Vercel Sandbox vs Maritime

Agent access

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

   weight 2 · round to Maritime
   Vercel Sandbox✓fullprobed8/10

   Probes confirm both a working llms.txt at the root domain (vercel.com/llms.txt, HTTP 200) and a markdown-formatted docs endpoint for Sandbox specifically (sandbox.md, HTTP 200), which are exactly the agent-oriented docs formats this story asks for. Missing for 10: no explicit evidence llms.txt itself deep-links to sandbox-specific docs, and no independent/community confirmation of an agent successfully consuming these files.

   • [probe] “PROBE llms.txt: HTTP 200 at https://vercel.com/llms.txt # Vercel > Vercel is a cloud platform for building, deploying, and scaling web appl…”
   • [probe] “PROBE docs-md: HTTP 200 at https://vercel.com/docs/sandbox.md --- title: Vercel Sandbox product: vercel url: /docs/sandbox canonical_url: "h…”
   • [claimed-docs] “Connect to and manage Sandboxes from the dashboard”

   proof ↗⚑ flag

   Maritime✓fullprobed9/10

   Maritime hosts a live llms.txt (confirmed by a 200 probe) plus a dedicated CLI llms.txt and an ai-agents docs page explicitly designed for agents (Claude Code, Cursor, etc.) to consume, authenticate, and act on — directly matching the story. Missing for 10: no independent third-party confirmation that agents actually navigate these docs successfully in practice.

   • [probe] “PROBE llms.txt: HTTP 200 at https://maritime.sh/llms.txt # Maritime > Maritime is a cloud platform for deploying, hosting, and managing AI …”
   • [claimed-docs] “Templates are cloneable agent packages; publishers can share them via link or the public library.”
   • [claimed-docs] “Pass --json on EVERY command. Success: one JSON value on stdout, stderr empty.”
   • [claimed-docs] “The Maritime CLI is built to be operated by other AI agents, not just humans: machine-readable output on every command, stable exit codes, a…”
   • [claimed-docs] “Paste one prompt into Claude, Cursor, or any assistant and it can install the CLI, authenticate, and deploy agents for you.”

   proof ↗⚑ flag

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

   weight 2 · round drawn
   Vercel Sandbox✓fullprobed8/10

   Vercel Sandbox is designed for headless/programmatic use via SDKs (@vercel/sandbox) and a CLI, with access tokens explicitly documented for external CI/CD systems and non-Vercel environments, plus OpenAPI-backed REST access confirmed by probe. missing for 10: no explicit third-party/independent CI pipeline case study or GitHub Actions example demonstrating real-world headless automation.

   • [claimed-docs] “Use `@vercel/sandbox` for JavaScript and TypeScript”
   • [claimed-docs] “Use access tokens when `VERCEL_OIDC_TOKEN` is unavailable, such as in external CI/CD systems or non-Vercel environments.”
   • [claimed-docs] “The Sandbox SDKs are the recommended way to integrate Vercel Sandbox into your applications. They provide a programmatic interface to create…”
   • [claimed-docs] “For local development, run `vercel link` and `vercel env pull` to get a development token. In production on Vercel, authentication is automa…”
   • [probe] “PROBE openapi: HTTP 200 at https://vercel.com/openapi.json — contains "openapi" key”

   proof ↗⚑ flag

   Maritime✓fullprobed8/10

   Maritime provides a CLI with --json flag on every command, machine-readable output, stable exit codes, and env-var (non-interactive) authentication via MARITIME_TOKEN, explicitly designed for automation/CI use and control by other AI agents rather than humans, plus a RESTful API and idempotent provisioning for scripted workflows. This directly supports headless/CI operation across CLI and SDK/API paths. Missing for 10: no explicit CI pipeline example (e.g., GitHub Actions/GitLab CI snippet) or independent third-party confirmation of CI usage.

   • [claimed-docs] “Pass --json on EVERY command. Success: one JSON value on stdout, stderr empty.”
   • [claimed-docs] “The Maritime CLI is built to be operated by other AI agents, not just humans: machine-readable output on every command, stable exit codes, a…”
   • [claimed-docs] “Authenticate with an env var, never interactively. Mint a long-lived key (`mk_…`) with `maritime keys create` and export it as `MARITIME_TOK…”
   • [claimed-docs] “Mint a long-lived key (`mk_…`) with `maritime keys create` and export it as `MARITIME_TOKEN`. Keys never expire; browser-login tokens do.”
   • [claimed-docs] “RESTful API for managing agents programmatically.”
   • [claimed-docs] “provision is idempotent on it: it returns the existing agent if there is one, else creates a new one. Safe to call on every sign-in.”
   • [probe] “official CLI documented at https://maritime.sh/docs/cli”

   proof ↗⚑ flag

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

   weight 3 · round drawn
   Vercel Sandbox—none0/10

   The axis applies to this product kind (peer products hold positive or none verdicts on this story), so lack of evidence for an applicable capability is "none", never "na". (na/none harmonized at arena bring-up — see pipeline/scripts/na-harmonize.ts.)

   ⚑ flag

   Maritime—none0/10

   Maritime is a platform for deploying and hosting agents built with frameworks (CrewAI, LangGraph, etc.), but the evidence pack shows no mention of MCP server integration or a mechanism for agents to consume external MCP tool servers. Nothing describes plugging in MCP servers to extend tool use.

   ⚑ flag

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

   weight 3 · round drawn
   Vercel Sandbox—none0/10

   Vercel Sandbox is an execution/infrastructure product (not itself an agent), so an official MCP server exposing it to agents is a fair, applicable axis; however no evidence in the pack mentions MCP, an MCP server, or any agent-protocol integration—only SDKs, CLI, and REST/OpenAPI docs are described.

   ⚑ flag

   Maritime—none0/10

   Maritime is a platform for deploying/hosting agents (not itself an agent), so an official MCP server is a fair axis to expect, but no evidence pack item mentions MCP, an MCP server, or MCP-compatible endpoints — only REST API, CLI, SDKs, and webhooks are documented.

   • [claimed-docs] “RESTful API for managing agents programmatically.”
   • [claimed-docs] “The Maritime CLI lets you deploy and manage agents from your terminal.”
   • [claimed-docs] “Mint a personal API key with `maritime keys create` and send it as a Bearer token, the same token the CLI itself uses.”

   proof ↗⚑ flag

5. ai-native userUse an official CLI

   weight 2 · round to Maritime
   Vercel Sandbox✓fullclaimed7/10

   Vercel Sandbox ships an official `sandbox` CLI explicitly positioned for manual testing, agentic workflows, debugging, and one-off operations, with documented commands like `sandbox create --name` and `sandbox run`. missing for 10: independent/hands-on corroboration beyond vendor docs, and no CLI reference/full command list shown.

   • [claimed-docs] “Use the `sandbox` CLI for manual testing, agentic workflows, debugging, and one-off operations”
   • [claimed-docs] “Create with an explicit name sandbox create --name my-sandbox”
   • [claimed-docs] “sandbox run echo "Hello Sandbox!"”
   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”

   proof ↗⚑ flag

   Maritime✓fullprobed9/10

   Maritime ships an official npm-installable CLI explicitly designed for AI-native use: machine-readable/--json output on every command, stable exit codes, a self-describing contract, and env-var (non-interactive) authentication via MARITIME_TOKEN, all documented at /docs/cli and /docs/ai-agents. It also supports agent-driven install/auth/deploy flows via prompted assistants. Missing for 10: independent/hands-on verification beyond vendor docs.

   • [claimed-docs] “The Maritime CLI lets you deploy and manage agents from your terminal.”
   • [claimed-docs] “npm install -g maritime-cli”
   • [claimed-docs] “The Maritime CLI is built to be operated by other AI agents, not just humans: machine-readable output on every command, stable exit codes, a…”
   • [claimed-docs] “Authenticate with an env var, never interactively. Mint a long-lived key (`mk_…`) with `maritime keys create` and export it as `MARITIME_TOK…”
   • [claimed-docs] “Pass --json on EVERY command. Success: one JSON value on stdout, stderr empty.”
   • [claimed-docs] “Paste one prompt into Claude, Cursor, or any assistant and it can install the CLI, authenticate, and deploy agents for you.”
   • [probe] “official CLI documented at https://maritime.sh/docs/cli”

   proof ↗⚑ flag

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

   weight 3 · round to Vercel Sandbox
   Vercel Sandbox✓fullprobed9/10

   Vercel Sandbox is explicitly built for agentic/AI use cases and exposes a documented SDK (@vercel/sandbox), CLI, and REST API (openapi.json) for programmatic control—creating sandboxes, running commands, streaming logs, managing files, images, and persistence. This directly satisfies driving the product via a documented public API for AI-native workflows. Missing for 10: independent third-party corroboration beyond vendor docs/probe checks.

   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”
   • [claimed-docs] “Use `@vercel/sandbox` for JavaScript and TypeScript”
   • [claimed-docs] “Use the `sandbox` CLI for manual testing, agentic workflows, debugging, and one-off operations”
   • [claimed-docs] “The Sandbox SDKs are the recommended way to integrate Vercel Sandbox into your applications. They provide a programmatic interface to create…”
   • [probe] “PROBE docs-md: HTTP 200 at https://vercel.com/docs/sandbox.md --- title: Vercel Sandbox product: vercel url: /docs/sandbox canonical_url: "h…”
   • [probe] “PROBE openapi: HTTP 200 at https://vercel.com/openapi.json — contains "openapi" key”

   proof ↗⚑ flag

   Maritime✓fullprobed8/10

   Maritime documents a full public API (REST endpoints, TypeScript/Python SDKs, idempotent provisioning, webhooks, scoped bearer-token auth) plus a CLI explicitly built for AI agents with machine-readable JSON output and stable exit codes, directly enabling AI-native driving of the platform. missing for 10: a discoverable machine-readable spec (openapi.json/swagger.json returned 404) and independent third-party corroboration of the API's stability.

