[claude-code-user-docs-review] 🔍 Claude Code User Documentation Review - 2026-05-30 #35911
Closed
Replies: 1 comment
-
|
This discussion has been marked as outdated by Claude Code User Documentation Review. A newer discussion is available at Discussion #36117. |
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 developer who uses Claude Code (and avoids GitHub Copilot products) can adopt gh-aw — Claude is a first-class engine with a dedicated
reference/engines.md, anANTHROPIC_API_KEYnote in Quick Start, andadd-wizard/init --engine claudepaths. The friction is defaults and discoverability, not missing capability: Copilot is the implicit default engine, the Quick Start narrative never setsengine: claudeinline (only optional Step 4), and the most likely persona trap —CLAUDE_CODE_OAUTH_TOKENis silently ignored — is buried inauth.mdxand absent from Quick Start.Key finding: A Claude Code user's instinct is to reuse their subscription via OAuth. gh-aw does not support
CLAUDE_CODE_OAUTH_TOKEN(auth.mdx:121-123) — only a pay-as-you-goANTHROPIC_API_KEY. Quick Start never warns about this, so the persona's first attempt can fail with no in-context explanation.Persona Context
Severity Findings
🚫 Critical Blockers (0 hard blockers — one silent auth trap)
No absolute blockers: the full flow completes via
gh aw add-wizard(engine menu includes Claude,quick-start.mdx:70) orgh aw init --engine claude(cli.md:141). The closest thing to a blocker is a silent auth trap:CLAUDE_CODE_OAUTH_TOKENunsupported & undocumented in Quick Start. OnlyANTHROPIC_API_KEYworks; OAuth / Claude Max / Teams subscription billing is explicitly not supported, and the OAuth token is ignored if set (auth.mdx:121-123, 199). Quick Start'sANTHROPIC_API_KEYnote (quick-start.mdx:84-88) omits this. For the Claude Code persona this is the most likely first-run failure.engine:resolves to Copilot (engines.md:24;how-they-work.mdx:26). Of 236 workflow files, 29 declare no engine and default to Copilot. A Claude-only user copying such a sample needsCOPILOT_GITHUB_TOKEN, which they cannot obtain.engine: claudeinline. It installsdaily-repo-statusand showsengine: claudeonly in optional Step 4 (quick-start.mdx:129-133). The sample's own default engine is unstated — if Copilot, the first run fails until overridden.gh aw initgenerates Copilot-specific artifacts by default (dispatcher skill + custom agent); user must pass--engine claudeto skip them (cli.md:135).COPILOT_GITHUB_TOKENis listed first in prerequisites and the secrets step, with the most detailed PAT walkthrough (quick-start.mdx:71, 77-82). Claude users must scan past Copilot setup.💡 Minor Confusion (paper cuts)
quick-start.mdx:29), wizard list (:70),secrets bootstrapexamples (cli.md:246).engines.md:28) — true, but no "why pick Claude" counterpoint beyondmax-turns.engines.mdis not linked from Quick Start, so the persona may never find the (excellent) Claude engine reference, feature table, and permission-mode security model.engines.md:40;tools.md:65), but only the Codex behavior is spelled out — Claude users get no heads-up.Engine Comparison
Ratings /5 (Setup clarity / Examples / Auth docs), from sub-agent data.
engines.mdstrong; QS note presentOPENAI_API_KEY/CODEX_API_KEYapi-targetdocumentedGEMINI_API_KEY, BYOK varsTool Availability
14 of 15 documented tools are engine-agnostic; none are copilot/claude/codex-only at config level.
edit,github,bash,web-fetch,playwright,cache-memory,repo-memory,qmd,agentic-workflows,cli-proxy,mcp-servers,tools.timeout,tools.startup-timeout, registry.web-search— the one engine-dependent entry. Native opt-in on Codex; Claude & most engines route via a third-party MCP server (tools.md:62-67;engines.md:40).engines.md:34-45):engine.agentcustom agent files ❌,engine.harness❌,max-continuations/autopilot ❌ are Copilot-only. Claude uniquely getsmax-turns✅. Tooling is portable; a few engine features are Copilot-exclusive.copilotBYOK config (COPILOT_PROVIDER_*,engines.md:180-214) is Copilot-only — irrelevant to a Copilot-avoiding persona.Authentication Gaps
COPILOT_GITHUB_TOKENGITHUB_TOKENANTHROPIC_API_KEYCLAUDE_CODE_OAUTH_TOKENignored (auth.mdx:121-123); no 401/403 help in QSOPENAI_API_KEY/CODEX_API_KEYCODEX_API_KEYwins at runtimeGEMINI_API_KEYClaude-only persona gaps: (1) OAuth-unsupported warning missing from QS; (2) no pointer to extra config for Claude + GitHub MCP (
auth.mdx:125); (3) no Claude 401/403 troubleshooting outside auth.mdx; (4) manualengine: claudeusers get no standalone "setANTHROPIC_API_KEY" instruction.Example Parity
From 236
.github/workflows/*.mdfiles:Claude is well represented (2nd, 63 examples) — not a parity blocker. Real gaps are Gemini/custom (≤1 each).
Recommended Actions
Priority 1 — remove the silent auth trap
quick-start.mdx:84-88: "CLAUDE_CODE_OAUTH_TOKEN/ Claude Max & Teams subscriptions are not supported — use a pay-as-you-goANTHROPIC_API_KEYfrom the Anthropic Console."engines.mdandauth.mdxClaude troubleshooting from Quick Start.Priority 2 — neutralize Copilot-first defaults in onboarding
engine: claudeinline in the main Quick Start narrative (not just optional Step 4), or state the sample's default engine and how to override before first run.Priority 3 — discoverability & parity paper cuts
engines.md:26-28(max-turns, no Copilot license, model choice).web-searchMCP requirement alongside the Codex note intools.md.gh aw init --engine claudeguidance earlier (already atcli.md:135).Conclusion
Can Claude Code users adopt gh-aw? Yes — overall 7.5/10. Claude is a genuine first-class engine:
engines.mdis thorough (feature table, permission-mode security model, BYOK, timeouts), tools are 14/15 engine-agnostic, and 63 Claude examples exist. The deductions are about defaults and discoverability for the Copilot-avoiding persona: Copilot is the silent default, the Quick Start is Copilot-shaped, and theCLAUDE_CODE_OAUTH_TOKEN-ignored trap — the single most persona-relevant pitfall — is absent from Quick Start. Fixing Priority 1 & 2 raises this to ~9/10.References: Workflow run §26684175861 — https://github.com/github/gh-aw/actions/runs/26684175861
Beta Was this translation helpful? Give feedback.
All reactions