greptile✓fullclaimed7/10

   Greptile documents a self-hosted deployment option via Docker Compose or Kubernetes, supporting AWS/GCP/Azure/air-gapped environments and custom/self-hosted LLMs, with logs stored only on customer servers. This is first-party documentation only, without independent/hands-on corroboration of a working self-hosted deployment. Missing for 10: independent verification that self-hosting actually works end-to-end, and clarity on whether 'core product' (all review/agent features) is fully feature-complete in self-hosted mode.

   • [claimed-docs] “Deploy Greptile in your own infrastructure with Docker Compose. Supports AWS, GCP, Azure, air-gapped environments, and custom LLM configurat…”
   • [claimed-docs] “Docker Compose: Up to 100 developers. Single VM, simpler operations. Kubernetes: 100+ developers. Horizontal scaling, high availability.”
   • [claimed-docs] “If your GitHub or GitLab instance restricts inbound traffic to specific IP addresses, you need to allowlist Greptile's IP range for code rev…”
   • [claimed-docs] “Security page, self-hosting and logging: "For self-hosted Greptile services, Custom Apps are hosted using your own infrastructure - such as …”

   proof ↗⚑ flag

   cubic—none0/10

   cubic is presented entirely as a hosted SaaS (GitHub app, cloud-based review engine, subscription/seats, analytics API); there is no evidence of a self-hosted/on-prem deployment option, Docker image, or open-source core anywhere in the docs.

   ⚑ flag

Platforms

1. developerThe reviewer installs as a GitHub/GitLab app and posts reviews as native inline comments on my pull requests within minutes

   weight 3 · round to cubic
   Greptile~partialcommunity7/10

   Docs clearly describe installing via GitHub/GitLab, automated PR reviews within 5 minutes (greptile-docs-5), and native review comments with actionable buttons (greptile-docs-1, greptile-docs-2, greptile-docs-36). Community testimony (greptile-comm-8) corroborates real-world use catching issues on PRs, but other community reports describe reliability issues (stuck processing, internal errors) with the underlying platform, and no independent source specifically confirms the 'minutes to install' claim or inline-comment UX end-to-end. missing for 10: independent hands-on confirmation of install speed and native inline-comment rendering, and resolution of community-reported reliability issues.

   • [claimed-docs] “Greptile is an AI code review agent that automatically reviews every pull request with complete understanding of your codebase.”
   • [claimed-docs] “Every review comment includes a Fix with your Agent button that sends the issue — with file paths, line numbers, and suggested code — straig…”
   • [claimed-docs] “Set up Greptile AI code reviews in 5 minutes. Connect GitHub or GitLab, configure review triggers, and get automated feedback on your first …”
   • [claimed-docs] “Every Greptile review comment includes a **Fix with your Agent** button. Click it, and the issue gets sent straight to your coding agent wit…”
   • [community] “My employer uses greptile and I'm pretty happy with it. Sometimes it can be a bit overzealous but more often than not it catches real issues…”
   • [community] “Not a single repo I've tried works. A lot of them seem not to have finished processing, but even the ones that have finished don't work.”
   • [community] “Ran it on my OSS project - stuck at 99% loading for ~30 minutes, then got 'Error: Internal error while locating sources' on every query. 'Th…”

   proof ↗⚑ flag

   cubic✓fullcommunity8/10

   Docs confirm installation as a GitHub/GitLab app with automatic reviews on new PRs, inline PR comments (@cubic-dev-ai, interactive comments, Fix with cubic), and a quickstart flow explicitly framed as 'minutes' with a free trial. Community feedback corroborates real-world usefulness of PR-level comments. Missing for 10: independent third-party confirmation of install speed and hands-on GitLab app usage (most evidence is first-party docs).

   • [claimed-docs] “Once installed, cubic automatically reviews new pull requests.”
   • [claimed-docs] “Spots bugs and improvements: Comments on bugs and improvements in pull requests”
   • [claimed-docs] “Interact with cubic in PR comments to ask questions, trigger reviews, and fix issues.”
   • [claimed-docs] “Get started with cubic AI code review in minutes.”
   • [claimed-docs] “Post this comment on GitHub to start a review: text theme={null} @cubic-dev-ai review this PR ”
   • [claimed-docs] “7 days free, no credit card required.”
   • [community] “I really liked it - it hit the mark. The current balance works very well, and it genuinely surprised me. It provides more technical explanat…”

   proof ↗⚑ flag

Suggestions

1. developerReview comments include committable suggested diffs I can apply with one click

   weight 3 · round to cubic
   Greptile~partialclaimed5/10

   Greptile's review comments include a one-click 'Fix with your Agent' button that sends suggested code with file/line context to an external coding agent (Claude Code, Cursor, Codex, Devin) to apply the fix, and a 'Fix All' button for bulk sends (greptile-docs-2, greptile-docs-3, greptile-docs-36, greptile-docs-10). This is a one-click flow but requires routing through a separate coding agent to actually commit the change, rather than a native inline 'commit suggestion' diff applied directly in the PR. Missing for 10: evidence of a native GitHub-style suggested-diff with a direct 'commit suggestion' button that applies without invoking an external agent, and independent/hands-on confirmation that the agent-applied fix actually commits cleanly.

