[claude-code-user-docs-review] 🔍 Claude Code User Documentation Review - 2026-06-05 #37113
Closed
Replies: 1 comment
-
|
This discussion has been marked as outdated by Claude Code User Documentation Review. A newer discussion is available at Discussion #37309. |
Beta Was this translation helpful? Give feedback.
0 replies
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 (GitHub + Claude, no Copilot, no Copilot CLI) can adopt gh-aw — there are no hard blockers. The interactive
add-wizardlets them pick Claude and the docs fully coverANTHROPIC_API_KEY. The persistent friction is framing: Copilot is the silent compile-time default, and the single most important caveat for this persona — that a Claude subscription /CLAUDE_CODE_OAUTH_TOKENis not supported — lives only inreference/auth.mdxand is never surfaced in the Quick Start.Key finding: A Claude Code user almost certainly authenticates Claude Code via a Max/Pro subscription token (
claude setup-token→CLAUDE_CODE_OAUTH_TOKEN). gh-aw ignores that token; it requires a separate pay-as-you-goANTHROPIC_API_KEYfrom the Anthropic Console. This is documented (auth.mdx:121-123) but absent from the onboarding path where it matters.Persona Context
Severity Findings
No 🚫 Critical Blockers found — every onboarding step has a working Claude path.
M1 — Subscription/OAuth token unsupported, and the caveat is buried.
docs/.../reference/auth.mdx:121-123: "CLAUDE_CODE_OAUTH_TOKENis not supported... Provider-based OAuth authentication (such as billing through a Claude Teams or Claude Max subscription) is not supported... it will be ignored." This is the #1 trap for the persona: a Claude Code user's natural credential does not work, andsetup/quick-start.mdx(the page they read) never mentions it. They must obtain a separate billed API key. Fix: add a one-line callout to theANTHROPIC_API_KEYNOTE block inquick-start.mdx:84-88. (Tracked but unresolved — seeREADME.md:237.)M2 — Copilot is the silent compile-time default.
introduction/how-they-work.mdx:26andreference/engines.md:24("Copilot CLI is the default —engine:can be omitted"). A Claude user who edits a.mdwithout addingengine: claude(only shown as optional Step 4,quick-start.mdx:127-133) silently compiles a Copilot workflow that then fails for lack ofCOPILOT_GITHUB_TOKEN. Fix: in Quick Start, state the default explicitly and showengine: claudein the primary path, not an optional afterthought.M3 —
gh aw initgenerates Copilot-specific artifacts by default.setup/cli.md:135,140: must rungh aw init --engine claudeto "skip Copilot-specific artifacts." This is in the CLI reference but not surfaced inquick-start.mdx, so a Claude user following onboarding generates unused Copilot scaffolding. Fix: surface--engine claudein the Quick Start init step.💡 Minor Confusion (paper cuts)
m1 — Copilot-first ordering throughout onboarding.
quick-start.mdx:71listsCOPILOT_GITHUB_TOKENfirst, and:77-81gives it a dedicated NOTE before the ANTHROPIC one (:84-88). README (:40) and how-they-work (:26) also list Copilot first. Reads as "Copilot is the intended engine." Fix: lead with engine-neutral language or alphabetize.m2 — No "why Claude vs Copilot?" guidance. No doc helps the persona understand the trade-off or confirm feature parity; they must infer it from example counts.
m3 — No pre-flight check that the Anthropic key has billing/credits.
auth.mdx:197-199covers this reactively (401/403 troubleshooting) but Quick Start offers no proactive note that a key without credit will fail.Engine Comparison
--engine claudein CLI refsmoke-claude.md+ MCP examplesANTHROPIC_API_KEY(but OAuth caveat buried)OPENAI_API_KEY/CODEX_API_KEYdocumentedsetup/page at allshared/genaiscript.md)reference/engines.md+auth.mdx; no key-acquisition guidanceTool Availability
Of the 15 tools documented in
reference/tools.md, all are engine-agnostic exceptweb-search, which is engine-dependent (native opt-in on Codex pertools.md:62-67; via third-party MCP on others). There is nocopilottool —copilot/claude/codexare engine names, not tools.agentic-workflows,playwright,github,bash,cache-memory,safe-outputsare all engine-agnostic. No tool is locked to the Copilot engine, so Claude users lose no documented tool capability. Only engine-specific defaults differ (tool timeout: Claude 60s, Codex 120s —tools.md:157).Authentication Gaps
COPILOT_GITHUB_TOKEN(fine-grained PAT)ANTHROPIC_API_KEYquick-start.mdx:86)OPENAI_API_KEY(orCODEX_API_KEY)auth.mdx:135)*_BASE_URLsetup/page; assumes existing provider credsAuthoritative auth reference:
reference/auth.mdx. No per-engine auth pages exist undersetup/.Example Parity
Engine declarations across
.github/workflows/*.md: Copilot 126 · Claude 63 · Codex 14 · Custom 1, with 32 workflows using no explicit engine (implicit Copilot). The ~2:1 Copilot lead is largely because Copilot is the default and gets restated in nestedengine.idblocks. Qualitatively, Claude has comparable coverage — dedicatedsmoke-claude.md(on par withsmoke-copilot*.md), MCP examples, and presence across the same workflow categories.docs/.../examples/is engine-agnostic (relies on default). No example-coverage blocker for Claude.Recommended Actions
Priority 1 (close the persona trap): Add the
CLAUDE_CODE_OAUTH_TOKEN-not-supported caveat to theANTHROPIC_API_KEYNOTE inquick-start.mdx:84-88(resolves the open item atREADME.md:237).Priority 2 (de-Copilot the default path): State the compile-time default explicitly in Quick Start, show
engine: claudein the primary flow (not optional Step 4), and surfacegh aw init --engine claude.Priority 3 (polish): Add a short "Choosing an engine" section (why Claude, parity note), use engine-neutral ordering in secret lists, and add a billing/credit pre-check note for the Anthropic key.
Conclusion
Can Claude Code users adopt gh-aw? Yes — with avoidable friction. No hard blockers; a Claude-only path exists end-to-end. The cost is that the persona must (1) discover that their subscription token won't work and get a billed API key, and (2) opt into Claude against Copilot-default framing at several steps.
Overall score: 7.5 / 10 — stable vs. the 2026-06-01 and 2026-06-04 reviews (both 7.5). No regressions; the Priority-1 fix (
README.md:237) remains open across all three runs.References:
Beta Was this translation helpful? Give feedback.
All reactions