   • [claimed-docs] “RESTful API for managing agents programmatically.”
   • [claimed-docs] “Mint a personal API key with `maritime keys create` and send it as a Bearer token, the same token the CLI itself uses.”
   • [claimed-docs] “Tag each agent with your own id via `externalId`. `provision` is **idempotent** on it: it returns the existing agent if there is one, else c…”
   • [claimed-docs] “provision is idempotent on it: it returns the existing agent if there is one, else creates a new one. Safe to call on every sign-in.”
   • [claimed-docs] “Get notified when a customer's agent changes state, instead of polling. Subscribe a URL, and Maritime POSTs a signed JSON body on every life…”
   • [claimed-docs] “The Maritime CLI is built to be operated by other AI agents, not just humans: machine-readable output on every command, stable exit codes, a…”
   • [claimed-docs] “Authenticate with an env var, never interactively. Mint a long-lived key (`mk_…`) with `maritime keys create` and export it as `MARITIME_TOK…”
   • [claimed-docs] “Pass --json on EVERY command. Success: one JSON value on stdout, stderr empty.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://maritime.sh/openapi.json, https://maritime.sh/swagger.json, https://maritime.sh/api/openapi.…”

   proof ↗⚑ flag

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

   weight 2 · round to Maritime
   Vercel Sandbox—none0/10

   Evidence only describes authentication mechanisms (automatic OIDC token in production, access tokens for CI/CD) but nowhere documents scoped or least-privilege permission models for these credentials — no mention of granular scopes, role restrictions, or per-agent credential limits.

   • [claimed-docs] “In production on Vercel, authentication is automatic.”
   • [claimed-docs] “Use access tokens when `VERCEL_OIDC_TOKEN` is unavailable, such as in external CI/CD systems or non-Vercel environments.”
   • [claimed-docs] “For local development, run `vercel link` and `vercel env pull` to get a development token. In production on Vercel, authentication is automa…”

   proof ↗⚑ flag

   Maritime✓fullclaimed8/10

   Maritime documents scoped API keys explicitly: every key carries scopes, users should hand subsystems the narrowest key they need, and a dedicated example shows a Computers-only key being refused on all other endpoints — a concrete least-privilege mechanism. This pairs with CLI/env-based key minting (mk_ tokens) suited to agent automation. Missing for 10: an exhaustive scope catalog/reference and independent (non-vendor) verification of scope enforcement in practice.

   • [claimed-docs] “Hand a subsystem the narrowest key it needs. A leaked key can then only do what its scope allows.”
   • [claimed-docs] “To hand out a key that reaches computers and nothing else, create a **Computers key** in the Computers section of Settings, API keys”
   • [claimed-docs] “Every key carries scopes. Hand a subsystem the _narrowest_ key it needs. A leaked key can then only do what its scope allows.”
   • [claimed-docs] “a key whose only scope is `computers` is refused on every other Maritime endpoint”
   • [claimed-docs] “Hand a subsystem the _narrowest_ key it needs. A leaked key can then only do what its scope allows.”
   • [claimed-docs] “Authenticate with an env var, never interactively. Mint a long-lived key (`mk_…`) with `maritime keys create` and export it as `MARITIME_TOK…”
   • [claimed-docs] “Mint a long-lived key (`mk_…`) with `maritime keys create` and export it as `MARITIME_TOKEN`. Keys never expire; browser-login tokens do.”

   proof ↗⚑ flag

8. ai-native userBuild against official SDKs

   weight 2 · round to Maritime
   Vercel Sandbox✓fullclaimed7/10

   Vercel Sandbox offers an official `@vercel/sandbox` SDK for JS/TS plus a CLI, with docs on authentication, commands, and sandbox lifecycle management — evidence that AI-native builders have a first-party SDK to integrate against. missing for 10: no mention of official SDKs for other languages (e.g., Python), no independent/hands-on corroboration of SDK usage beyond vendor docs.

   • [claimed-docs] “Use `@vercel/sandbox` for JavaScript and TypeScript”
   • [claimed-docs] “The Sandbox SDKs are the recommended way to integrate Vercel Sandbox into your applications. They provide a programmatic interface to create…”
   • [claimed-docs] “Use the `sandbox` CLI for manual testing, agentic workflows, debugging, and one-off operations”
   • [claimed-docs] “For local development, run `vercel link` and `vercel env pull` to get a development token. In production on Vercel, authentication is automa…”
   • [claimed-docs] “Commands can run in blocking mode (wait for completion) or detached mode (return immediately).”

   proof ↗⚑ flag

   Maritime✓fullprobed8/10

   Maritime documents official TypeScript and Python SDKs with idempotent agent provisioning, scoped authentication, webhooks, messaging, and a Files API, plus an npm-installable CLI and RESTful API, all clearly aimed at building AI-native integrations. missing for 10: no public OpenAPI/swagger spec found (404s) and no independent/hands-on corroboration beyond first-party docs.

   • [claimed-docs] “Tag each agent with your own id via `externalId`. `provision` is **idempotent** on it: it returns the existing agent if there is one, else c…”
   • [claimed-docs] “Mint a personal API key with `maritime keys create` and send it as a Bearer token, the same token the CLI itself uses.”
   • [claimed-docs] “Give every one of your users their own agent with the TypeScript or Python SDK.”
   • [claimed-docs] “Send a message and wait for the reply. Sleeping agents wake automatically.”
   • [claimed-docs] “RESTful API for managing agents programmatically.”
   • [claimed-docs] “The Maritime Files API: eight endpoints and both SDKs to browse, upload, download, and edit an agent's disk over HTTP, even while the agent …”
   • [claimed-docs] “provision is idempotent on it: it returns the existing agent if there is one, else creates a new one. Safe to call on every sign-in.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://maritime.sh/openapi.json, https://maritime.sh/swagger.json, https://maritime.sh/api/openapi.…”

   proof ↗⚑ flag

9. ai-native userSubscribe to events via webhooks

   weight 2 · round to Maritime
   Vercel Sandbox—none0/10

   Evidence covers sandbox creation, execution modes, persistence, and streaming logs from detached commands, but there is no mention of webhooks or event subscription mechanisms for external systems to receive Sandbox events.

   ⚑ flag

   Maritime✓fullclaimed7/10

   Docs explicitly describe subscribing a URL to receive signed JSON POSTs on every agent lifecycle event, replacing polling, which directly matches the story. missing for 10: independent/hands-on confirmation of webhook delivery reliability, payload schema details, and signature verification specifics.

   • [claimed-docs] “Get notified when a customer's agent changes state, instead of polling. Subscribe a URL, and Maritime POSTs a signed JSON body on every life…”
   • [claimed-docs] “Hand a subsystem the narrowest key it needs. A leaked key can then only do what its scope allows.”

   proof ↗⚑ flag

Agentic features

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

   weight 2 · round to Maritime
   Vercel Sandbox~partialclaimed5/10

   Vercel Sandbox supports running agent-generated code, detached/background commands with log streaming, timeouts/extension, and persistence/resume, which enable building autonomous background automations. However, there's no built-in scheduler, trigger system, or native 'automation' orchestration layer — users must build their own scheduling/trigger logic around the SDK/CLI. missing for 10: native scheduling/cron or event-trigger mechanism, independent evidence of long-running unattended automations in production, and explicit 'autonomous background job' framing beyond detached commands.

   • [claimed-docs] “Commands can run in blocking mode (wait for completion) or detached mode (return immediately).”
   • [claimed-docs] “Stream logs from a detached command”
   • [claimed-docs] “You can configure this using the `timeout` option when creating a sandbox, and extend it using `sandbox.extendTimeout()`.”
   • [claimed-docs] “Persistent sandboxes: Sandboxes that auto-save state on stop and resume where you left off. Persistence is the default. No manual snapshot m…”
   • [claimed-docs] “Sandboxes that auto-save state on stop and resume where you left off. Persistence is the default. No manual snapshot management needed.”
   • [claimed-docs] “Multi-agent isolation: Give each AI agent its own Linux user with a private home directory, and share files between agents with groups.”

   proof ↗⚑ flag

   Maritime✓fullclaimed8/10

   Maritime provides cron triggers with standard five-field expressions that wake agents on schedule and deliver prompts, plus agents can self-schedule their own wake triggers (e.g., 'send me a digest every weekday at 9'), and agents run persistently on always-on micro-VMs with sleep/wake economics, enabling true background autonomous automation. Missing for 10: independent/hands-on verification of trigger reliability and no mention of monitoring/alerting on automation failures.

   • [claimed-docs] “A cron trigger takes a standard five-field expression, wakes the agent on schedule, and delivers its `prompt` as a chat message on every fir…”
   • [claimed-docs] “Ask the agent (OpenClaw-family and Hermes templates): tell it "send me a digest every weekday at 9" and the platform syncs the schedule it s…”
   • [claimed-docs] “tell it "send me a digest every weekday at 9" and the platform syncs the schedule it sets for itself into real wake triggers”
   • [claimed-docs] “Always-on is a flat add-on for the rare agent that must react in milliseconds around the clock.”
   • [claimed-docs] “Send a message and wait for the reply. Sleeping agents wake automatically.”
   • [claimed-docs] “Maritime gives it what a laptop session can't: its own always-on micro-VM in the cloud, persistent storage, sleep/wake economics, and every …”

   proof ↗⚑ flag

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

   weight 2 · round to Maritime
   Vercel Sandbox—none0/10

   Vercel Sandbox is designed to run agent-generated code and is explicitly marketed for 'agent workflows,' but the evidence only describes CLI commands (`sandbox create`, `sandbox run`) and SDK calls, not a natural-language interface for operating the product itself. No documentation shows a chat/NL command layer for creating or managing sandboxes. Missing for 10: any NL command parser, chat-driven CLI, or natural-language API for controlling Sandbox operations.

   • [claimed-docs] “Use the `sandbox` CLI for manual testing, agentic workflows, debugging, and one-off operations”
   • [claimed-docs] “Create with an explicit name sandbox create --name my-sandbox”
   • [claimed-docs] “sandbox run echo "Hello Sandbox!"”

   proof ↗⚑ flag

   Maritime~partialclaimed6/10

   Maritime documents natural-language operation at the agent level (chat with agents, cron-triggered prompts set via natural language like 'send me a digest every weekday at 9') and CLI/API designed to be driven by AI agents via prompts (paste a prompt into Claude/Cursor to install and operate). However, the core Maritime platform itself (dashboard, agent management) is not shown to be operable purely via natural-language commands by the end user — the NL interface is mediated through the CLI/API being agent-friendly rather than a native NL command surface for platform operations. missing for 10: evidence of a native NL interface for platform-level operations (not just agent chat/prompts), independent/hands-on confirmation of this workflow succeeding.