   • [claimed-docs] “Every review comment includes a Fix with your Agent button that sends the issue — with file paths, line numbers, and suggested code — straig…”
   • [claimed-docs] “A Fix All button in the review summary sends every issue at once.”
   • [claimed-docs] “Launch any agent in one click with a Greptile comment, full context, and suggested fix.”
   • [claimed-docs] “Every Greptile review comment includes a **Fix with your Agent** button. Click it, and the issue gets sent straight to your coding agent wit…”
   • [claimed-docs] “Org admins can go to **Settings → Code Review → Default Coding Agents** to turn on Fix with your Agent for the org.”

   proof ↗⚑ flag

   cubic✓fullclaimed7/10

   Cubic explicitly supports one-click fixes: reviewers can click 'Fix with cubic' and cubic generates the fix and pushes it directly to the PR branch (cubic-docs-4, cubic-docs-29, cubic-docs-53), fulfilling the core 'one-click apply' story for review comments. Missing for 10: no evidence of an inline GitHub-style suggested-diff block shown for manual approval before commit, and no independent/hands-on corroboration of this specific flow beyond first-party docs.

   • [claimed-docs] “cubic generates the fix and pushes it to your PR branch.”
   • [claimed-docs] “cubic can automatically fix issues in the background. When the AI spots a problem, you can click the **Fix with cubic** button”
   • [claimed-docs] “cubic can automatically fix issues identified during code review. Request a targeted fix with one click.”
   • [claimed-docs] “Coding agents: Generate fixes when requested, using your configured provider”

   proof ↗⚑ flag

Summaries

1. developerEvery PR gets an auto-generated summary and change walkthrough so human reviewers orient fast

   weight 2 · round to cubic
   Greptile✓fullcommunity7/10

   Greptile's docs describe automatic review of every PR with full codebase context, a 0-5 merge-safety score, and auto-generated sequence diagrams/flowcharts for every PR specifically to help parse complex changes — directly matching the 'summary + walkthrough' story. Community evidence (comm-8) corroborates real-world use where Greptile catches issues before human review, giving reviewers a head start. Missing for 10: no independent verification of diagram/summary quality or accuracy, and no evidence separating 'summary' as a distinct named artifact from the diagrams/score/comments bundle.

   • [claimed-docs] “Greptile is an AI code review agent that automatically reviews every pull request with complete understanding of your codebase.”
   • [claimed-docs] “Know how safe a PR is to merge at a glance with a 0-5 score”
   • [claimed-docs] “Parse complex changes with sequence diagrams and flowcharts generated for every PR”
   • [community] “My employer uses greptile and I'm pretty happy with it. Sometimes it can be a bit overzealous but more often than not it catches real issues…”

   proof ↗⚑ flag

   cubic✓fullcommunity9/10

   cubic explicitly generates PR summaries/descriptions from code changes automatically on every PR (docs-21, docs-57), plus keeps the summary visible alongside the diff while scrolling for easier orientation (docs-43), and community feedback corroborates depth of explanations (cubic-comm-1). Missing for 10: independent/hands-on review specifically confirming summary accuracy or walkthrough quality at scale beyond docs and one community quote.

   • [claimed-docs] “PR summaries: Generates PR descriptions based on code changes”
   • [claimed-docs] “cubic helps your team spend less time writing PR descriptions automatically by generating clear, concise summaries.”
   • [claimed-docs] “Now, the group description stays beside the diff as you scroll, so you can read the code and its explanation together.”
   • [claimed-docs] “Spots bugs and improvements: Comments on bugs and improvements in pull requests”
   • [community] “I really liked it - it hit the mark. The current balance works very well, and it genuinely surprised me. It provides more technical explanat…”

   proof ↗⚑ flag

Updates

1. developerPushing new commits triggers an incremental re-review that tracks what was fixed instead of repeating old comments

   weight 2 · round to Greptile
   Greptile~partialclaimed7/10

   Docs describe Greptile tracking 'unaddressed' vs resolved comments and iterating until a PR reaches a perfect review (zero unresolved comments), implying it distinguishes fixed issues from outstanding ones across pushes, and the learning system adapts based on reactions to avoid repeating unwanted comments. However, there's no explicit first-party statement that a new commit push specifically triggers a fresh incremental review pass distinct from prior comments, and no independent/hands-on evidence confirming this exact re-review behavior in practice. Missing for 10: explicit doc on push-triggered incremental review mechanics, and community/hands-on confirmation that old fixed comments are not repeated after new commits.

   • [claimed-docs] “Iteratively fix a PR until Greptile gives a perfect review: 5/5 confidence, zero unresolved comments.”
   • [claimed-docs] “Fetch PR comments - Get unaddressed Greptile feedback for any PR”
   • [claimed-docs] “Greplooping Iteratively fix a PR until Greptile gives a perfect review: 5/5 confidence, zero unresolved comments.”
   • [claimed-docs] “List unaddressed Greptile comments for PR #5 in owner/repo”
   • [claimed-docs] “Your 👍/👎 reactions and replies teach Greptile what matters. After 2-3 weeks, it stops commenting on things you don't care about.”
   • [claimed-docs] “Greptile learns from your team's feedback to provide increasingly relevant suggestions. The primary training methods are emoji reactions and…”

   proof ↗⚑ flag

   cubic~partialclaimed6/10

   cubic auto-reviews new PRs and auto-resolves threads when issues are fixed (cubic-docs-1, cubic-docs-20), and its memory/learning system means feedback and prior comments inform future reviews (cubic-docs-10, cubic-docs-54), which together imply incremental re-review behavior rather than repeating stale comments. However, there's no explicit documentation describing a re-review triggered specifically by new commits that diffs against prior findings or explicitly states old fixed issues won't be re-flagged. missing for 10: explicit docs on commit-triggered incremental re-review logic, confirmation that previously-flagged-and-fixed issues are suppressed on subsequent pushes, and independent/hands-on evidence of this specific workflow.

