You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A Claude Code user reading the gh-aw documentation can successfully adopt the tool — there are no hard blockers, and engine: claude is a fully supported, well-documented option — but the path is noticeably steeper than the Copilot path at every step. The Quick Start, CLI reference, and most examples default to Copilot; Claude users have to read the engine table, override the default, and skip Copilot-only setup notes to make progress.
Key Finding: No critical blockers. The same four major obstacles and several minor paper-cuts identified in the previous four audits remain unfixed in this run — most notably a wrong Anthropic-Console URL at auth.mdx:109, an auth table that omits three supported engines, a Copilot-first Quick Start, and gh aw init silently creating a Copilot-only dispatcher agent file.
Persona Context
I reviewed this documentation as a developer who:
✅ Uses GitHub for version control
✅ Uses Claude Code as primary AI assistant
❌ Does NOT use GitHub Copilot
❌ Does NOT use Copilot CLI
❌ Does NOT have Copilot subscription
Question 1: Onboarding Experience
Can a Claude Code user understand and get started with gh-aw?
Yes, with caveats. The overview is product-neutral (README.md:40: "Supports GitHub Copilot, Claude (Anthropic), Codex (OpenAI), and Gemini (Google) — pick whichever AI account you already have"), and the Quick Start prerequisites list all four engines on equal footing (quick-start.mdx:29). gh aw add-wizard interactively prompts for an engine and the matching secret (quick-start.mdx:70), so the happy path does work for Claude.
However, the narrative of the Quick Start is Copilot-first:
The first NOTE box (quick-start.mdx:75-79) walks through COPILOT_GITHUB_TOKEN setup in detail.
The second NOTE box (quick-start.mdx:82-86) covers ANTHROPIC_API_KEY — shorter, no troubleshooting tip.
Codex and Gemini get no NOTE box at all, despite being listed as prerequisites.
The customization example at Step 4 (quick-start.mdx:127-131) shows engine: claude as the override — but only after the workflow has been added with whatever the wizard picked.
Specific Issues Found:
quick-start.mdx:75-90: setup tips cover only 2 of 4 advertised engines
engines.md:24: "Copilot CLI is the default" is a one-liner; nothing in the Quick Start mentions that omitting engine: lands on Copilot
README.md:39-40: "How It Works" link is fine, but the Overview never explains the engine-choice tradeoffs
Recommended Fixes:
Add parallel NOTE boxes for Codex and Gemini in quick-start.mdx
Add a one-line "Default engine is Copilot — set engine: claude in the frontmatter to change it" callout in Step 2 before the wizard runs
Surface the "Which engine should I choose?" paragraph from engines.md:26-28 in the Quick Start or How It Works page
Question 2: Inaccessible Features for Non-Copilot Users
What features or steps don't work without Copilot?
engines.md:34-45 is the single most honest piece of cross-engine documentation in the repo and clearly enumerates the asymmetries. From a Claude user's perspective:
Features That Require Copilot:
engine.agent — custom agent files in .github/agents/ (engines.md:41, engines.md:111-121)
COPILOT_PROVIDER_* BYOK mode for routing through other providers (engines.md:201-268)
The dispatcher agent file generated by gh aw init at .github/agents/agentic-workflows.agent.md (cli.md:134) is in Copilot agent-file format and is not used by other engines
Features That Work Without Copilot (engine-agnostic):
engine.permission-mode — actually Claude-only (engines.md:439-498), a positive Claude-specific extension
Missing Documentation:
cli.md:134 describes the dispatcher agent file created by gh aw init but never warns it is Copilot-only. A Claude user running gh aw init will end up with a .github/agents/ file the engine doesn't use, with no signal that this is OK.
No "what's the Claude equivalent of engine.agent?" guidance — the answer is "use engine.bare plus a self-contained prompt" but that's not stated anywhere.