   • [claimed-docs] “Ask the agent (OpenClaw-family and Hermes templates): tell it "send me a digest every weekday at 9" and the platform syncs the schedule it s…”
   • [claimed-docs] “Paste one prompt into Claude, Cursor, or any assistant and it can install the CLI, authenticate, and deploy agents for you.”
   • [claimed-docs] “tell it "send me a digest every weekday at 9" and the platform syncs the schedule it sets for itself into real wake triggers”
   • [claimed-docs] “The Maritime CLI is built to be operated by other AI agents, not just humans: machine-readable output on every command, stable exit codes, a…”
   • [claimed-docs] “point Claude Code or your Maritime agent at a provider's runbook page and it can inventory the old provider with read-only credentials, recr…”

   proof ↗⚑ flag

Api quality

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

   weight 2 · round drawn
   Vercel Sandbox—none0/10

   Evidence shows Vercel has an OpenAPI spec and Sandbox docs, but nothing describes an interactive API reference with runnable/live examples for Sandbox itself — no docs playground, no 'try it' console, no runnable code snippets in-page. missing for 10: interactive API reference UI, runnable example execution in docs, evidence of a docs playground feature.

   • [probe] “PROBE openapi: HTTP 200 at https://vercel.com/openapi.json — contains "openapi" key”
   • [claimed-docs] “The Sandbox SDKs are the recommended way to integrate Vercel Sandbox into your applications. They provide a programmatic interface to create…”

   proof ↗⚑ flag

   Maritime—none0/10

   Docs describe a RESTful API and CLI but there's no evidence of an interactive API reference (e.g., Swagger/OpenAPI explorer) with runnable examples; probes explicitly show no openapi.json/swagger.json spec found at any candidate path. missing for 10: interactive API console/playground, OpenAPI spec, runnable code examples in docs.

   • [claimed-docs] “RESTful API for managing agents programmatically.”
   • [claimed-docs] “Mint a personal API key with `maritime keys create` and send it as a Bearer token, the same token the CLI itself uses.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://maritime.sh/openapi.json, https://maritime.sh/swagger.json, https://maritime.sh/api/openapi.…”

   proof ↗⚑ flag

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

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox✓fullprobed9/10

   A probe confirms Vercel publishes a machine-readable OpenAPI spec at https://vercel.com/openapi.json which validly contains the 'openapi' key, and Sandbox is part of Vercel's documented API/SDK surface. Missing for 10: no explicit confirmation that Sandbox-specific endpoints are included in that spec versus general Vercel API.

   • [probe] “PROBE openapi: HTTP 200 at https://vercel.com/openapi.json — contains "openapi" key”
   • [claimed-docs] “The Sandbox SDKs are the recommended way to integrate Vercel Sandbox into your applications. They provide a programmatic interface to create…”

   proof ↗⚑ flag

   Maritime—none0/10

   Maritime documents a RESTful API and CLI with --json output, but there is no evidence of a downloadable machine-readable spec (OpenAPI/Swagger); active probes for openapi.json, swagger.json, and similar paths all returned 404.

   • [probe] “PROBE openapi: all candidate paths 404 (https://maritime.sh/openapi.json, https://maritime.sh/swagger.json, https://maritime.sh/api/openapi.…”
   • [claimed-docs] “RESTful API for managing agents programmatically.”
   • [claimed-docs] “Mint a personal API key with `maritime keys create` and send it as a Bearer token, the same token the CLI itself uses.”

   proof ↗⚑ flag

3. ai-native userTest against a sandbox environment without touching production data

   weight 1 · round to Vercel Sandbox
   Vercel Sandbox✓fullclaimed8/10

   Vercel Sandbox is explicitly designed to run untrusted/agent-generated code in isolated Linux microVMs 'without exposing your production systems,' with per-agent isolation, ephemeral or persistent state, and no shared access to production infrastructure. This directly matches the AI-native story of testing agent code safely away from production data. missing for 10: independent/hands-on evidence of real-world isolation guarantees, and explicit documentation on data-boundary controls preventing production data leakage into sandboxes.

   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”
   • [claimed-docs] “Execute untrusted code safely: Run AI agent output, user uploads, or third-party scripts without exposing your production systems.”
   • [claimed-docs] “Multi-agent isolation: Give each AI agent its own Linux user with a private home directory, and share files between agents with groups.”
   • [claimed-docs] “Test in isolation: Preview how user-submitted or agent-generated code behaves in a self-contained environment with access to logs, file edit…”
   • [claimed-docs] “Test in isolation: Preview how user-submitted or agent-generated code behaves in a self”

   proof ↗⚑ flag

   Maritime—none0/10

   Maritime's docs cover deployment, keys/scopes, templates, triggers, and agent lifecycle, but nothing describes a sandbox/staging environment isolated from production data for testing purposes. Missing for 10: any mention of a sandbox/test mode, staging environment, or data isolation guarantees between test and production agents.

   ⚑ flag

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

   weight 2 · round drawn
   Vercel Sandbox—none0/10

   The evidence pack shows SDK/CLI docs, features, and an OpenAPI spec, but no mention of API versioning scheme or a documented deprecation policy for Vercel Sandbox's APIs. Some features are explicitly marked 'beta' (docs-14, docs-44), suggesting instability without a stated deprecation process. missing for 10: explicit API version numbers, a deprecation/sunset policy, changelog entries documenting breaking-change handling.

   ⚑ flag

   Maritime—none0/10

   No evidence of API versioning scheme (e.g., v1/v2 paths) or any documented deprecation policy; OpenAPI spec probes all 404, and no docs mention version support windows or breaking-change policy.

   • [probe] “PROBE openapi: all candidate paths 404 (https://maritime.sh/openapi.json, https://maritime.sh/swagger.json, https://maritime.sh/api/openapi.…”
   • [claimed-docs] “RESTful API for managing agents programmatically.”

   proof ↗⚑ flag

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

   weight 2 · round to Maritime
   Vercel Sandbox—none0/10

   Vercel Sandbox docs describe creating/running individual sandboxes, multi-agent isolation, and SDK-driven command execution, but there is no evidence of a bulk/batch operation feature for processing many items at once (e.g., batch job APIs, fan-out orchestration, or bulk file/task processing).

   ⚑ flag

   Maritime~partialclaimed4/10

   There is explicit support for bulk import of config/env vars (maritime-docs-11) and a scriptable CLI/API with JSON output (maritime-docs-9, maritime-docs-36) that an AI-native user could loop over to perform bulk actions, but there is no documented batch/bulk endpoint for operating on many agents or resources at once. missing for 10: a dedicated bulk-operation API/CLI command (e.g. bulk create/delete/update agents), evidence of scaling to many items reliably, and independent confirmation of bulk workflows in practice.

   • [claimed-docs] “Bulk import from a .env file (or stdin), then hot-reload”
   • [claimed-docs] “Pass --json on EVERY command. Success: one JSON value on stdout, stderr empty.”
   • [claimed-docs] “RESTful API for managing agents programmatically.”
   • [claimed-docs] “The Maritime CLI is built to be operated by other AI agents, not just humans: machine-readable output on every command, stable exit codes, a…”

   proof ↗⚑ flag

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

   weight 3 · round to Maritime
   Vercel Sandbox—none0/10

   The axis applies to this product kind (peer products hold positive or none verdicts on this story), so lack of evidence for an applicable capability is "none", never "na". (na/none harmonized at arena bring-up — see pipeline/scripts/na-harmonize.ts.)

   ⚑ flag

   Maritime~partialclaimed6/10

   Maritime supports triggers (cron with five-field expressions, natural-language schedule setting, webhooks for lifecycle events) that wake agents and fire actions automatically, which covers rule-based automation on events. However, this is event/schedule-based triggering rather than a general-purpose conditional rules engine (e.g., no evidence of arbitrary if/then logic across multiple event types or complex conditions). missing for 10: evidence of a general conditional rules/policy engine beyond cron and webhook triggers, and independent/hands-on confirmation of trigger reliability.

   • [claimed-docs] “A cron trigger takes a standard five-field expression, wakes the agent on schedule, and delivers its `prompt` as a chat message on every fir…”
   • [claimed-docs] “Ask the agent (OpenClaw-family and Hermes templates): tell it "send me a digest every weekday at 9" and the platform syncs the schedule it s…”
   • [claimed-docs] “tell it "send me a digest every weekday at 9" and the platform syncs the schedule it sets for itself into real wake triggers”
   • [claimed-docs] “Get notified when a customer's agent changes state, instead of polling. Subscribe a URL, and Maritime POSTs a signed JSON body on every life…”
   • [claimed-docs] “Always-on is a flat add-on for the rare agent that must react in milliseconds around the clock.”

   proof ↗⚑ flag

3. ai-native userSchedule recurring jobs or workflows

   weight 2 · round to Maritime
   Vercel Sandbox—none0/10

   Vercel Sandbox provides on-demand ephemeral/persistent execution environments with manual creation, timeouts, and resumption, but no evidence of a scheduler, cron-like trigger, or recurring job orchestration feature; scheduling would need to be built externally (e.g., Vercel Cron) which isn't documented here as part of Sandbox itself.

   • [claimed-docs] “You can configure this using the `timeout` option when creating a sandbox, and extend it using `sandbox.extendTimeout()`.”
   • [claimed-docs] “The default timeout is 5 minutes. You can configure this using the `timeout` option when creating a sandbox, and extend it using `sandbox.ex…”
   • [claimed-docs] “Or retrieve an existing sandbox by name (resumes if stopped)”

   proof ↗⚑ flag

   Maritime✓fullclaimed8/10

   Maritime explicitly supports cron triggers with standard five-field expressions that wake agents and deliver prompts on schedule, plus a natural-language flow where an agent sets its own recurring schedule ('send me a digest every weekday at 9') that syncs into real wake triggers. This directly covers recurring job/workflow scheduling for AI-native users. Missing for 10: independent/hands-on verification beyond vendor docs, and no detail on managing/editing/cancelling existing scheduled triggers via CLI/API.