   • [claimed-docs] “Once installed, cubic automatically reviews new pull requests.”
   • [claimed-docs] “Auto-resolve threads: Enable automatic thread resolution to close findings when the issue is fixed”
   • [claimed-docs] “cubic automatically learns when you: Reply to its comments... React to suggestions: Thumbs up/down help cubic calibrate what's useful vs noi…”
   • [claimed-docs] “cubic learns your team's unwritten rules from two sources: direct feedback on its comments, and patterns extracted from your senior engineer…”

   proof ↗⚑ flag

1. ai-native userChoose where my data is stored (region/residency)

   weight 2 · round to Greptile
   Greptile~partialclaimed6/10

   Greptile lets customers self-host the entire platform on their own infrastructure (AWS, GCP, Azure, or air-gapped) inside their own VPC, which effectively gives full control over where data physically resides, and logging can be disabled entirely in that mode. However, there is no documented region-selection feature for the standard hosted/SaaS offering (e.g., no EU vs US data-residency toggle) — residency is achieved only via the heavier self-hosting path, not as a first-class SaaS setting. Missing for 10: explicit hosted multi-region data residency options, third-party confirmation of self-host residency compliance (SOC2/GDPR data residency certifications).

   • [claimed-docs] “Deploy Greptile in your own infrastructure with Docker Compose. Supports AWS, GCP, Azure, air-gapped environments, and custom LLM configurat…”
   • [claimed-docs] “Docker Compose: Up to 100 developers. Single VM, simpler operations. Kubernetes: 100+ developers. Horizontal scaling, high availability.”
   • [claimed-docs] “Security page, self-hosting and logging: "For self-hosted Greptile services, Custom Apps are hosted using your own infrastructure - such as …”
   • [claimed-docs] “Security page, De-Identified Data and AI Training (vendor's own carve-out): "Greptile may aggregate and anonymize Customer Data ('De-Identif…”

   proof ↗⚑ flag

   cubic—none0/10

   No evidence of data residency/region selection options in cubic's documentation; privacy/security docs mention SOC 2 compliance and no third-party model training, but nothing about choosing storage region or data residency controls.

   • [claimed-docs] “Privacy & security: "No third-party model training on Customer Code — Our AI model providers (e.g., OpenAI, Anthropic) contractually commit …”
   • [claimed-docs] “Privacy & security: "cubic is SOC 2 Type 1 compliant, demonstrating our commitment to security and trust for our customers. We maintain comp…”

   proof ↗⚑ flag

2. ai-native userPrevent my data from being used to train AI models

   weight 3 · round to cubic
   Greptile~partialclaimed4/10

   Greptile's own security page states it may use de-identified customer data for 'AI Training and Learnings' as a standing carve-out, with no explicit toggle to opt out of this for standard hosted customers — only self-hosted deployments (with logging fully disabled and data kept in the customer's own VPC) offer a documented path to keep data out of Greptile's hands. This gives privacy-focused users a workaround (self-host) but not a clean, guaranteed 'no AI training' control for the mainstream SaaS offering. Missing for 10: an explicit opt-out/toggle for hosted customers to disable AI-training use of their (even de-identified) data, and independent confirmation that self-hosted mode fully excludes any data from training.

   • [claimed-docs] “Security page, self-hosting and logging: "For self-hosted Greptile services, Custom Apps are hosted using your own infrastructure - such as …”
   • [claimed-docs] “Security page, De-Identified Data and AI Training (vendor's own carve-out): "Greptile may aggregate and anonymize Customer Data ('De-Identif…”

   proof ↗⚑ flag

   cubic✓fullclaimed8/10

   cubic's privacy/security docs explicitly state model providers contractually commit to not train on customer code, and offer a full opt-out to block AI features entirely for a workspace. missing for 10: independent/third-party audit confirming the no-training claim, and detail on how the opt-out mechanism is enforced technically.

   • [claimed-docs] “Privacy & security: "No third-party model training on Customer Code — Our AI model providers (e.g., OpenAI, Anthropic) contractually commit …”

   proof ↗⚑ flag

3. ai-native userControl data retention and deletion

   weight 2 · round to Greptile
   Greptile~partialclaimed5/10

   Greptile's security docs let customers self-host so logs stay only on their own servers and let customers turn off logging entirely for 100% private chats, which is meaningful retention control (greptile-supp-1). However, the same page discloses a standing carve-out that Greptile may aggregate/anonymize Customer Data for AI training even when logging is limited (greptile-supp-2), and there is no documented self-service deletion mechanism, retention period, or explicit 'delete my data' control for the standard SaaS offering. Missing for 10: explicit data-deletion API/workflow for cloud customers, documented retention windows, and clarity on how the AI-training carve-out interacts with opt-out settings.

   • [claimed-docs] “Security page, self-hosting and logging: "For self-hosted Greptile services, Custom Apps are hosted using your own infrastructure - such as …”
   • [claimed-docs] “Security page, De-Identified Data and AI Training (vendor's own carve-out): "Greptile may aggregate and anonymize Customer Data ('De-Identif…”

   proof ↗⚑ flag

   cubic~partialclaimed3/10

   cubic's privacy/security docs mention no third-party model training and the ability to block AI features for a workspace, showing some data-handling controls, but there is no explicit mention of data retention periods, deletion requests, or a self-service deletion mechanism. missing for 10: explicit retention policy, user-initiated deletion process, confirmation of deletion timelines/scope.

   • [claimed-docs] “Privacy & security: "No third-party model training on Customer Code — Our AI model providers (e.g., OpenAI, Anthropic) contractually commit …”
   • [claimed-docs] “Privacy & security: "cubic is SOC 2 Type 1 compliant, demonstrating our commitment to security and trust for our customers. We maintain comp…”

   proof ↗⚑ flag

4. ai-native userOpt out of telemetry and usage tracking

   weight 2 · round to Greptile
   Greptile~partialclaimed4/10

   Self-hosted deployments let customers 'turn off logging and make chats 100% private' (greptile-supp-1), but the standard/SaaS offering explicitly reserves the right to aggregate and anonymize customer data for 'AI Training and Learnings' (greptile-supp-2) with no documented opt-out mechanism for that telemetry. Missing for 10: an explicit SaaS-tier telemetry/usage-tracking opt-out toggle, and independent confirmation that opting out actually stops data collection.