No "what's the Claude equivalent of max-continuations?" guidance — answer is "Claude has max-turns but does not have autopilot continuations" but that's only inferable from the table at engines.md:34-45.
Question 3: Documentation Gaps and Assumptions
Where does the documentation assume Copilot usage?
Copilot-Centric Language Found In:
cli.md:134 — "Initialize repository for agentic workflows. Configures .gitattributes, creates the dispatcher agent file (.github/agents/agentic-workflows.agent.md)" — describes a Copilot agent-file format without flagging the engine dependency
architecture.mdx:228 — sample firewall config example uses engine: copilot with no Claude variant shown
engines.md:62-69, engines.md:111-121, engines.md:295-301 — extended-config examples all use id: copilot
auth.mdx:17-23 — the "Which secret do I need?" lookup table only lists 4 of 7 supported engines; Crush, OpenCode, and Pi are absent despite being documented in engines.md:20-22
auth.mdx:109 — **the API key creation link still points to (platform.claude.com/redacted) (a docs landing page, not a key-creation page). The correct URL is used at quick-start.mdx:84andauth.mdx:199 (https://console.anthropic.com/settings/keys` / https://console.anthropic.com/). This was flagged in 4 prior audits and is still not fixed.
how-they-work.mdx:26 — engine intro line — Copilot listed first and tagged "(default)"; ordering choice acceptable, but no compensating "you can use any of these" emphasis
Missing Alternative Instructions:
No worked Claude example next to Copilot examples in architecture.mdx and engines.md extended-config sections
No "porting a workflow from Copilot to Claude" mini-guide despite ~30 files containing engine: copilot and only 6 containing engine: claude
The engine-env-secrets-to-engine-config codemod (cli.md:268) is described without explaining which engines benefit from it (it applies regardless of engine)
Severity-Categorized Findings
🚫 Critical Blockers (Score: 0/10)
No critical blockers identified. A Claude user can complete the Quick Start, run a workflow, and customize it. The add-wizard interactive engine selector and gh aw secrets bootstrap --engine claude (cli.md:240-245) bridge the worst gaps.
⚠️ Major Obstacles (Score: 4/10)
Obstacle 1: Wrong Anthropic API key URL in auth reference
Impact: A Claude user clicking the link in auth.mdx to create their API key lands on a docs landing page instead of the keys console.
Current State:auth.mdx:109 reads Create an API key at (platform.claude.com/redacted) The correct URL https://console.anthropic.com/settings/keys` is already used at quick-start.mdx:84 and https://console.anthropic.com/ at auth.mdx:199.
Why It's Problematic: Three URLs in three places for the same task in the same docs site — and the most authoritative one (the auth reference) is wrong. Flagged in 5 consecutive audits.
Suggested Fix: Change auth.mdx:109 to Create an API key in [Anthropic Console](https://console.anthropic.com/settings/keys). to match quick-start.mdx:84.
Impact: Users of Crush, OpenCode, and Pi cannot find their required secret in the auth reference.
Current State:auth.mdx:17-23 lists Copilot, Claude, Codex, Gemini. engines.md:20-22 lists Crush (COPILOT_GITHUB_TOKEN), OpenCode (COPILOT_GITHUB_TOKEN), and Pi (COPILOT_GITHUB_TOKEN or provider-specific).
Why It's Problematic: The auth page is the canonical "which secret do I need" reference. Three engines are silently missing from it. Flagged in 5 consecutive audits.
Suggested Fix: Add three rows to the auth.mdx:17-23 table mirroring engines.md:20-22. Note the experimental status inline.
Obstacle 3: `gh aw init` Copilot-only behavior undocumented
Impact: A Claude user running the recommended init command gets a .github/agents/agentic-workflows.agent.md file that their engine ignores, with no signal that this is expected.
Current State:cli.md:134 documents gh aw init as "Initialize repository for agentic workflows. Configures .gitattributes, creates the dispatcher agent file (.github/agents/agentic-workflows.agent.md)". No mention that the dispatcher agent file is a Copilot agent-file format.