   • [claimed-docs] “A cron trigger takes a standard five-field expression, wakes the agent on schedule, and delivers its `prompt` as a chat message on every fir…”
   • [claimed-docs] “Ask the agent (OpenClaw-family and Hermes templates): tell it "send me a digest every weekday at 9" and the platform syncs the schedule it s…”
   • [claimed-docs] “tell it "send me a digest every weekday at 9" and the platform syncs the schedule it sets for itself into real wake triggers”
   • [claimed-docs] “Always-on is a flat add-on for the rare agent that must react in milliseconds around the clock.”

   proof ↗⚑ flag

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

   weight 1 · round drawn
   Vercel Sandbox—none0/10

   The axis applies to this product kind (peer products hold positive or none verdicts on this story), so lack of evidence for an applicable capability is "none", never "na". (na/none harmonized at arena bring-up — see pipeline/scripts/na-harmonize.ts.)

   ⚑ flag

   Maritime—none0/10

   No evidence of version history, review workflows, or rollback for automations/agent configs; docs cover deployment, triggers, builds, and CLI but nothing about versioning or reverting agent configuration/automation changes.

   ⚑ flag

Agent workloads

1. ai-native userRun coding agents like Claude Code or Codex inside the sandbox following the vendor's own recipe

   weight 2 · round to Maritime
   Vercel Sandbox~partialclaimed6/10

   Docs confirm the default 'vercel/sandbox/universal' image ships with coding agents preinstalled and highlight agent workflows (e.g. Claude Code/Codex-style use), but the evidence pack lacks an explicit vendor 'recipe' walkthrough naming Claude Code or Codex step-by-step. missing for 10: a dedicated vendor tutorial/recipe explicitly running Claude Code or Codex inside the sandbox, and independent hands-on confirmation of this exact workflow.

   • [claimed-docs] “Sandboxes run Linux images, with Ubuntu, Arch Linux, or any other Linux distribution you need. The default image is `vercel/sandbox/universa…”
   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”
   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs.”
   • [claimed-docs] “Multi-agent isolation: Give each AI agent its own Linux user with a private home directory, and share files between agents with groups.”
   • [claimed-docs] “Use the `sandbox` CLI for manual testing, agentic workflows, debugging, and one-off operations”

   proof ↗⚑ flag

   Maritime✓fullclaimed8/10

   Maritime documents dedicated template recipes for running Claude Code and Codex inside its sandboxed micro-VMs (persistent storage, sleep/wake, metered LLM access by default, chat channels), directly matching the vendor's own recipe framing for these named coding agents. missing for 10: independent/hands-on verification beyond vendor docs, and no direct evidence of a step-by-step 'follow this exact recipe' walkthrough output for a live run.

   • [claimed-docs] “Maritime gives it what a laptop session can't: its own always-on micro-VM in the cloud, persistent storage, sleep/wake economics, and every …”
   • [claimed-docs] “No key is required: the agent uses Maritime's metered LLM access by default.”
   • [claimed-docs] “Every ready-made template is a real agent framework, pre-configured and one image pull from live. Pick one, name it, deploy.”
   • [claimed-docs] “point Claude Code or your Maritime agent at a provider's runbook page and it can inventory the old provider with read-only credentials, recr…”
   • [claimed-docs] “Paste one prompt into Claude, Cursor, or any assistant and it can install the CLI, authenticate, and deploy agents for you.”

   proof ↗⚑ flag

Workloads

1. developerRun a headless browser or full desktop environment inside the sandbox

   weight 2 · round to Maritime
   Vercel Sandbox—none0/10

   Evidence describes Linux microVMs, custom OCI images, and general code execution, but nothing mentions headless browser support, display/VNC, GUI desktop environments, or browser automation tooling. missing for 10: any mention of headless browser (e.g., Chromium/Playwright) support, GPU/display capabilities, or desktop environment provisioning.

   ⚑ flag

   Maritime✓fullclaimed7/10

   Maritime docs explicitly describe 'Computers': a persistent Linux desktop (XFCE, Chromium, LibreOffice) per user driven by a model, and the changelog confirms agents get a real Linux desktop with screenshot/mouse/keyboard control and live takeover — this covers full desktop environments including a browser (Chromium) inside the sandbox. missing for 10: no independent/hands-on corroboration of performance or headless-only browser mode, and no detail on resource limits or concurrency.

   • [claimed-docs] “A persistent Linux desktop (XFCE, Chromium, LibreOffice) for each of _your_ end users, driven by _your_ model.”
   • [claimed-docs] “Agents are getting a real Linux desktop they can drive: screenshot, mouse, keyboard, and a live view you can take over.”
   • [claimed-docs] “To hand out a key that reaches computers and nothing else, create a **Computers key** in the Computers section of Settings, API keys”

   proof ↗⚑ flag

2. developerAttach GPUs to sandboxed workloads

   weight 2 · round drawn
   Vercel Sandbox—none0/10

   No evidence anywhere in the docs mentions GPU support or GPU attachment for Vercel Sandbox; sandboxes are described only as CPU-based Linux microVMs with system-privileged process support, storage/drives, and networking. missing for 10: any mention of GPU hardware, GPU-enabled images, or GPU pricing/quota.

   • [claimed-docs] “Sandboxes run Linux images, with Ubuntu, Arch Linux, or any other Linux distribution you need.”
   • [claimed-docs] “System-privileged processes: Run workloads that need system-level privileges, such as container runtimes like Docker, VPN clients, and FUSE …”
   • [claimed-docs] “Vercel Sandbox usage is metered across several dimensions. This page explains how billing works for each plan, what quotas apply, and how to…”

   proof ↗⚑ flag

   Maritime—none0/10

   No evidence anywhere in the pack mentions GPUs, GPU-accelerated instances, or hardware acceleration for sandboxed agent workloads; all compute descriptions reference micro-VMs, sleep/wake, and CPU-based containers only. Missing for 10: any mention of GPU attachment, GPU instance types, or hardware accelerator support.

   • [claimed-docs] “Maritime gives it what a laptop session can't: its own always-on micro-VM in the cloud, persistent storage, sleep/wake economics, and every …”
   • [claimed-docs] “Any Docker container becomes an agent: three endpoints get you sleep/wake, channels, and triggers.”

   proof ↗⚑ flag

Files

1. developerRead, write, upload, and download files in the sandbox filesystem via the SDK

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox~partialclaimed6/10

   Docs confirm the SDK provides a programmatic interface to manage files (mount remote storage, drives, filesystem persistence via snapshots) and general file access is implied through 'file edits' and FUSE-mounted storage, but no explicit documentation of discrete SDK methods like sandbox.writeFiles(), readFile(), or upload/download APIs is present. missing for 10: explicit SDK method documentation/examples for read, write, upload, and download file operations, and independent/hands-on confirmation of these specific API calls.

   • [claimed-docs] “The Sandbox SDKs are the recommended way to integrate Vercel Sandbox into your applications. They provide a programmatic interface to create…”
   • [claimed-docs] “Mount an external object store such as Amazon S3 with a FUSE driver, and read and write remote files through the sandbox filesystem.”
   • [claimed-docs] “Drives (beta): Attach persistent filesystem storage to sandboxes and reuse data across sandbox runs.”
   • [claimed-docs] “Test in isolation: Preview how user-submitted or agent-generated code behaves in a self-contained environment with access to logs, file edit…”
   • [claimed-docs] “when a sandbox stops, the SDK automatically snapshots its filesystem, and the sandbox configuration is preserved across sessions, so both ar…”

   proof ↗⚑ flag

   Maritime—none0/10

   No evidence pack item mentions SDK-based file read/write/upload/download operations against a sandbox filesystem; the docs cover agent provisioning, webhooks, auth scoping, triggers, computers (desktop), CLI, and migration but nothing about filesystem file I/O via SDK. Missing for full: no filesystem API/SDK method docs, no upload/download endpoint references, no code samples showing file operations.

   ⚑ flag

Runtimes

1. developerDefine custom sandbox templates or bring my own container image

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox✓fullclaimed9/10

   Docs explicitly support custom sandbox templates and BYO container images: users can build custom OCI images stored in Vercel Container Registry, start sandboxes from managed, shared/public, or custom images, and Docker-image-based workflows are directly supported. Also supports snapshotting for reusable custom environments. missing for 10: independent/hands-on corroboration of custom image builds beyond first-party docs.

   • [claimed-docs] “Build your system packages and tooling into an image, and share it across projects and teams.”
   • [claimed-docs] “If you already use Docker images to define your environment, store the image in Vercel Container Registry (VCR) and create the sandbox with …”
   • [claimed-docs] “Start sandboxes from Vercel's Managed Images, a shared or public image, or your own OCI images stored in Vercel Container Registry.”
   • [claimed-docs] “Managed and custom images: Start sandboxes from Vercel's Managed Images, a shared or public image, or your own OCI images stored in Vercel C…”
   • [claimed-docs] “Sandboxes run Linux images, with Ubuntu, Arch Linux, or any other Linux distribution you need. The default image is `vercel/sandbox/universa…”

   proof ↗⚑ flag

   Maritime~partialclaimed6/10

   Docs confirm bring-your-own-image support via a documented three-endpoint contract that works with any Docker image (CrewAI, LangGraph, AutoGen, custom scripts), plus a build log and template system for packaging custom agents. However, there's no detailed spec of the contract itself, no example custom Dockerfile, and no independent/hands-on confirmation of successfully running a fully custom image end-to-end. missing for 10: full technical spec of the three-endpoint contract, a worked example of a custom image deployment, independent verification of custom image support beyond docs claims.