   • [claimed-docs] “Security page, self-hosting and logging: "For self-hosted Greptile services, Custom Apps are hosted using your own infrastructure - such as …”
   • [claimed-docs] “Security page, De-Identified Data and AI Training (vendor's own carve-out): "Greptile may aggregate and anonymize Customer Data ('De-Identif…”

   proof ↗⚑ flag

   cubic—none0/10

   Evidence covers data-training and access-control privacy commitments (no third-party model training, SOC2 compliance, ability to block AI features workspace-wide) but no documented telemetry/usage-tracking toggle or opt-out mechanism specifically for product analytics/telemetry.

   ⚑ flag

Analytics

1. engineering leadI see dashboards of findings, acceptance rates, and review coverage across my org

   weight 1 · round to cubic
   Greptile—none0/10

   The evidence describes per-PR review comments, strictness/noise-control settings, learning from feedback, and CLI/MCP integrations, but no org-level dashboard aggregating findings, acceptance rates, or review coverage metrics across teams/repos is documented anywhere in the pack.

   ⚑ flag

   cubic✓fullclaimed7/10

   cubic's Analytics dashboard explicitly covers PR-level findings/fixes, AI coding and AI review impact tabs, and delivery/effectiveness across repos, directly matching the engineering-lead dashboard story (cubic-docs-16, 41, 42, 59, 60). Missing for 10: explicit org-wide rollup UI screenshots, exact 'acceptance rate' terminology/metric definition, and independent/hands-on corroboration beyond vendor docs.

   • [claimed-docs] “The Analytics API gives you PR-level data on how many issues were flagged, how many were fixed, how much AI code was authored, etc.”
   • [claimed-docs] “The AI coding tab shows how AI assistants contribute to merged pull requests in your repositories.”
   • [claimed-docs] “The AI review tab shows whether cubic reviews are helping your team ship better code faster.”
   • [claimed-docs] “The analytics dashboard shows how your team ships code across three lenses: AI coding usage, AI review impact, and delivery speed.”
   • [claimed-docs] “The Delivery and effectiveness tab focuses on how quickly PRs move from creation to merge and where bottlenecks appear.”

   proof ↗⚑ flag

Gates

1. engineering leadThe reviewer can gate merges — a required status check or blocking review that enforces resolution of critical findings

   weight 2 · round drawn
   Greptile—none0/10

   Greptile documents review scoring (0-5 safety score, 5/5 confidence) and comment resolution tracking, but no evidence describes a required GitHub/GitLab status check, branch-protection integration, or a blocking review mechanism that would actually prevent a merge until critical findings are resolved.

   ⚑ flag

   cubic—none0/10

   Cubic documents auto-approval (skipping human review for low-risk PRs) and auto-resolving comment threads, but there is no evidence of a required GitHub status check or blocking review policy that prevents merge until critical findings are resolved — the auto-approval feature actually works in the opposite direction (approving rather than blocking). Missing for full credit: documentation of a required/blocking check integration, branch-protection enforcement, or explicit 'block merge on critical findings' policy.

   • [claimed-docs] “Enable [auto-approval](/ai-review/auto-approval) to let cubic approve PRs that meet your review policy.”
   • [claimed-docs] “Auto-resolve threads: Enable automatic thread resolution to close findings when the issue is fixed”
   • [claimed-docs] “Auto-approval lets you skip human review for pull requests that cubic determines are low risk and issue-free.”

   proof ↗⚑ flag

Detection

1. developerThe reviewer catches real bugs in my PR — logic errors, race conditions, broken edge cases — not just style nits

   weight 3 · round to cubic
   Greptile~partialcommunity6/10

   Docs describe substantive bug-catching capability beyond style — TREX runtime sandbox that surfaces race conditions/logic bugs via traces and screenshots, static+AI security scanning, and v5's parallel agent swarm claiming fewer false positives and more real bugs caught. Community evidence partially corroborates real-world value (comm-8: 'catches real issues... gives author a chance to fix before human reviews') but also shows mixed accuracy experiences on complex code (comm-3: incomplete/incorrect answers on C++ templates) and reliability hiccups reported for the platform generally. Missing for 10: independent, PR-review-specific benchmarks or hands-on accounts confirming it reliably finds logic errors/race conditions (not just Q&A accuracy), and no rebuttal to the noted accuracy gaps.

   • [claimed-docs] “TREX runs your PR branch in a sandbox to find the bugs that only show up at runtime. It starts services, mocks inputs, clicks through UI flo…”
   • [claimed-docs] “Greptile pairs static scanning with an AI security agent to catch vulnerabilities in every pull request.”
   • [claimed-docs] “v5 is faster, catches more bugs, and flags fewer false positives by running a swarm of narrowly scoped agents in parallel.”
   • [community] “My employer uses greptile and I'm pretty happy with it. Sometimes it can be a bit overzealous but more often than not it catches real issues…”
   • [community] “Tried it on my own C++ codebase - impressed it found the C++ standard used, but it's really bad at analyzing templates; answers always incom…”

   proof ↗⚑ flag

   cubic✓fullcommunity7/10

   Cubic's docs explicitly claim to catch logic errors, bugs, and security vulnerabilities (not just style), with deep Ultrareview passes for complex refactors, and one independent HN commenter corroborates that it 'provides more technical explanations than just basic checks' and is 'really powerful.' However, race conditions and specific edge-case handling are not explicitly evidenced, and most support is vendor documentation rather than hands-on benchmarks. Missing for 10: independent benchmark/hands-on evidence of catching race conditions or subtle logic bugs specifically, broader third-party validation beyond a single HN comment.