Why It's Problematic: Silent format choice masks an engine assumption. A Claude user editing the file would expect Claude to read it. Flagged in 5 consecutive audits.
Suggested Fix: Add one sentence to cli.md:134: "The dispatcher agent file uses Copilot's .agent.md format — it is loaded by the Copilot engine only; other engines ignore it. Use --no-mcp to skip if you are not using Copilot." Also document the --no-mcp flag implications more clearly.
Affected Files:docs/src/content/docs/setup/cli.md
Obstacle 4: Quick Start setup tips only cover 2 of 4 advertised engines
Impact: Codex and Gemini users have no Quick-Start-level setup tips; they have to jump to the auth.mdx reference mid-flow.
Current State:quick-start.mdx:75-79 covers COPILOT_GITHUB_TOKEN setup. quick-start.mdx:82-86 covers ANTHROPIC_API_KEY setup. No parallel NOTE boxes for OPENAI_API_KEY or GEMINI_API_KEY despite both being in the prerequisites list (quick-start.mdx:29) and the wizard selector (quick-start.mdx:70).
Why It's Problematic: Breaks the parity the prerequisites list promised. Flagged in 4 consecutive audits.
Suggested Fix: Add two more NOTE boxes after quick-start.mdx:86, one each for Codex and Gemini, with the same shape (key-creation URL + gh aw secrets set snippet).
Issue 1: Engine timeout defaults split across two reference pages — tools.md:155 mentions Claude (60s) and Codex (120s) only; the full table is at engines.md:381-388. No cross-link. File: docs/src/content/docs/reference/tools.md:155
Issue 2:architecture.mdx:228 firewall example uses engine: copilot — substitute or duplicate with engine: claude for symmetry. File: docs/src/content/docs/introduction/architecture.mdx:228
Issue 3: No "what's the Claude equivalent of X?" mapping section anywhere — would close most lingering parity questions in one place
Issue 4: Engine extended-config examples (engines.md:62-91, 111-121, 270-303) all use id: copilot; even a single id: claude worked example would help
Issue 5:how-they-work.mdx:26 lists Copilot first with "(default)" but doesn't say engine: can be omitted only for Copilot — that fact lives only at engines.md:24
Issue 6:engine: custom is not actually a valid engine value per engines.md:14-22, yet this audit prompt and prior agent reports occasionally reference it — the docs are consistent here, but the audit prompt template itself should drop the term
Engine Comparison Analysis
Available Engines
Per engines.md:14-22, gh-aw supports seven engines:
tools.web-search for Codex is native and opt-in; for all others it goes through a third-party MCP server (tools.md:67, engines.md:52)
Tool allowlists work for Copilot, Claude, Codex, Gemini, Pi — not Crush or OpenCode (engines.md:45)
Unclear/Undocumented:
Default tool timeouts only listed for Claude and Codex in tools.md:155. Full table appears in engines.md:381-388 but tools.md doesn't cross-reference it.
Authentication Requirements
Current Documentation
Quick Start guide covers authentication for:
✅ Copilot — detailed NOTE box at quick-start.mdx:75-79
✅ Claude — present at quick-start.mdx:82-86 (shorter than Copilot's)
⚠️ Codex — listed in prerequisites, missing NOTE box
⚠️ Gemini — listed in prerequisites, missing NOTE box
⚠️ Crush / OpenCode / Pi — omitted entirely from quick-start
Missing for Claude Users
auth.mdx:109 link is wrong (see Obstacle 1)
No "is CLAUDE_CODE_OAUTH_TOKEN supported?" cross-link from quick-start.mdx — only auth.mdx:121-123 answers it
No troubleshooting tip in the quick-start.mdx:82-86 NOTE box (Copilot's NOTE box at line 75-79 is similarly thin, but Copilot also has a troubleshooting section in auth.mdx:97-99)
Copilot Examples: Saturate the docs; every architectural and config example defaults to Copilot. This is consistent and expected given the default-engine choice.