   • [claimed-docs] “CrewAI, LangGraph, AutoGen, your own Python script: any Docker image that speaks a three-endpoint contract gets the whole platform.”
   • [claimed-docs] “Templates are cloneable agent packages; publishers can share them via link or the public library.”
   • [claimed-docs] “The agent page streams the build log live under **Overview → Builds**.”
   • [claimed-docs] “Every ready-made template is a real agent framework, pre-configured and one image pull from live. Pick one, name it, deploy.”

   proof ↗⚑ flag

2. developerExecute code in multiple language runtimes (Python, JavaScript, and more) and get rich results back

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox✓fullclaimed7/10

   Docs confirm sandboxes run Linux images with Python 3.14 and Node.js LTS pre-installed by default, plus support for any other Linux distro via custom OCI images, enabling multi-language execution (JS/TS, Python, and more) (docs-5, docs-8, docs-34, docs-30). Execution results are surfaced via logs, streamed output, and file access (docs-19, docs-20, docs-26), but there's no explicit mention of structured/rich result formats (e.g., images, dataframes, plots) beyond logs and files. Missing for 10: explicit support for rich/structured output types (charts, tables, media) and independent hands-on confirmation of multi-language fidelity.

   • [claimed-docs] “Use `@vercel/sandbox` for JavaScript and TypeScript”
   • [claimed-docs] “Sandboxes run Linux images, with Ubuntu, Arch Linux, or any other Linux distribution you need.”
   • [claimed-docs] “Sandboxes run Linux images, with Ubuntu, Arch Linux, or any other Linux distribution you need. The default image is `vercel/sandbox/universa…”
   • [claimed-docs] “Start sandboxes from Vercel's Managed Images, a shared or public image, or your own OCI images stored in Vercel Container Registry.”
   • [claimed-docs] “Commands can run in blocking mode (wait for completion) or detached mode (return immediately).”
   • [claimed-docs] “Stream logs from a detached command”
   • [claimed-docs] “Test in isolation: Preview how user-submitted or agent-generated code behaves in a self-contained environment with access to logs, file edit…”

   proof ↗⚑ flag

   Maritime—none0/10

   Maritime's evidence describes deploying persistent Docker-based agents (any Docker image with sleep/wake/channels/triggers), a Files API, and a Linux desktop 'Computers' feature, but nothing describes a code-execution/sandbox API that runs snippets in multiple language runtimes (Python, JS, etc.) and returns structured/rich results (e.g., stdout, images, dataframes) the way a code-interpreter tool would.

   • [claimed-docs] “Any Docker container becomes an agent: three endpoints get you sleep/wake, channels, and triggers.”
   • [claimed-docs] “A persistent Linux desktop (XFCE, Chromium, LibreOffice) for each of _your_ end users, driven by _your_ model.”
   • [claimed-docs] “The Maritime Files API: eight endpoints and both SDKs to browse, upload, download, and edit an agent's disk over HTTP, even while the agent …”
   • [claimed-docs] “CrewAI, LangGraph, AutoGen, your own Python script: any Docker image that speaks a three-endpoint contract gets the whole platform.”

   proof ↗⚑ flag

Untrusted code

1. developerExecute untrusted, AI-generated code without risking my own infrastructure

   weight 3 · round to Vercel Sandbox
   Vercel Sandbox✓fullclaimed9/10

   Vercel Sandbox's docs directly describe running untrusted or AI agent-generated code in isolated Linux microVMs without exposing production systems, with per-agent isolation, configurable timeouts, and managed images—precisely matching the story. Missing for 10: independent/hands-on verification beyond first-party docs.

   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”
   • [claimed-docs] “Execute untrusted code safely: Run AI agent output, user uploads, or third-party scripts without exposing your production systems.”
   • [claimed-docs] “Multi-agent isolation: Give each AI agent its own Linux user with a private home directory, and share files between agents with groups.”
   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs.”
   • [claimed-docs] “Test in isolation: Preview how user-submitted or agent-generated code behaves in a self-contained environment with access to logs, file edit…”

   proof ↗⚑ flag

   Maritime~partialprobed5/10

   Maritime hosts agents as Docker images on 'serverless infrastructure' (maritime-docs-1, maritime-probe-1) and provides persistent isolated Linux desktops per end user (maritime-docs-5), which implies agent code runs off the developer's own machines/infra. However, there is no explicit documentation of sandboxing guarantees, resource/network isolation, or security boundaries specifically for running untrusted or AI-generated code safely. Missing for 10: explicit sandbox/isolation security model, resource-limit guarantees, and any independent verification that untrusted code cannot escape or affect other tenants/infra.

   • [claimed-docs] “CrewAI, LangGraph, AutoGen, your own Python script: any Docker image that speaks a three-endpoint contract gets the whole platform.”
   • [claimed-docs] “A persistent Linux desktop (XFCE, Chromium, LibreOffice) for each of _your_ end users, driven by _your_ model.”
   • [probe] “PROBE llms.txt: HTTP 200 at https://maritime.sh/llms.txt # Maritime > Maritime is a cloud platform for deploying, hosting, and managing AI …”

   proof ↗⚑ flag

2. developerRun arbitrary shell commands and install packages inside the sandbox

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox✓fullclaimed8/10

   Docs clearly show running arbitrary commands (blocking/detached modes, `sandbox run echo ...`, CLI and SDK) inside Linux microVMs, with custom images and full Linux distros allowing package installs (e.g., Ubuntu/Arch, custom OCI images, system-privileged processes). missing for 10: no independent/hands-on report confirming package installation workflows beyond first-party docs.

   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”
   • [claimed-docs] “Sandboxes run Linux images, with Ubuntu, Arch Linux, or any other Linux distribution you need.”
   • [claimed-docs] “System-privileged processes: Run workloads that need system-level privileges, such as container runtimes like Docker, VPN clients, and FUSE …”
   • [claimed-docs] “Commands can run in blocking mode (wait for completion) or detached mode (return immediately).”
   • [claimed-docs] “Use the `sandbox` CLI for manual testing, agentic workflows, debugging, and one-off operations”
   • [claimed-docs] “Sandboxes run Linux images, with Ubuntu, Arch Linux, or any other Linux distribution you need. The default image is `vercel/sandbox/universa…”
   • [claimed-docs] “sandbox run echo "Hello Sandbox!"”
   • [claimed-docs] “Start sandboxes from Vercel's Managed Images, a shared or public image, or your own OCI images stored in Vercel Container Registry.”

   proof ↗⚑ flag

   Maritime—none0/10

   Maritime's docs describe deploying agents via Docker images, templates, and a custom framework contract, but there is no evidence of an interactive shell/terminal access or arbitrary package installation inside a running sandbox for a developer to use directly. The 'Computers' feature offers a persistent desktop for end-users, not a shell for arbitrary command execution by the developer.

   ⚑ flag

Agent lifecycle

1. ai-native userMy agent can provision its own sandbox, execute code, read the results, and tear it down — end to end without a human

   weight 3 · round to Vercel Sandbox
   Vercel Sandbox✓fullclaimed8/10

   Vercel Sandbox provides a full programmatic SDK/CLI to create sandboxes (with names, custom images, regions), run blocking or detached commands, stream logs/output, and stop/tear down sandboxes, with API-based authentication (OIDC/tokens) suitable for unattended agent use — no human-in-the-loop required. missing for 10: no independent/hands-on third-party report confirming an agent autonomously completing the full provision-execute-teardown loop, and no explicit example showing an agent reading command output/results programmatically end-to-end.

   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”
   • [claimed-docs] “Execute untrusted code safely: Run AI agent output, user uploads, or third-party scripts without exposing your production systems.”
   • [claimed-docs] “Commands can run in blocking mode (wait for completion) or detached mode (return immediately).”
   • [claimed-docs] “Stream logs from a detached command”
   • [claimed-docs] “You can also stop sandboxes from the Vercel Dashboard by navigating to Observability > Sandboxes and clicking Stop Sandbox.”
   • [claimed-docs] “The Sandbox SDKs are the recommended way to integrate Vercel Sandbox into your applications. They provide a programmatic interface to create…”
   • [claimed-docs] “For local development, run `vercel link` and `vercel env pull` to get a development token. In production on Vercel, authentication is automa…”
   • [claimed-docs] “Create with an explicit name sandbox create --name my-sandbox”
   • [claimed-docs] “sandbox run echo "Hello Sandbox!"”

   proof ↗⚑ flag

   Maritime~partialclaimed6/10

   Maritime's CLI and SDK let an agent provision (idempotent) itself, run in a sandboxed micro-VM/container, and be managed via authenticated API/CLI calls with JSON output designed for machine consumption, and agents can be torn down/put to sleep automatically when idle. However, the evidence is docs-only with no hands-on trace of an agent actually reading back code-execution results or explicit 'delete/destroy' teardown call, and no independent verification of the full autonomous loop. missing for 10: concrete evidence of an agent reading execution output programmatically, an explicit teardown/delete API example, and independent/hands-on confirmation of the full provision→execute→read→teardown loop.

   • [claimed-docs] “Tag each agent with your own id via `externalId`. `provision` is **idempotent** on it: it returns the existing agent if there is one, else c…”
   • [claimed-docs] “Pass --json on EVERY command. Success: one JSON value on stdout, stderr empty.”
   • [claimed-docs] “The Maritime CLI is built to be operated by other AI agents, not just humans: machine-readable output on every command, stable exit codes, a…”
   • [claimed-docs] “Authenticate with an env var, never interactively. Mint a long-lived key (`mk_…`) with `maritime keys create` and export it as `MARITIME_TOK…”
   • [claimed-docs] “Send a message and wait for the reply. Sleeping agents wake automatically.”
   • [claimed-docs] “provision is idempotent on it: it returns the existing agent if there is one, else creates a new one. Safe to call on every sign-in.”
   • [claimed-docs] “Any Docker container becomes an agent: three endpoints get you sleep/wake, channels, and triggers.”

   proof ↗⚑ flag

Isolation

1. platform-engineerRely on a documented hard isolation boundary (microVM or equivalent) between sandboxes and my systems

   weight 3 · round to Vercel Sandbox
   Vercel Sandbox~partialclaimed6/10

   Vercel's first-party docs repeatedly and explicitly claim microVM-based isolation for running untrusted/agent-generated code without exposing production systems, plus additional isolation features (per-agent Linux users, system-privileged process support). However, all evidence is vendor documentation with no independent security audit, third-party penetration test, or technical deep-dive into the underlying microVM technology (e.g. Firecracker) confirming the boundary in practice. missing for 10: independent/third-party verification of the isolation boundary, technical detail on the hypervisor/microVM implementation, and any hands-on security assessment.