   • [claimed-docs] “Spots bugs and improvements: Comments on bugs and improvements in pull requests”
   • [claimed-docs] “The AI automatically scans every pull request for a wide range of issues, including logic errors, style inconsistencies, and potential secur…”
   • [claimed-docs] “When a pull request needs extra scrutiny, you can ask cubic for an Ultrareview.”
   • [claimed-docs] “Ultrareview is cubic's deepest review. It runs a longer, multi-pass analysis using cubic's most capable review models, and typically takes a…”
   • [community] “I really liked it - it hit the mark. The current balance works very well, and it genuinely surprised me. It provides more technical explanat…”

   proof ↗⚑ flag

Learning

1. developerPush back on a bad review comment and the reviewer learns — it stops repeating the same rejected feedback

   weight 2 · round to cubic
   Greptile~partialclaimed7/10

   Docs explicitly describe a learning loop where 👍/👎 reactions and reply comments train Greptile to stop repeating feedback team members don't want, with noise control filtering repetitive comments (greptile-docs-4, greptile-docs-33, greptile-docs-32). However, this is described as a team-wide, multi-week aggregate learning process rather than an immediate per-developer 'push back once and it never repeats' guarantee, and no independent/hands-on evidence confirms the mechanism works as described. Missing for 10: independent corroboration of the learning behavior, and clarity on whether a single developer's pushback (vs. team consensus over weeks) suppresses a specific rejected comment.

   • [claimed-docs] “Your 👍/👎 reactions and replies teach Greptile what matters. After 2-3 weeks, it stops commenting on things you don't care about.”
   • [claimed-docs] “Greptile learns from your team's feedback to provide increasingly relevant suggestions. The primary training methods are emoji reactions and…”
   • [claimed-docs] “With noise control, Greptile limits reviews to high-signal insights, skipping low impact or repetitive feedback.”

   proof ↗⚑ flag

   cubic✓fullcommunity7/10

   cubic explicitly documents that replying to its comments teaches it 'what to do differently' and it 'remembers for next time,' with learning drawn from direct feedback and thumbs up/down reactions plus senior engineers' review patterns — directly matching the story of pushing back and the reviewer not repeating rejected feedback. However, this is entirely first-party documentation with no independent/hands-on confirmation that the learning actually persists or stops repeat comments in practice; one community comment expresses general skepticism about YAML-based rule interpretation but doesn't concretely show the learning failing. Missing for 10: independent hands-on verification that cubic stops repeating rejected feedback, and details on how long/broadly the learned preference applies.

   • [claimed-docs] “cubic automatically learns when you: Reply to its comments... React to suggestions: Thumbs up/down help cubic calibrate what's useful vs noi…”
   • [claimed-docs] “Reply to its comments: Type a response explaining what cubic should do differently, and it remembers for next time”
   • [claimed-docs] “React to suggestions: Thumbs up/down help cubic calibrate what's useful vs noise for your specific codebase”
   • [claimed-docs] “cubic learns your team's unwritten rules from two sources: direct feedback on its comments, and patterns extracted from your senior engineer…”
   • [community] “The 'in-codebase' approach is the right one, but a YAML file with plain text is a half-measure. The most reliable rule that 'lives next to t…”

   proof ↗⚑ flag

Noise

1. engineering leadThe reviewer keeps noise low — few false positives, deduplicated comments, severity labels — so my team doesn't tune it out

   weight 3 · round drawn
   Greptile~partialcommunity6/10

   Greptile documents multiple noise-reduction mechanisms: adjustable strictness/nitpickiness (docs-20, docs-32), a learning system trained via 👍/👎 reactions that stops flagging things teams don't care about (docs-4, docs-33), custom rules to catch team-specific issues (docs-21), and a v5 update claiming 'fewer false positives' via parallel narrow-scope agents (docs-26). However, hands-on community evidence is mixed: one enterprise user reports Greptile is 'sometimes overzealous' even while generally catching real issues (greptile-comm-8), indicating the noise-reduction promise isn't fully realized in practice, and there's no independent data on deduplication or severity-labeling accuracy. Missing for 10: independent verification of false-positive rate reduction, evidence of comment deduplication specifically, and confirmation severity labels are reliably applied and trusted by teams.

   • [claimed-docs] “Your 👍/👎 reactions and replies teach Greptile what matters. After 2-3 weeks, it stops commenting on things you don't care about.”
   • [claimed-docs] “Control how strict Greptile is about leaving comments with the strictness setting (1–3).”
   • [claimed-docs] “With noise control, Greptile limits reviews to high-signal insights, skipping low impact or repetitive feedback.”
   • [claimed-docs] “Greptile learns from your team's feedback to provide increasingly relevant suggestions. The primary training methods are emoji reactions and…”
   • [claimed-docs] “v5 is faster, catches more bugs, and flags fewer false positives by running a swarm of narrowly scoped agents in parallel.”
   • [claimed-docs] “Create custom rules that catch team-specific issues”
   • [community] “My employer uses greptile and I'm pretty happy with it. Sometimes it can be a bit overzealous but more often than not it catches real issues…”

   proof ↗⚑ flag

   cubic~partialcommunity6/10

   cubic documents concrete noise-reduction mechanisms: it learns from thumbs-up/down and reply feedback to calibrate 'what's useful vs noise' (cubic-docs-10, cubic-docs-32/33), learns unwritten rules from senior engineers' historical reviews (cubic-docs-54), supports ignore patterns/custom agents via cubic.yaml (cubic-docs-11), and auto-resolves threads once issues are fixed (cubic-docs-20). One community reviewer corroborates quality/technical depth (cubic-comm-1), while another raises skepticism about YAML-based rule reliability (cubic-comm-2), though this doesn't concretely contradict the noise-reduction claims. Missing for 10: explicit severity labeling of comments, explicit deduplication-of-comments mechanism, and independent data/benchmarks on false-positive rates.