Claude Examples: Sufficient to learn the engine — engines.md:322-325, 344-352, 408-415, 447-451 all show idiomatic Claude config — but no end-to-end Claude workflow file is featured the way Copilot examples are in architecture.mdx. A Claude user has to mentally translate.
Recommended Actions
Priority 1: Critical Documentation Fixes
Fix auth.mdx:109 Anthropic URL — Change to https://console.anthropic.com/settings/keys to match quick-start.mdx:84. File: docs/src/content/docs/reference/auth.mdx
Warn that gh aw init dispatcher agent file is Copilot-only — One sentence at cli.md:134. File: docs/src/content/docs/setup/cli.md
Add Codex and Gemini NOTE boxes to Quick Start — Mirror the Claude/Copilot tip-boxes for parity. File: docs/src/content/docs/setup/quick-start.mdx
Priority 2: Major Improvements
Add "Default engine = Copilot" callout in Quick Start Step 2 — Before the wizard runs, mention that the default applies if engine: is omitted. File: docs/src/content/docs/setup/quick-start.mdx
Cross-link timeout defaults — tools.md:155 should link to the full per-engine table in engines.md:381-388. File: docs/src/content/docs/reference/tools.md
Add at least one Claude end-to-end example in architecture.mdx — Mirror the Copilot example at architecture.mdx:228. File: docs/src/content/docs/introduction/architecture.mdx
Add a "Choosing an engine" page or section in How It Works — Surface the paragraph at engines.md:26-28 earlier in the funnel. File: docs/src/content/docs/introduction/how-they-work.mdx
Priority 3: Nice-to-Have Enhancements
"Claude equivalent of X" mapping — Concise table explaining: engine.agent → no Claude equivalent (use prompt-only); max-continuations → use max-turns for in-run iteration only; etc.
A "Why Claude vs Copilot" decision aid — Expand engines.md:26-28 with concrete decision criteria (token cost, max-turns control, ecosystem fit).
Drop engine: custom references from any audit/prompt templates — it is not a real engine value.
Positive Findings
What Works Well
✅ README.md:40 puts all four engines on equal footing in the overview paragraph
✅ quick-start.mdx:29 lists all four engines as equally valid prerequisites
✅ quick-start.mdx:71 inline-cites all four secret docs anchors — strong wayfinding
✅ add-wizard interactively prompts for engine — no manual frontmatter editing required for first run
✅ engines.md:439-498 provides a substantial Claude-specific security/permission section
✅ auth.mdx:121-123 explicitly documents CLAUDE_CODE_OAUTH_TOKEN is unsupported — saves Claude Code Max/Teams subscribers from confusion
✅ gh aw secrets bootstrap --engine claude exists for Claude-specific secret validation (cli.md:240-245)
✅ Tool surface is mostly engine-agnostic — Claude users get the full power of playwright, cache-memory, safe-outputs, agentic-workflows, etc.
✅ gh aw new --engine claude (cli.md:202) injects engine: claude into generated frontmatter
Conclusion
Can Claude Code Users Successfully Adopt gh-aw?
Answer: Yes, with moderate friction.
Reasoning: No critical blocker prevents adoption. A Claude user who reads engines.md and the auth reference, and runs gh aw add-wizard, will get a working Claude workflow on the first try. The pain is cumulative: a wrong URL here, a missing NOTE box there, a Copilot-only file silently created during init, and a sea of Copilot-flavored examples to mentally translate. Each is small; together they signal "Copilot is the first-class citizen and Claude is a supported alternative."
The good news: every issue identified is a docs fix, not an engineering gap. The product itself genuinely supports Claude at parity for most authoring workflows. The same four major obstacles have been flagged in five consecutive audits — they are tractable, low-risk PRs.