   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”
   • [claimed-docs] “Execute untrusted code safely: Run AI agent output, user uploads, or third-party scripts without exposing your production systems.”
   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs.”
   • [claimed-docs] “System-privileged processes: Run workloads that need system-level privileges, such as container runtimes like Docker, VPN clients, and FUSE …”
   • [claimed-docs] “Multi-agent isolation: Give each AI agent its own Linux user with a private home directory, and share files between agents with groups.”

   proof ↗⚑ flag

   Maritime—none0/10

   No evidence pack item documents a specific isolation mechanism (microVM, gVisor, container hardening, VM boundary) between sandboxes and host systems; docs mention Docker images, serverless infra, and scoped API keys but not the underlying execution isolation boundary. Missing for 10: any explicit statement of sandbox isolation technology, security model documentation, or third-party audit/confirmation of hard isolation.

   ⚑ flag

2. ai-native userGive an agent a sandbox where host secrets and credentials are unreachable by the code it runs

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox✓fullclaimed8/10

   Vercel Sandbox docs explicitly describe running agent-generated/untrusted code in isolated Linux microVMs 'without exposing your production systems,' and support per-agent Linux user isolation, which directly implies host secrets/credentials are not reachable by sandboxed code. missing for 10: explicit documentation of how host secrets/env vars are excluded or scoped from the sandbox environment, and independent/hands-on verification of this isolation guarantee.

   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”
   • [claimed-docs] “Execute untrusted code safely: Run AI agent output, user uploads, or third-party scripts without exposing your production systems.”
   • [claimed-docs] “Multi-agent isolation: Give each AI agent its own Linux user with a private home directory, and share files between agents with groups.”
   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs.”

   proof ↗⚑ flag

   Maritime~partialclaimed6/10

   Maritime's docs describe per-agent micro-VM isolation with secrets 'injected only into that agent's VM' and encrypted at rest, plus scoped API keys so a leaked key 'can only do what its scope allows,' which together imply host-level credentials aren't reachable by agent code. However, there's no explicit documentation of sandbox mechanics like blocking cloud metadata endpoints, filesystem isolation from the host, or independent security testing confirming secrets are truly unreachable. missing for 10: explicit host-credential isolation guarantees (e.g., no metadata service access, filesystem boundary details), independent/hands-on security verification.

   • [claimed-docs] “Variables are **secret by default**: stored encrypted at rest, injected only into that agent's VM, and always masked in the UI.”
   • [claimed-docs] “Hand a subsystem the narrowest key it needs. A leaked key can then only do what its scope allows.”
   • [claimed-docs] “To hand out a key that reaches computers and nothing else, create a **Computers key** in the Computers section of Settings, API keys”
   • [claimed-docs] “a key whose only scope is `computers` is refused on every other Maritime endpoint”
   • [claimed-docs] “Every key carries scopes. Hand a subsystem the _narrowest_ key it needs. A leaked key can then only do what its scope allows.”

   proof ↗⚑ flag

Network policy

1. platform-engineerRestrict or allow the sandbox's network egress with explicit policy

   weight 3 · round drawn
   Vercel Sandbox—none0/10

   No evidence in the pack mentions network egress controls, firewall rules, or outbound traffic policy for Vercel Sandbox—coverage focuses on isolation via microVMs, filesystem, images, and multi-agent user separation but never explicit egress allow/deny configuration.

   ⚑ flag

   Maritime—none0/10

   No evidence pack item mentions network egress controls, firewall rules, or any explicit allow/deny policy for outbound traffic from the sandbox/agent VM; the closest material covers API key scoping and secret management, not network policy.

   ⚑ flag

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

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox✓fullprobed7/10

   Vercel Sandbox is fundamentally API/SDK-driven (creation, commands, snapshots, tags, drives, custom images all documented via SDK/CLI), and the dashboard only offers a subset (viewing/stopping sandboxes) that mirrors API capability rather than adding unique UI-only features, and an OpenAPI spec exists confirming programmatic access. Missing for 10: an explicit vendor statement confirming full UI/API parity, and independent confirmation that every dashboard action (e.g., tag/drive management) is scriptable via API with no UI-exclusive features.

   • [claimed-docs] “The Sandbox SDKs are the recommended way to integrate Vercel Sandbox into your applications. They provide a programmatic interface to create…”
   • [claimed-docs] “Use the `sandbox` CLI for manual testing, agentic workflows, debugging, and one-off operations”
   • [claimed-docs] “You can also stop sandboxes from the Vercel Dashboard by navigating to Observability > Sandboxes and clicking Stop Sandbox.”
   • [claimed-docs] “Connect to and manage Sandboxes from the dashboard”
   • [probe] “PROBE openapi: HTTP 200 at https://vercel.com/openapi.json — contains "openapi" key”

   proof ↗⚑ flag

   Maritime~partialprobed6/10

   Docs show a broad API/SDK/CLI surface (provisioning, keys, webhooks, triggers, computers, templates, builds) and CLI --json mode implying the CLI itself is just an API wrapper, suggesting an API-first architecture. However there's no explicit statement of full UI/API parity and no discoverable OpenAPI spec (probes return 404s for all schema locations), so completeness of API coverage relative to every UI feature can't be confirmed. missing for 10: explicit parity statement, published OpenAPI/API reference, independent confirmation that every UI action has an API equivalent.

   • [claimed-docs] “Mint a personal API key with `maritime keys create` and send it as a Bearer token, the same token the CLI itself uses.”
   • [claimed-docs] “Pass --json on EVERY command. Success: one JSON value on stdout, stderr empty.”
   • [claimed-docs] “Tag each agent with your own id via `externalId`. `provision` is **idempotent** on it: it returns the existing agent if there is one, else c…”
   • [claimed-docs] “Hand a subsystem the narrowest key it needs. A leaked key can then only do what its scope allows.”
   • [claimed-docs] “A persistent Linux desktop (XFCE, Chromium, LibreOffice) for each of _your_ end users, driven by _your_ model.”
   • [claimed-docs] “The agent page streams the build log live under **Overview → Builds**.”
   • [claimed-docs] “Templates are cloneable agent packages; publishers can share them via link or the public library.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://maritime.sh/openapi.json, https://maritime.sh/swagger.json, https://maritime.sh/api/openapi.…”

   proof ↗⚑ flag

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

   weight 3 · round drawn
   Vercel Sandbox—none0/10

   Vercel Sandbox stores state via drives, snapshots, and container images, so data portability is a fair question, but no evidence describes an explicit data-export feature or guarantee of open, exportable formats for that stored state. Missing for 10: any documented export/download-all-data capability, explicit open-format guarantees for drives/snapshots, or migration tooling to leave the platform.

   • [claimed-docs] “Drives (beta): Attach persistent filesystem storage to sandboxes and reuse data across sandbox runs.”
   • [claimed-docs] “Snapshotting: Save the state of a running sandbox to resume later. Skip dependency installation on subsequent runs.”
   • [claimed-docs] “If you already use Docker images to define your environment, store the image in Vercel Container Registry (VCR) and create the sandbox with …”

   proof ↗⚑ flag

   Maritime—none0/10

   Evidence shows import/migration INTO Maritime (from AWS/Heroku/VPS, .env bulk import) and a CLI with --json output, but nothing documents exporting agent configs, data, secrets, or state OUT in open formats to leave the platform. missing for 10: an export command or documented data-portability feature, open-format export of agent state/configs, any migration-out guide.

   • [claimed-docs] “Bulk import from a .env file (or stdin), then hot-reload”
   • [claimed-docs] “Migrate from AWS, Heroku, or a VPS”
   • [claimed-docs] “Pass --json on EVERY command. Success: one JSON value on stdout, stderr empty.”

   proof ↗⚑ flag

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

   weight 2 · round drawn
   Vercel Sandbox—none0/10

   No evidence that Vercel Sandbox's source code is published under an open license; the evidence pack only covers docs, features, pricing, and API surface, with no mention of a public repo or license.

   ⚑ flag

   Maritime—none0/10

   Maritime is a closed commercial SaaS platform; there is no evidence of an open-source license or public source repository anywhere in the docs or probes. missing for 10: an open-source license, a public source code repository, any mention of licensing terms for the platform's own code.

   ⚑ flag

4. ai-native userSelf-host the core product

   weight 3 · round drawn
   Vercel Sandbox—none0/10

   Vercel Sandbox is a fully managed, hosted service running on Vercel's own microVM infrastructure (iad1/sfo1/cle1/cdg1 regions); there is no evidence of any self-hostable core, open-source runtime, or on-prem deployment option — it is inherently tied to Vercel's cloud.

   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”
   • [claimed-docs] “Vercel Sandbox is available in the `iad1` (default), `sfo1`, `cle1`, and `cdg1` regions. You can choose the region per sandbox or set a proj…”
   • [claimed-docs] “Vercel Sandbox usage is metered across several dimensions. This page explains how billing works for each plan, what quotas apply, and how to…”

   proof ↗⚑ flag

   Maritime—none0/10

   Maritime is explicitly described as a cloud platform with agents running on serverless infrastructure, and no documentation offers a self-hosted deployment option for the core platform itself (Docker images are for agent frameworks running atop Maritime's cloud, not for self-hosting Maritime).

   • [probe] “PROBE llms.txt: HTTP 200 at https://maritime.sh/llms.txt # Maritime > Maritime is a cloud platform for deploying, hosting, and managing AI …”
   • [claimed-docs] “Migrate from AWS, Heroku, or a VPS”

   proof ↗⚑ flag

Latency

1. developerStart sandboxes with documented sub-second-to-few-second cold starts

   weight 3 · round drawn
   Vercel Sandbox—none0/10

   The docs describe microVM isolation, snapshotting for faster restarts, and persistence, but no evidence gives concrete cold-start latency figures (e.g., sub-second or few-second boot times) for starting a fresh sandbox. Missing for 10: documented benchmark numbers or explicit cold-start time claims, any independent performance measurement.