   • [claimed-docs] “cubic automatically learns when you: Reply to its comments... React to suggestions: Thumbs up/down help cubic calibrate what's useful vs noi…”
   • [claimed-docs] “Auto-resolve threads: Enable automatic thread resolution to close findings when the issue is fixed”
   • [claimed-docs] “Reply to its comments: Type a response explaining what cubic should do differently, and it remembers for next time”
   • [claimed-docs] “React to suggestions: Thumbs up/down help cubic calibrate what's useful vs noise for your specific codebase”
   • [claimed-docs] “cubic learns your team's unwritten rules from two sources: direct feedback on its comments, and patterns extracted from your senior engineer…”
   • [claimed-docs] “`cubic.yaml` lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, …”
   • [community] “I really liked it - it hit the mark. The current balance works very well, and it genuinely surprised me. It provides more technical explanat…”
   • [community] “The 'in-codebase' approach is the right one, but a YAML file with plain text is a half-measure. The most reliable rule that 'lives next to t…”

   proof ↗⚑ flag

Security

1. security engineerReviews flag security problems in the diff — injection risks, leaked secrets, insecure patterns — alongside functional bugs

   weight 2 · round to Greptile
   Greptile✓fullcommunity6/10

   Greptile has a dedicated security-check feature explicitly pairing static scanning with an AI security agent to catch vulnerabilities on every PR, alongside its general review pipeline that flags functional bugs and code-quality issues in the same diff comments. However, missing for 10: independent/hands-on evidence confirming detection of specific vuln classes like injection or leaked secrets, and no community reports specifically validating security-finding accuracy (community feedback only covers general bug-catching, not security).

   • [claimed-docs] “Greptile pairs static scanning with an AI security agent to catch vulnerabilities in every pull request.”
   • [claimed-docs] “Greptile is an AI code review agent that automatically reviews every pull request with complete understanding of your codebase.”
   • [claimed-docs] “v5 is faster, catches more bugs, and flags fewer false positives by running a swarm of narrowly scoped agents in parallel.”
   • [community] “My employer uses greptile and I'm pretty happy with it. Sometimes it can be a bit overzealous but more often than not it catches real issues…”

   proof ↗⚑ flag

   cubic~partialcommunity6/10

   cubic docs explicitly claim it scans PRs for 'security vulnerabilities' alongside logic errors and style issues, and offers an 'Ultrareview' deep-pass mode explicitly recommended for security-sensitive changes, plus custom agents to enforce org-specific rules. However, there's no documented evidence of specific detection categories like secret leakage or injection-pattern examples, nor independent/hands-on evidence confirming accuracy on real security bugs (community quotes address general usefulness, not security-specific findings). Missing for 10: explicit mention of secret-scanning/injection-detection categories, and independent verification of security-finding accuracy.

   • [claimed-docs] “The AI automatically scans every pull request for a wide range of issues, including logic errors, style inconsistencies, and potential secur…”
   • [claimed-docs] “When a pull request needs extra scrutiny, you can ask cubic for an Ultrareview.”
   • [claimed-docs] “Ultrareview is cubic's deepest review. It runs a longer, multi-pass analysis using cubic's most capable review models, and typically takes a…”
   • [claimed-docs] “Custom agents are review rules that enforce your organization's specific best practices across pull requests.”
   • [community] “I really liked it - it hit the mark. The current balance works very well, and it genuinely surprised me. It provides more technical explanat…”

   proof ↗⚑ flag

Cli

1. developerI run reviews from a CLI against local diffs or in CI scripts, with machine-readable output my tooling can consume

   weight 2 · round drawn
   Greptile~partialprobed5/10

   Greptile ships an official CLI (`greptile review`, `greptile onboard`) that reviews local/feature-branch diffs and returns a 0-5 safety score, and docs note coding agents can invoke the CLI to review their own code, implying scriptable/automatable use. However, there is no explicit documentation of a machine-readable output format (e.g., JSON flag) or dedicated CI pipeline integration examples — only an OpenAPI/CLI probe confirming the CLI page exists with no schema found. Missing for 10: documented JSON/structured output mode, explicit CI script examples, and independent confirmation of CLI reliability in automation contexts.

   • [claimed-docs] “Run `greptile review`, get a review of the feature branch diff in 60 seconds.”
   • [claimed-docs] “Coding agents can use the greptile cli to review their own code”
   • [claimed-docs] “Set up Greptile from your terminal with `greptile onboard` — or hand this page to your coding agent and have it run the setup for you.”
   • [claimed-docs] “`greptile onboard` creates your organization, connects GitHub or GitLab, enables repositories, and imports your existing AI rules files — th…”
   • [probe] “official CLI documented at https://www.greptile.com/docs/code-review/greptile-cli”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.greptile.com/openapi.json, https://www.greptile.com/swagger.json, https://www.greptile.c…”

   proof ↗⚑ flag

   cubic~partialprobed5/10

   cubic ships an official CLI (`cubic-docs-7`, `cubic-probe-5`) that reviews local diffs before push and outputs a prompt for a coding agent to fix issues, satisfying the 'local diff' half of the story. However there is no documented evidence of CI-specific usage (e.g., a CI mode/flag) or of structured machine-readable output (JSON/SARIF) that external tooling could consume — the CLI output is described as a natural-language prompt, not a machine-parseable format. missing for 10: CI integration docs/flags, machine-readable (JSON/SARIF) output format, examples of scripting/tooling consumption.