Overall Assessment Score: 6/10
Breakdown:
Clarity for non-Copilot users: 6/10
Claude engine documentation: 7/10
Alternative approaches provided: 6/10
Engine parity: 5/10
Next Steps
Ship the four Priority-1 doc fixes — these have been outstanding for five audits and would shift the score to 7-8/10 immediately.
Add at least one Claude-engine worked example to architecture.mdx and the Quick Start.
Consider an "engine parity matrix" page that surfaces the comparison table from engines.md:34-45 as a first-class navigational entry.
Cross-checked: directory listings of docs/src/content/docs/examples/, docs/src/content/docs/guides/, and engine: literal occurrence counts across all docs
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
-
Executive Summary
A Claude Code user reading the gh-aw documentation can successfully adopt the tool — there are no hard blockers, and
engine: claudeis a fully supported, well-documented option — but the path is noticeably steeper than the Copilot path at every step. The Quick Start, CLI reference, and most examples default to Copilot; Claude users have to read the engine table, override the default, and skip Copilot-only setup notes to make progress.Key Finding: No critical blockers. The same four major obstacles and several minor paper-cuts identified in the previous four audits remain unfixed in this run — most notably a wrong Anthropic-Console URL at
auth.mdx:109, an auth table that omits three supported engines, a Copilot-first Quick Start, andgh aw initsilently creating a Copilot-only dispatcher agent file.Persona Context
I reviewed this documentation as a developer who:
Question 1: Onboarding Experience
Can a Claude Code user understand and get started with gh-aw?
Yes, with caveats. The overview is product-neutral (
README.md:40: "Supports GitHub Copilot, Claude (Anthropic), Codex (OpenAI), and Gemini (Google) — pick whichever AI account you already have"), and the Quick Start prerequisites list all four engines on equal footing (quick-start.mdx:29).gh aw add-wizardinteractively prompts for an engine and the matching secret (quick-start.mdx:70), so the happy path does work for Claude.However, the narrative of the Quick Start is Copilot-first:
quick-start.mdx:75-79) walks throughCOPILOT_GITHUB_TOKENsetup in detail.quick-start.mdx:82-86) coversANTHROPIC_API_KEY— shorter, no troubleshooting tip.quick-start.mdx:127-131) showsengine: claudeas the override — but only after the workflow has been added with whatever the wizard picked.Specific Issues Found:
quick-start.mdx:75-90: setup tips cover only 2 of 4 advertised enginesengines.md:24: "Copilot CLI is the default" is a one-liner; nothing in the Quick Start mentions that omittingengine:lands on CopilotREADME.md:39-40: "How It Works" link is fine, but the Overview never explains the engine-choice tradeoffsRecommended Fixes:
quick-start.mdxengine: claudein the frontmatter to change it" callout in Step 2 before the wizard runsengines.md:26-28in the Quick Start or How It Works pageQuestion 2: Inaccessible Features for Non-Copilot Users
What features or steps don't work without Copilot?
engines.md:34-45is the single most honest piece of cross-engine documentation in the repo and clearly enumerates the asymmetries. From a Claude user's perspective:Features That Require Copilot:
engine.agent— custom agent files in.github/agents/(engines.md:41,engines.md:111-121)engine.harness— custom harness script (engines.md:44,engines.md:293-315)max-continuations— autopilot multi-run mode (engines.md:38)COPILOT_PROVIDER_*BYOK mode for routing through other providers (engines.md:201-268)gh aw initat.github/agents/agentic-workflows.agent.md(cli.md:134) is in Copilot agent-file format and is not used by other enginesFeatures That Work Without Copilot (engine-agnostic):
edit,bash,github,web-fetch,playwright,cache-memory,repo-memory,qmd,agentic-workflows,cli-proxy,mcp-servers(tools.md:18-209)architecture.mdx:168-323)timeout-minutes,tools.timeout,tools.startup-timeout,max-runs,engine.api-target,engine.bare,engine.env,engine.args,engine.command(engines.md:43,431-437)engine.permission-mode— actually Claude-only (engines.md:439-498), a positive Claude-specific extensionMissing Documentation:
cli.md:134describes the dispatcher agent file created bygh aw initbut never warns it is Copilot-only. A Claude user runninggh aw initwill end up with a.github/agents/file the engine doesn't use, with no signal that this is OK.engine.agent?" guidance — the answer is "useengine.bareplus a self-contained prompt" but that's not stated anywhere.max-continuations?" guidance — answer is "Claude hasmax-turnsbut does not have autopilot continuations" but that's only inferable from the table atengines.md:34-45.Question 3: Documentation Gaps and Assumptions
Where does the documentation assume Copilot usage?