   • [claimed-docs] “Using a snapshot is much faster than creating from scratch because it avoids reinstalling dependencies and repeating setup steps.”
   • [claimed-docs] “when a sandbox stops, the SDK automatically snapshots its filesystem, and the sandbox configuration is preserved across sessions, so both ar…”

   proof ↗⚑ flag

   Maritime—none0/10

   Maritime's docs describe sleep/wake behavior (e.g., 'Sleeping agents wake automatically', 'puts it to sleep when idle, and wakes it on the next visit') and an 'always-on' add-on for millisecond reaction, but no evidence anywhere documents actual cold-start latency numbers (sub-second to few-second) for waking a sandboxed agent from sleep.

   • [claimed-docs] “Send a message and wait for the reply. Sleeping agents wake automatically.”
   • [claimed-docs] “Any repo with a Dockerfile can be served on a public, no-login HTTPS URL. Maritime builds it, runs it serverlessly, puts it to sleep when id…”
   • [claimed-docs] “Always-on is a flat add-on for the rare agent that must react in milliseconds around the clock.”

   proof ↗⚑ flag

Scale

1. platform-engineerRun large concurrent fleets of sandboxes with documented concurrency limits

   weight 3 · round to Vercel Sandbox
   Vercel Sandbox~partialclaimed3/10

   Docs mention that Sandbox usage is metered and that quotas apply per plan, plus region availability and spend management for cost control, but no concrete documented concurrency limits (e.g., max simultaneous sandboxes) are shown in the evidence. missing for 10: explicit concurrency limit numbers, guidance on running large fleets of sandboxes concurrently, and any performance/scale benchmarks.

   • [claimed-docs] “Vercel Sandbox usage is metered across several dimensions. This page explains how billing works for each plan, what quotas apply, and how to…”
   • [claimed-docs] “Vercel Sandbox is available in the `iad1` (default), `sfo1`, `cle1`, and `cdg1` regions. You can choose the region per sandbox or set a proj…”
   • [claimed-docs] “To control costs, configure Spend Management to receive alerts or pause projects when you reach a specified amount.”

   proof ↗⚑ flag

   Maritime—none0/10

   No evidence of documented concurrency limits or guidance for running large fleets of sandboxes at scale; docs cover provisioning, idempotency, and billing per-agent but never state concurrency ceilings, throughput benchmarks, or fleet-scale operational guidance.

   ⚑ flag

2. developerKeep a sandbox session running for hours or days for long agent tasks

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox~partialclaimed6/10

   Docs show sandboxes have a default 5-minute timeout that can be explicitly extended via `timeout` option or `sandbox.extendTimeout()`, and persistent sandboxes auto-snapshot on stop so you can quickly resume a long agent task across sessions rather than losing state. However, there's no evidence of an unbounded or multi-day continuous session—developers must actively manage/extend timeouts or rely on stop/resume snapshotting rather than a single always-on session. missing for 10: documented maximum extendable duration or explicit support for a single continuous multi-hour/day session without developer-managed timeout extensions, independent corroboration of long-running stability.

   • [claimed-docs] “You can configure this using the `timeout` option when creating a sandbox, and extend it using `sandbox.extendTimeout()`.”
   • [claimed-docs] “The default timeout is 5 minutes. You can configure this using the `timeout` option when creating a sandbox, and extend it using `sandbox.ex…”
   • [claimed-docs] “Persistent sandboxes: Sandboxes that auto-save state on stop and resume where you left off. Persistence is the default. No manual snapshot m…”
   • [claimed-docs] “Sandboxes that auto-save state on stop and resume where you left off. Persistence is the default. No manual snapshot management needed.”
   • [claimed-docs] “when a sandbox stops, the SDK automatically snapshots its filesystem, and the sandbox configuration is preserved across sessions, so both ar…”
   • [claimed-docs] “Using a snapshot is much faster than creating from scratch because it avoids reinstalling dependencies and repeating setup steps.”

   proof ↗⚑ flag

   Maritime~partialprobed5/10

   Maritime advertises serverless infra by default with an 'always-on' add-on for agents that must run continuously, and a persistent Linux desktop model for end users, suggesting long-lived sessions are possible but only via a paid add-on rather than a default guarantee for hours/days-long sandbox sessions. missing for 10: explicit documentation of session/session-timeout limits, confirmation that a sandbox session can run unattended for hours or days without the always-on add-on, and independent verification of long-running task stability.

   • [claimed-docs] “Always-on is a flat add-on for the rare agent that must react in milliseconds around the clock.”
   • [claimed-docs] “A persistent Linux desktop (XFCE, Chromium, LibreOffice) for each of _your_ end users, driven by _your_ model.”
   • [probe] “PROBE llms.txt: HTTP 200 at https://maritime.sh/llms.txt # Maritime > Maritime is a cloud platform for deploying, hosting, and managing AI …”

   proof ↗⚑ flag

Pricing

1. developerStart building with a free tier or included credits without talking to sales

   weight 1 · round to Maritime
   Vercel Sandbox—none0/10

   The evidence pack mentions billing, plans, and quotas (e.g., vercel-sandbox-docs-42, docs-24) but never explicitly states a free tier or included credits usable without contacting sales. Missing for 10: explicit free-tier/plan pricing details, evidence of self-serve signup without sales contact.

   • [claimed-docs] “Vercel Sandbox usage is metered across several dimensions. This page explains how billing works for each plan, what quotas apply, and how to…”
   • [claimed-docs] “To control costs, configure Spend Management to receive alerts or pause projects when you reach a specified amount.”

   proof ↗⚑ flag

   Maritime✓fullclaimed8/10

   Docs explicitly state the free plan includes 3 agents with no card required, and quickstart/CLI docs show a self-serve path (npm install, deploy in under 5 minutes) with no sales contact implied anywhere. missing for 10: independent/third-party confirmation of the free tier's real-world limits and no hands-on verification of the signup flow.

   • [claimed-docs] “the free plan includes 3 agents, no card required”
   • [claimed-docs] “Deploy your first agent and talk to it, in under 5 minutes.”
   • [claimed-docs] “npm install -g maritime-cli”
   • [claimed-docs] “One subscription, a set number of agents, flat prices for extras. No meters, no hourly rates: how much your agents run never changes the bil…”

   proof ↗⚑ flag

2. platform-engineerPay per second only for the compute a sandbox actually uses

   weight 3 · round to Vercel Sandbox
   Vercel Sandbox~partialclaimed4/10

   Docs confirm Sandbox usage is metered on a pricing page with configurable timeouts and spend-management alerts to control costs, implying granular, usage-based billing, but no evidence explicitly states billing is charged 'per second' of compute used. missing for 10: explicit per-second billing granularity documentation, real-world billing statement/invoice examples showing sub-minute metering.

   • [claimed-docs] “Vercel Sandbox usage is metered across several dimensions. This page explains how billing works for each plan, what quotas apply, and how to…”
   • [claimed-docs] “You can configure this using the `timeout` option when creating a sandbox, and extend it using `sandbox.extendTimeout()`.”
   • [claimed-docs] “The default timeout is 5 minutes. You can configure this using the `timeout` option when creating a sandbox, and extend it using `sandbox.ex…”
   • [claimed-docs] “To control costs, configure Spend Management to receive alerts or pause projects when you reach a specified amount.”

   proof ↗⚑ flag

   Maritime—none0/10

   Maritime's own billing docs explicitly state the opposite of per-second usage billing: 'One subscription, a set number of agents, flat prices for extras. No meters, no hourly rates: how much your agents run never changes the bill' and 'Always-on is a flat add-on.' There is no evidence of any per-second or usage-metered compute pricing for sandboxes.

   • [claimed-docs] “One subscription, a set number of agents, flat prices for extras. No meters, no hourly rates: how much your agents run never changes the bil…”
   • [claimed-docs] “Always-on is a flat add-on for the rare agent that must react in milliseconds around the clock.”

   proof ↗⚑ flag

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

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox~partialclaimed5/10

   Docs confirm region selection (iad1, sfo1, cle1, cdg1) can be set per sandbox or as a project default, giving users some control over where sandboxes execute, and one changelog note claims availability 'in all regions.' However, there is no explicit data residency/compliance guarantee, no mention of where persistent Drives/snapshots/data-at-rest are stored, and no discussion of data sovereignty controls beyond compute region choice. missing for 10: explicit data-at-rest residency guarantees, compliance/certification details tied to region choice, documentation on storage location for Drives/snapshots, and independent confirmation of region enforcement.

   • [claimed-docs] “Vercel Sandbox is available in the `iad1` (default), `sfo1`, `cle1`, and `cdg1` regions. You can choose the region per sandbox or set a proj…”
   • [claimed-docs] “Vercel Sandbox is now available in all regions”

   proof ↗⚑ flag

   Maritime—none0/10

   No evidence in the pack mentions data residency, region selection, or geographic storage controls anywhere in Maritime's docs or probes; nothing addresses where agent/user data is physically stored.

   ⚑ flag

2. ai-native userControl data retention and deletion

   weight 2 · round drawn
   Vercel Sandbox—none0/10

   Docs describe sandbox lifecycle (timeouts, stopping, persistence, snapshots) but there is no explicit statement about data retention policies, deletion guarantees, or privacy controls for stored sandbox data/drives. missing for 10: explicit data retention policy, explicit deletion/purge controls, privacy/compliance documentation for stored data.

   • [claimed-docs] “You can also stop sandboxes from the Vercel Dashboard by navigating to Observability > Sandboxes and clicking Stop Sandbox.”
   • [claimed-docs] “You can configure this using the `timeout` option when creating a sandbox, and extend it using `sandbox.extendTimeout()`.”
   • [claimed-docs] “Sandboxes that auto-save state on stop and resume where you left off. Persistence is the default. No manual snapshot management needed.”
   • [claimed-docs] “when a sandbox stops, the SDK automatically snapshots its filesystem, and the sandbox configuration is preserved across sessions, so both ar…”

   proof ↗⚑ flag

   Maritime—none0/10

   No evidence describes data retention policies, deletion controls, or export/purge mechanisms for agent data, files, or session logs; docs mention encrypted-at-rest secrets and append-only session logs but nothing about user-controlled retention or deletion.