   • [claimed-docs] “The **cubic CLI** reviews local changes before you push. It finds bugs and generates a prompt that your coding agent can use to fix them.”
   • [probe] “official CLI documented at https://docs.cubic.dev/ide/cli-review”

   proof ↗⚑ flag

Ide

1. developerI get the same review inside my IDE before I push, catching issues while the code is still in my editor

   weight 2 · round to cubic
   Greptile~partialprobed6/10

   Greptile's core product reviews pull requests after they are opened, but its CLI (`greptile review`) can review a feature-branch diff locally in ~60 seconds, and its MCP server lets Cursor/Claude Code/VS Code/Codex surface review comments and patterns directly in the editor, which together approximate a pre-push, in-IDE review loop. However, this is CLI/MCP-triggered rather than a live, continuous in-editor reviewer, and there is no independent/hands-on evidence confirming the pre-push workflow works as described. Missing for 10: evidence of automatic/real-time in-editor annotations (not just CLI/MCP triggered), independent user confirmation of the pre-push flow, and clarity that MCP comments are available before a PR/push exists rather than only fetching existing PR feedback.

   • [claimed-docs] “Run `greptile review`, get a review of the feature branch diff in 60 seconds.”
   • [claimed-docs] “Coding agents can use the greptile cli to review their own code”
   • [claimed-docs] “Connect Cursor, Claude Code, VS Code, or Codex to Greptile MCP with OAuth.”
   • [claimed-docs] “With MCP, you can view, search, and create patterns from your IDE.”
   • [claimed-docs] “View and resolve review comments directly from Claude Code.”
   • [probe] “official MCP server documented at https://www.greptile.com/docs/mcp-v2/overview”
   • [probe] “official CLI documented at https://www.greptile.com/docs/code-review/greptile-cli”

   proof ↗⚑ flag

   cubic✓fullprobed7/10

   cubic ships a dedicated CLI that "reviews local changes before you push. It finds bugs and generates a prompt that your coding agent can use to fix them" (cubic-docs-7, cubic-probe-5), plus IDE/agent integration via cubic skills and an MCP server that lets coding agents read review findings and codebase context (cubic-docs-8, cubic-docs-35, cubic-probe-4), directly matching the story of catching issues pre-push in the editor. Missing for 10: independent/hands-on validation of the CLI's pre-push review quality and no community evidence specifically confirming the IDE experience (community citations discuss PR-level review, not the local CLI).

   • [claimed-docs] “The **cubic CLI** reviews local changes before you push. It finds bugs and generates a prompt that your coding agent can use to fix them.”
   • [claimed-docs] “Connect cubic's MCP server to your coding agent to read review findings and codebase context, request PR reviews, and triage PR or codebase …”
   • [claimed-docs] “Install and connect cubic in Cursor, Claude Code, VS Code, Codex, Gemini CLI, or another coding tool.”
   • [claimed-docs] “cubic skills give your coding agent instructions for reviewing code, handling PR comments, and using your team's codebase context.”
   • [probe] “official MCP server documented at https://docs.cubic.dev/ide/mcp-server”
   • [probe] “official CLI documented at https://docs.cubic.dev/ide/cli-review”

   proof ↗⚑ flag

Config

1. engineering leadI configure the reviewer with a versioned config file in my repo — path filters, per-path instructions, review profiles

   weight 3 · round to cubic
   Greptile~partialclaimed6/10

   Greptile documents a versioned `greptile.json` config file placed in the repo root that is read from the PR's source branch and overrides dashboard settings, supporting per-repo, version-controlled configuration; related settings like custom rules and strictness/nitpickiness controls are also documented as configurable dials. However, the evidence pack never explicitly confirms path-based filters, per-path instructions, or named 'review profiles' within that config schema — only that dashboard-level settings (rules, strictness) can be overridden via the file. Missing for 10: explicit greptile.json schema showing path-filter syntax, per-path instruction blocks, and named review-profile support.

   • [claimed-docs] “Place `greptile.json` in your repository root. Settings are read from the source branch of the PR and override dashboard settings.”
   • [claimed-docs] “Create custom rules that catch team-specific issues”
   • [claimed-docs] “Control how strict Greptile is about leaving comments with the strictness setting (1–3).”
   • [claimed-docs] “`greptile onboard` creates your organization, connects GitHub or GitLab, enables repositories, and imports your existing AI rules files — th…”

   proof ↗⚑ flag

   cubic✓fullcommunity7/10

   cubic.yaml is documented as a versioned, repo-root config file that is the 'source of truth for AI review behavior, ignore patterns, PR descriptions, and custom agents,' directly covering path filters (ignore patterns) and review profiles (custom agents, Ultrareview). Custom context files (README.md, AGENTS.md, .cursorrules, .ai/ dirs) further support per-path/contextual instructions. missing for 10: an explicit documented example of per-path instruction syntax within cubic.yaml, and independent/hands-on verification of the config file working as described (one community comment even questions whether YAML-based rules are as robust as code-based checks).

