Published 2026-09-21 · Tested 2026-08-20
Warp
B+Warp received 7 PASS votes and passed 5 of five agent surface checks. The clearest finding came from the find the exact limits task.
Three AI models, GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, each read Warp’s public documentation independently and attempted five first-hour developer jobs: run the first cloud agent, find the exact limits, recover from a conflict, authenticate headless agent runs, use the TypeScript SDK.
No accounts, API calls, or code execution were used. Every verdict came from public pages and every published quotation passed a live verification check. Preflight substitutions: T3: A documented HTTP 409 agent-run conflict replaces HTTP 429. T4: API-key authentication replaces webhooks.
Freshness
How rechecks work- Category
- Coding agents
- Tested
- Quotes verified
- Surface rechecked
No change since the test.
4 of 5 quoted passages still appear on the live pages. The quoted page changed for Find the exact limits.
88.8% · 71/80 · AI Agent Readiness Score · reading 30 pts · surface 50 pts
Agent surface checks · 50/50
The Reading Test
| Task | GPT 5.6 Sol | Opus 5 | DeepSeek v4F | Consensus |
|---|---|---|---|---|
| Run the first cloud agent | PASS | PASS | PARTIAL | PASS |
| Find the exact limits | PARTIAL | PARTIAL | PARTIAL | PARTIAL |
| Recover from a conflict | PASS | PARTIAL | PASS | PASS |
| Authenticate headless agent runs | PASS | PASS | PASS | PASS |
| Use the TypeScript SDK | PARTIAL | FAIL | PARTIAL | PARTIAL |
docs platform: Astro Starlight (unscored) · verified 2026-08-20
What to fix first
These 2 fixes could add up to 7 points to the AI Agent Readiness Score. The list ranks each fix by the points it would add. How the ranking works
- 1
Found: The README's configuration example imports warp-sdk and reads task_id. Opus 5 also found no SDK code on docs.warp.dev, only a link to GitHub.
Fix: Add an oz-agent-sdk install and create-run example to docs.warp.dev, and use oz-agent-sdk and run_id in every README example.
Evidence: raw.githubusercontent.com/
warpdotdev/ oz-sdk-typescript/ main/ README (opens in a new tab) - 2
Found: The billing docs defer per-plan credits and seats to warp.dev/pricing. Two pages disagree on whether auto-reload fires at or below 100 credits.
Fix: List each plan's credits and seats in the billing docs, and use one auto-reload trigger on every billing page.
Evidence: docs.warp.dev/
support-and-community/ plans-and-billing/ pricing-faqs (opens in a new tab)
What the docs get right
- Authenticate headless agent runs: 3 PASS votes. The key page covers both key types, WARP_API_KEY, Oz CLI authentication, and REST bearer authentication through the linked quickstart.
- Run the first cloud agent: 2 PASS votes. The linked key, environment, run, status, and session steps form one complete API path.
- 5 of 5 agent surface checks. Present: llms.txt, llms-full.txt, markdown mirrors, an MCP server, docs AI.
Run the first cloud agent
PASSPASS consensus from 2 PASS, 1 PARTIAL.
The linked key, environment, run, status, and session steps form one complete API path. The API & SDK quickstart gives one numbered path with copyable curl for key export, run creation against POST /api/v1/agent/run, status polling on GET /agent/runs/<RUN_ID>, and the session_link, and it names how to obtain the environment ID with oz environment list. The API/SDK quickstart documents creating an API key, submitting a run, polling state, and viewing the session link via curl, but creating the required cloud environment redirects to the Cloud Agents Quickstart which uses the Warp desktop app, not the API. An agent starting from zero cannot create an environment purely through the documented API.
Find the exact limits
PARTIALPARTIAL consensus from 3 PARTIAL.
Plan values are findable, but two pages disagree on whether auto-reload fires at or below 100 credits. Add-on expiry (12 months from purchase), the 100-credit auto-reload threshold, and the $200 default monthly spend limit are stated exactly, but every per-plan credit allowance and seat limit is deferred off-docs to warp.dev/pricing, and the reset anchor is given three different ways across pages. Monthly credit allowances are findable (Build ~1,500 credits implied from Max's 12× reference, Max 18,000, Business 1,500/seat, Enterprise custom), but the Build plan's allowance is only implied, not stated explicitly. The documentation consistently delegates to the external marketing pricing page for exact numbers. Unused credits do not roll over; add-on credits roll over for 12 months; auto-reload triggers at 100 credits with a default $200/month spend cap.
Recover from a conflict
PASSPASS consensus from 2 PASS, 1 PARTIAL.
The 409 page identifies pending as the blocked state and tells readers to wait for in progress before retrying. The wait-and-retry correction is stated and the error is marked retryable, but the page's example response points cancellation at /api/v1/agent/tasks/abc123/cancel while the real endpoint is POST /agent/runs/{runId}/cancel, and its lowercase "pending" and "in progress" are not the literal RunState tokens an agent must compare against. The conflict error page documents the exact 409 scenario (cancelling a task in PENDING state), provides the full error response schema, marks the error as retryable, and gives a clear two-step resolution: wait for the task to transition to INPROGRESS, then retry the cancel request.
Authenticate headless agent runs
PASSPASS consensus from 3 PASS.
The key page covers both key types, WARP_API_KEY, Oz CLI authentication, and REST bearer authentication through the linked quickstart. The API keys page covers the whole path end to end: personal versus agent key semantics, creation in the Oz web app, the Warp app, or oz api-key create, the wk- prefix, export WARP_API_KEY, and the quickstart supplies the matching Authorization: Bearer $WARP_API_KEY header for REST. The API Keys page documents creating personal and agent API keys in the Oz web app or Warp app, setting WARP_API_KEY, authenticating the Oz CLI and REST API, and managing keys. The API/SDK quickstart reinforces the same flow with curl examples.
Use the TypeScript SDK
PARTIALPARTIAL consensus from 2 PARTIAL, 1 FAIL.
The main example reads run_id, but its configuration example switches to warp-sdk and reads task_id. No page on docs.warp.dev contains a package name, an install command, an import, a client constructor, or a single line of TypeScript, and even the SDK-titled demo page carries zero code blocks, so the docs alone cannot get an agent to a run_id. The SDK README shows npm install oz-agent-sdk, client initialization with WARP_API_KEY, and creating a run. The first example correctly uses response.run_id, but the second example uses response.task_id (deprecated per the SDK source, which says 'Use run_id instead') and imports from 'warp-sdk' instead of the installed 'oz-agent-sdk', creating inconsistency.
The receipt
Auto-reload automatically purchases more credits whenever your balance reaches 100 credits, ensuring uninterrupted access to premium AI features.
Plan values are findable, but two pages disagree on whether auto-reload fires at or below 100 credits.
Agent surface notes
Initialize returned an MCP Bearer OAuth challenge with valid resource metadata.
The live docs expose a public Ask Warp assistant.
Show the score
Paste this into a readme:
[](https://docsforagents.com/reports/warp-docs-ai-agent-readiness/) Method note
This is a reading test of public documentation, not an execution test. No accounts were created and no API calls were run. The AI Agent Readiness Score counts fifteen reading votes at PASS 2, PARTIAL 1, and FAIL 0, for 30 possible points. Five agent surface checks add 10 points each. The total is 80. Consensus chips show each row majority and do not affect scoring. The panel split on 3 of five tasks. Quotes shown here were re-fetched and confirmed verbatim on 2026-08-20.
Put another docs site through the battery.
Nominate a docs site