   ⚑ flag

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

   weight 2 · round drawn
   Vercel Sandbox—none0/10

   No evidence in the pack addresses telemetry/usage-tracking opt-out for Vercel Sandbox; documentation covers execution, isolation, images, and pricing but nothing about telemetry controls.

   ⚑ flag

   Maritime—none0/10

   No evidence pack item mentions telemetry, usage tracking, analytics collection, or an opt-out mechanism anywhere in Maritime's docs; the closest topics (scoped API keys, secret variables) address access control, not telemetry preferences.

   ⚑ flag

Lifecycle

1. developerSet timeouts so sandboxes shut down automatically and stop billing when idle or done

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox✓fullclaimed8/10

   Docs explicitly state sandboxes have a default 5-minute timeout, configurable via the `timeout` option at creation, extendable via `sandbox.extendTimeout()`, and can be manually stopped from the dashboard, directly supporting automatic shutdown to stop billing when idle or done. Missing for 10: independent/hands-on confirmation that billing actually stops immediately upon timeout, and documentation of any maximum timeout limits or edge cases.

   • [claimed-docs] “You can configure this using the `timeout` option when creating a sandbox, and extend it using `sandbox.extendTimeout()`.”
   • [claimed-docs] “The default timeout is 5 minutes. You can configure this using the `timeout` option when creating a sandbox, and extend it using `sandbox.ex…”
   • [claimed-docs] “You can also stop sandboxes from the Vercel Dashboard by navigating to Observability > Sandboxes and clicking Stop Sandbox.”
   • [claimed-docs] “Vercel Sandbox usage is metered across several dimensions. This page explains how billing works for each plan, what quotas apply, and how to…”

   proof ↗⚑ flag

   Maritime—none0/10

   The docs describe Maritime as serverless agent hosting and mention an 'always-on' paid add-on (implying default agents are not always running), but no evidence describes a developer-configurable idle timeout or automatic shutdown-to-stop-billing mechanism. Missing for 10: any documented timeout setting, idle-detection policy, or billing-pause behavior tied to inactivity.

   • [claimed-docs] “Always-on is a flat add-on for the rare agent that must react in milliseconds around the clock.”

   proof ↗⚑ flag

2. developerSpin up an isolated sandbox with one API/SDK call and get a live environment in seconds

   weight 3 · round to Vercel Sandbox
   Vercel Sandbox✓fullclaimed8/10

   Docs show a clear SDK/CLI path (`@vercel/sandbox`, `sandbox create`/`sandbox run`) to spin up isolated Linux microVMs, with fast startup via managed images or snapshots that skip dependency installation, matching the 'one call, live environment in seconds' story. Missing for 10: independent/hands-on timing benchmarks confirming 'seconds' latency and third-party corroboration beyond first-party docs.

   • [claimed-docs] “Run untrusted or agent-generated code in isolated Linux microVMs. Use Vercel Sandbox for agent workflows, debugging, and one-off commands.”
   • [claimed-docs] “Use `@vercel/sandbox` for JavaScript and TypeScript”
   • [claimed-docs] “Use the `sandbox` CLI for manual testing, agentic workflows, debugging, and one-off operations”
   • [claimed-docs] “The Sandbox SDKs are the recommended way to integrate Vercel Sandbox into your applications. They provide a programmatic interface to create…”
   • [claimed-docs] “sandbox run echo "Hello Sandbox!"”
   • [claimed-docs] “Using a snapshot is much faster than creating from scratch because it avoids reinstalling dependencies and repeating setup steps.”
   • [claimed-docs] “Start sandboxes from Vercel's Managed Images, a shared or public image, or your own OCI images stored in Vercel Container Registry.”

   proof ↗⚑ flag

   Maritime✓fullprobed7/10

   Maritime's SDK exposes an idempotent `provision` call that creates/returns a live agent environment, and docs claim full agent deployment (image pull → live, chat-ready) in under 5 minutes, with CLI/API equivalents (`maritime keys create`, RESTful API). This matches the 'one call, live environment fast' story well, though the fastest documented time is minutes not literal seconds and there's no independent/hands-on latency benchmark or public OpenAPI spec confirming the exact call shape. Missing for 10: independent timing/benchmark evidence corroborating 'seconds', and a verifiable OpenAPI/API schema (probe shows openapi.json 404).

   • [claimed-docs] “Tag each agent with your own id via `externalId`. `provision` is **idempotent** on it: it returns the existing agent if there is one, else c…”
   • [claimed-docs] “provision is idempotent on it: it returns the existing agent if there is one, else creates a new one. Safe to call on every sign-in.”
   • [claimed-docs] “Deploy your first agent and talk to it, in under 5 minutes.”
   • [claimed-docs] “Every ready-made template is a real agent framework, pre-configured and one image pull from live. Pick one, name it, deploy.”
   • [claimed-docs] “RESTful API for managing agents programmatically.”
   • [claimed-docs] “The Maritime CLI lets you deploy and manage agents from your terminal.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://maritime.sh/openapi.json, https://maritime.sh/swagger.json, https://maritime.sh/api/openapi.…”

   proof ↗⚑ flag

Preview access

1. developerExpose a port from the sandbox on a public preview URL to reach services running inside

   weight 2 · round to Maritime
   Vercel Sandbox~partialclaimed5/10

   Docs mention running dev servers with 'live previews' and testing agent-generated code with access to 'live previews', implying some port-exposure/public-URL capability, but no evidence pack item describes the actual API/CLI mechanism (e.g., a domain/getUrl method, port argument, or URL format) for exposing a sandboxed port publicly. missing for 10: explicit documentation of the port-exposure API or CLI command, example of the resulting public URL, and confirmation of access control/authentication on that URL.

   • [claimed-docs] “Run development servers: Spin up and test applications with live previews.”
   • [claimed-docs] “Test in isolation: Preview how user-submitted or agent-generated code behaves in a self-contained environment with access to logs, file edit…”
   • [claimed-docs] “Test in isolation: Preview how user-submitted or agent-generated code behaves in a self”

   proof ↗⚑ flag

   Maritime~partialclaimed6/10

   Maritime's docs describe deploying any Dockerfile-based repo to a public, no-login HTTPS URL that reaches the running container's service, matching the core story of exposing an internal service via a public preview URL (maritime-docs-29, maritime-docs-47). However, the docs never explicitly describe port-selection mechanics (e.g., multiple ports, custom port binding) and all evidence is first-party with no independent/hands-on corroboration. missing for 10: explicit port-selection/multi-port docs, independent verification that the public URL reaches an arbitrary internal port.

   • [claimed-docs] “Any repo with a Dockerfile can be served on a public, no-login HTTPS URL. Maritime builds it, runs it serverlessly, puts it to sleep when id…”
   • [claimed-docs] “Any repo with a Dockerfile can be served on a public, no-login HTTPS URL.”

   proof ↗⚑ flag

Snapshots

1. developerPause a running sandbox and resume it later with filesystem and memory state intact

   weight 2 · round to Vercel Sandbox
   Vercel Sandbox✓fullclaimed8/10

   Docs explicitly describe persistent sandboxes that auto-save filesystem state on stop and resume where left off by default, with configuration preserved across sessions, plus explicit snapshotting to skip dependency reinstallation and resuming an existing sandbox by name. This directly matches the pause/resume-with-state story via first-party documentation. Missing for 10: independent hands-on verification that memory/process state (not just filesystem) is preserved across pause/resume, and no third-party corroboration beyond vendor docs.

   • [claimed-docs] “Persistent sandboxes: Sandboxes that auto-save state on stop and resume where you left off. Persistence is the default. No manual snapshot m…”
   • [claimed-docs] “Snapshotting: Save the state of a running sandbox to resume later. Skip dependency installation on subsequent runs.”
   • [claimed-docs] “Or retrieve an existing sandbox by name (resumes if stopped)”
   • [claimed-docs] “Sandboxes that auto-save state on stop and resume where you left off. Persistence is the default. No manual snapshot management needed.”
   • [claimed-docs] “Using a snapshot is much faster than creating from scratch because it avoids reinstalling dependencies and repeating setup steps.”
   • [claimed-docs] “when a sandbox stops, the SDK automatically snapshots its filesystem, and the sandbox configuration is preserved across sessions, so both ar…”

   proof ↗⚑ flag

   Maritime—none0/10

   No evidence in the pack mentions pausing/suspending a sandbox or preserving filesystem and memory state across a pause/resume cycle; docs cover provisioning, webhooks, triggers, templates, and CLI/API auth but nothing about snapshot or suspend-resume persistence.

   ⚑ flag

2. developerSnapshot a sandbox and later restore or fork new sandboxes from that snapshot

   weight 3 · round to Vercel Sandbox
   Vercel Sandbox~partialclaimed6/10

   Vercel Sandbox docs clearly describe automatic snapshotting of a sandbox's filesystem/state on stop and restoring it on resume, with named sandboxes resumable and snapshots speeding up subsequent runs by skipping setup. However, the evidence only shows resuming the *same* sandbox from its own snapshot, not explicitly forking multiple independent new sandboxes from one saved snapshot. Missing for 10: explicit support/documentation for forking multiple new sandboxes off a single snapshot, and any hands-on/independent confirmation of this fork workflow.

   • [claimed-docs] “Snapshotting: Save the state of a running sandbox to resume later. Skip dependency installation on subsequent runs.”
   • [claimed-docs] “Sandboxes that auto-save state on stop and resume where you left off. Persistence is the default. No manual snapshot management needed.”
   • [claimed-docs] “when a sandbox stops, the SDK automatically snapshots its filesystem, and the sandbox configuration is preserved across sessions, so both ar…”
   • [claimed-docs] “Or retrieve an existing sandbox by name (resumes if stopped)”
   • [claimed-docs] “Using a snapshot is much faster than creating from scratch because it avoids reinstalling dependencies and repeating setup steps.”

   proof ↗⚑ flag

   Maritime—none0/10

   No evidence in the pack mentions snapshotting a sandbox/agent state or restoring/forking new instances from such a snapshot; templates (maritime-docs-7) allow cloning a pre-configured agent package but not capturing/restoring runtime state.

   ⚑ flag