Copilot-Centric Language Found In:
cli.md:134— "Initialize repository for agentic workflows. Configures.gitattributes, creates the dispatcher agent file (.github/agents/agentic-workflows.agent.md)" — describes a Copilot agent-file format without flagging the engine dependencyarchitecture.mdx:228— sample firewall config example usesengine: copilotwith no Claude variant shownengines.md:62-69,engines.md:111-121,engines.md:295-301— extended-config examples all useid: copilotauth.mdx:17-23— the "Which secret do I need?" lookup table only lists 4 of 7 supported engines; Crush, OpenCode, and Pi are absent despite being documented inengines.md:20-22auth.mdx:109— **the API key creation link still points to(platform.claude.com/redacted) (a docs landing page, not a key-creation page). The correct URL is used atquick-start.mdx:84andauth.mdx:199(https://console.anthropic.com/settings/keys` /https://console.anthropic.com/). This was flagged in 4 prior audits and is still not fixed.how-they-work.mdx:26— engine intro line — Copilot listed first and tagged "(default)"; ordering choice acceptable, but no compensating "you can use any of these" emphasisMissing Alternative Instructions:
architecture.mdxandengines.mdextended-config sectionsengine: copilotand only 6 containingengine: claudeengine-env-secrets-to-engine-configcodemod (cli.md:268) is described without explaining which engines benefit from it (it applies regardless of engine)Severity-Categorized Findings
🚫 Critical Blockers (Score: 0/10)
No critical blockers identified. A Claude user can complete the Quick Start, run a workflow, and customize it. The
add-wizardinteractive engine selector andgh aw secrets bootstrap --engine claude(cli.md:240-245) bridge the worst gaps.Obstacle 1: Wrong Anthropic API key URL in auth reference
Impact: A Claude user clicking the link in
auth.mdxto create their API key lands on a docs landing page instead of the keys console.Current State:
auth.mdx:109readsCreate an API key at (platform.claude.com/redacted) The correct URLhttps://console.anthropic.com/settings/keys` is already used atquick-start.mdx:84andhttps://console.anthropic.com/atauth.mdx:199.Why It's Problematic: Three URLs in three places for the same task in the same docs site — and the most authoritative one (the auth reference) is wrong. Flagged in 5 consecutive audits.
Suggested Fix: Change
auth.mdx:109toCreate an API key in [Anthropic Console](https://console.anthropic.com/settings/keys).to matchquick-start.mdx:84.Affected Files:
docs/src/content/docs/reference/auth.mdxObstacle 2: Auth lookup table omits 3 supported engines
Impact: Users of Crush, OpenCode, and Pi cannot find their required secret in the auth reference.
Current State:
auth.mdx:17-23lists Copilot, Claude, Codex, Gemini.engines.md:20-22lists Crush (COPILOT_GITHUB_TOKEN), OpenCode (COPILOT_GITHUB_TOKEN), and Pi (COPILOT_GITHUB_TOKENor provider-specific).Why It's Problematic: The auth page is the canonical "which secret do I need" reference. Three engines are silently missing from it. Flagged in 5 consecutive audits.
