Repository navigation
[claude-code-user-docs-review] 🔍 Claude Code User Documentation Review - 2026-10-31 #73504
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Executive Summary
A Claude Code user can complete onboarding without Copilot, but the docs are Copilot-first and the Claude path is thin. Key finding: a Claude Pro/Max or
claude logintoken is explicitly unsupported, and the docs never say that the requiredANTHROPIC_API_KEYis billed separately from a Claude subscription.Scope: README plus 5 core docs (quick-start, how-they-work, architecture, tools, cli). The linked Claude engine page and auth reference were not in the reviewed set, so some gaps below may be answered there.
Snapshot
Severity Findings
Critical Blockers
ANTHROPIC_API_KEYor Anthropic WIF works. A Claude Code subscriber with no API account is stuck, and nothing warns them up front.quick-start.mdx:148,quick-start.mdx:153,cli.md:207,cli.md:287. A workaround exists (create an API key), so this is a surprise-cost and dead-end issue rather than a hard stop.Major Obstacles
engine: claude.quick-start.mdx:155,how-they-work.mdx:33engine.authYAML or federation-rule steps.quick-start.mdx:151,cli.md:207engine:silently selects Copilot, and no failure mode is documented for someone who copies a Copilot workflow.how-they-work.mdx:26,cli.md:183-185gh aw initwithout--engine claudecreates Copilot-only artifacts.cli.md:153,cli.md:174-181ANTHROPIC_API_KEYonly on the pull request path; the direct-push path is unexplained.cli.md:207,quick-start.mdx:127tools.md:253,tools.md:262Minor Confusion
quick-start.mdx:70,quick-start.mdx:51cli.md:180web-search:is declared; native vs MCP is not explained, andweb-fetchhas no Claude note.tools.md:128,tools.md:116-124gh aw doctordoes not checkANTHROPIC_API_KEY.cli.md:303-322anthropic/model needstools.github.mode: gh-proxyandtools.cli-proxy: true.quick-start.mdx:175-177architecture.mdx:303cli.md:410workflow_dispatch(all engines).cli.md:561-563Engine & Tool Matrix
Example counts come from top-level
.github/workflows/*.mdfrontmatter (shared fragments not scanned); they are repo workflows, not docs examples.quick-start.mdx:133-143)copilot-requests: writeorCOPILOT_GITHUB_TOKENinit --engine claude(cli.md:159)ANTHROPIC_API_KEYor WIF; OAuth rejectedquick-start.mdx:157-163)CODEX_API_KEYorOPENAI_API_KEYquick-start.mdx:167-177)Parity observations: Claude has about 47% of Copilot example volume. Of the 65 Claude declarations, 30 use the nested
id:form and 35 the inline form. Core-doc mentions are roughly 26 Claude vs 41 Copilot lines. No workflow declares enginecustom.Tool classification (14 configurable entries, 19 rows total):
edit,bash,githubtools.md:20-39,102-114)web-fetch,web-searchtools.md:128)linear,jiratools.md:41-100)cache-memory,repo-memory,drive-memory(private preview)tools.md:146-175)qmd,agentic-workflows,playwrighttools.md:134-213)cli-proxy,tools.timeout,startup-timeouttools.md:215-267)Auth Gaps
ANTHROPIC_API_KEY; keyless: Anthropic WIF withpermissions.id-token: write. Unsupported:CLAUDE_CODE_OAUTH_TOKEN(quick-start.mdx:147-153).quick-start.mdx:148); WIFengine.authexample (quick-start.mdx:151);doctorcheck for engine keys (cli.md:303-322); behavior on invalid or expired key; Claude model and cost guidance (architecture.mdx:809); whetherGITHUB_TOKENalone suffices.copilot-requests: write,assign-to-agent(needs a PAT; engine interaction unstated,cli.md:320), and Copilot agent/MCP wiring files not generated for Claude (cli.md:180-181).Recommended Actions
Priority 1
quick-start.mdx:148-153: Claude subscriptions do not work, API keys are billed separately, with a link to Anthropic Console.engine.authYAML example.Priority 2
quick-start.mdx:51,:70).engine:is omitted and how to convert a Copilot example to Claude (cli.md:183-185).cli.md:207).Priority 3
gh aw doctorand its docs to checkANTHROPIC_API_KEY.cli.md:180) and a web-search/web-fetch engine matrix (tools.md:116-128).tools.md:253).Note:
claudeexample counts shift from 63 to 65 since the last run due to the counting method (top-level workflows only); trend is not a real change.All reactions