   • [claimed-docs] “`cubic.yaml` lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, …”
   • [claimed-docs] “cubic.yaml lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, an…”
   • [claimed-docs] “Custom agents are review rules that enforce your organization's specific best practices across pull requests.”
   • [claimed-docs] “Custom agents: Enforce your team's coding standards”
   • [claimed-docs] “cubic automatically detects and uses context files in your repository, including `README.md`, `context.md`, `AGENTS.md`, `.cursorrules`”
   • [claimed-docs] “cubic automatically detects and uses context files in your repository, including `README.md`, `context.md`, `AGENTS.md`, `.cursorrules`, age…”
   • [community] “The 'in-codebase' approach is the right one, but a YAML file with plain text is a half-measure. The most reliable rule that 'lives next to t…”

   proof ↗⚑ flag

Governance

1. engineering leadI roll out org-level review defaults across hundreds of repos and manage exceptions centrally

   weight 1 · round to Greptile
   Greptile~partialclaimed7/10

   Docs show org-level dashboard settings that apply by default, with per-repo greptile.json files that override those defaults for exceptions (greptile-docs-22), org admin controls like Settings → Code Review → Default Coding Agents (greptile-docs-34), and CLI onboarding that creates an org, connects GitHub/GitLab, and enables repositories in bulk (greptile-docs-31). This covers the core mechanics of setting org-wide defaults and per-repo exceptions, but there's no direct evidence of managing hundreds of repos at scale, bulk exception dashboards, role-based admin permissions, or audit/reporting tools for centralized governance. Missing for 10: explicit multi-repo/org-scale management UI, RBAC for admins, and evidence of exception tracking across hundreds of repos.

   • [claimed-docs] “Place `greptile.json` in your repository root. Settings are read from the source branch of the PR and override dashboard settings.”
   • [claimed-docs] “Org admins can go to **Settings → Code Review → Default Coding Agents** to turn on Fix with your Agent for the org.”
   • [claimed-docs] “`greptile onboard` creates your organization, connects GitHub or GitLab, enables repositories, and imports your existing AI rules files — th…”
   • [claimed-docs] “Control how strict Greptile is about leaving comments with the strictness setting (1–3).”
   • [claimed-docs] “Create custom rules that catch team-specific issues”

   proof ↗⚑ flag

   cubic~partialclaimed5/10

   cubic.yaml provides per-repo config (source of truth for review behavior, ignore patterns, custom agents) and custom agents/RBAC roles suggest some centralized governance, but evidence shows configuration is per-repository (cubic.yaml lives in each repo root) rather than an org-level policy pushed/enforced across hundreds of repos with a central exceptions dashboard. missing for 10: an org-wide policy/template propagation mechanism across many repos, a central exceptions-management UI/workflow, and independent evidence of usage at hundreds-of-repos scale.

   • [claimed-docs] “`cubic.yaml` lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, …”
   • [claimed-docs] “cubic.yaml lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, an…”
   • [claimed-docs] “Custom agents are review rules that enforce your organization's specific best practices across pull requests.”
   • [claimed-docs] “cubic uses a role-based access control system to manage who can make changes to your team's subscription and settings. Every team member has…”
   • [claimed-docs] “Custom agents: Enforce your team's coding standards”

   proof ↗⚑ flag

Rules

1. engineering leadI encode my team's own review guidelines — natural-language rules, AST patterns, or linked style guides — and the reviewer enforces them

   weight 2 · round to Greptile
   Greptile✓fullclaimed8/10

   Greptile supports custom rules for team-specific issues, a greptile.json config file that overrides dashboard settings, indexing of existing rule files (Claude.md, AGENTS.md, cursor.rules), strictness/noise-control tuning, and a learning system that adapts from emoji reactions/comments — collectively covering natural-language rules, linked style-guide-like imports, and reinforcement learning of preferences. Missing for 10: explicit documentation of AST-pattern-based rule authoring and independent/community validation that custom rules are reliably enforced in practice.

   • [claimed-docs] “Create custom rules that catch team-specific issues”
   • [claimed-docs] “Place `greptile.json` in your repository root. Settings are read from the source branch of the PR and override dashboard settings.”
   • [claimed-docs] “Automatically index existing rule files like Claude.md, AGENTS.md, and cursor.rules for richer review context.”
   • [claimed-docs] “Control how strict Greptile is about leaving comments with the strictness setting (1–3).”
   • [claimed-docs] “With noise control, Greptile limits reviews to high-signal insights, skipping low impact or repetitive feedback.”
   • [claimed-docs] “Greptile learns from your team's feedback to provide increasingly relevant suggestions. The primary training methods are emoji reactions and…”
   • [claimed-docs] “Your 👍/👎 reactions and replies teach Greptile what matters. After 2-3 weeks, it stops commenting on things you don't care about.”

   proof ↗⚑ flag

   cubic~partialcommunity7/10

   cubic supports engineering-lead-authored review rules via 'Custom agents' (natural-language rules enforcing team standards), a cubic.yaml as source of truth, auto-detected context files (README, context.md, AGENTS.md, .cursorrules) and cross-repo linking for shared style guides, plus adoption of community rule sets. However, there is no explicit evidence of AST-pattern-based rule authoring, and one community comment expresses skepticism that plain-text/YAML rules are a 'half-measure' versus more robust structural checks, though this is opinion rather than a documented failure. Missing for 10: explicit AST-pattern rule support, independent hands-on validation that custom agents reliably enforce guidelines in practice.

   • [claimed-docs] “Custom agents are review rules that enforce your organization's specific best practices across pull requests.”
   • [claimed-docs] “`cubic.yaml` lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, …”
   • [claimed-docs] “Custom agents: Enforce your team's coding standards”
   • [claimed-docs] “You can also adopt existing agents from the community library with a single click.”
   • [claimed-docs] “cubic automatically detects and uses context files in your repository, including `README.md`, `context.md`, `AGENTS.md`, `.cursorrules`”
   • [claimed-docs] “cubic automatically detects and uses context files in your repository, including `README.md`, `context.md`, `AGENTS.md`, `.cursorrules`, age…”
   • [claimed-docs] “cubic.yaml lives in the root of your repository and becomes the source of truth for AI review behavior, ignore patterns, PR descriptions, an…”
   • [community] “The 'in-codebase' approach is the right one, but a YAML file with plain text is a half-measure. The most reliable rule that 'lives next to t…”

   proof ↗⚑ flag