Suggested Fix: Add three rows to the
auth.mdx:17-23table mirroringengines.md:20-22. Note the experimental status inline.Affected Files:
docs/src/content/docs/reference/auth.mdxObstacle 3: `gh aw init` Copilot-only behavior undocumented
Impact: A Claude user running the recommended init command gets a
.github/agents/agentic-workflows.agent.mdfile that their engine ignores, with no signal that this is expected.Current State:
cli.md:134documentsgh aw initas "Initialize repository for agentic workflows. Configures.gitattributes, creates the dispatcher agent file (.github/agents/agentic-workflows.agent.md)". No mention that the dispatcher agent file is a Copilot agent-file format.Why It's Problematic: Silent format choice masks an engine assumption. A Claude user editing the file would expect Claude to read it. Flagged in 5 consecutive audits.
Suggested Fix: Add one sentence to
cli.md:134: "The dispatcher agent file uses Copilot's.agent.mdformat — it is loaded by the Copilot engine only; other engines ignore it. Use--no-mcpto skip if you are not using Copilot." Also document the--no-mcpflag implications more clearly.Affected Files:
docs/src/content/docs/setup/cli.mdObstacle 4: Quick Start setup tips only cover 2 of 4 advertised engines
Impact: Codex and Gemini users have no Quick-Start-level setup tips; they have to jump to the
auth.mdxreference mid-flow.Current State:
quick-start.mdx:75-79coversCOPILOT_GITHUB_TOKENsetup.quick-start.mdx:82-86coversANTHROPIC_API_KEYsetup. No parallel NOTE boxes forOPENAI_API_KEYorGEMINI_API_KEYdespite both being in the prerequisites list (quick-start.mdx:29) and the wizard selector (quick-start.mdx:70).Why It's Problematic: Breaks the parity the prerequisites list promised. Flagged in 4 consecutive audits.
Suggested Fix: Add two more NOTE boxes after
quick-start.mdx:86, one each for Codex and Gemini, with the same shape (key-creation URL +gh aw secrets setsnippet).Affected Files:
docs/src/content/docs/setup/quick-start.mdx💡 Minor Confusion Points (Score: 6/10)
tools.md:155mentions Claude (60s) and Codex (120s) only; the full table is atengines.md:381-388. No cross-link. File:docs/src/content/docs/reference/tools.md:155architecture.mdx:228firewall example usesengine: copilot— substitute or duplicate withengine: claudefor symmetry. File:docs/src/content/docs/introduction/architecture.mdx:228engines.md:62-91, 111-121, 270-303) all useid: copilot; even a singleid: claudeworked example would helphow-they-work.mdx:26lists Copilot first with "(default)" but doesn't sayengine:can be omitted only for Copilot — that fact lives only atengines.md:24engine: customis not actually a valid engine value perengines.md:14-22, yet this audit prompt and prior agent reports occasionally reference it — the docs are consistent here, but the audit prompt template itself should drop the termEngine Comparison Analysis
Available Engines
Per
engines.md:14-22, gh-aw supports seven engines:engine: copilot— default, broadest feature set, well-documentedengine: claude— fully supported, well-documented, exemplary Claude-specific section atengines.md:439-498engine: codex— supported, lighter docs presenceengine: gemini— supported, lighter docs presenceengine: crush— experimental, very thin docsengine: opencode— experimental, very thin docsengine: pi— experimental, very thin docsengine: customis not a real engine value — ignore prior audit references to it.Documentation Quality by Engine
Tool Availability Analysis
Tools Review
Engine-Agnostic Tools (per
tools.mdandengines.md):edit,bash,github,web-fetch,playwright,cache-memory,repo-memory,qmd,agentic-workflows,cli-proxy,mcp-servers,tools.timeout,tools.startup-timeoutEngine-Specific Tools/Knobs:
tools.web-searchfor Codex is native and opt-in; for all others it goes through a third-party MCP server (tools.md:67,engines.md:52)engines.md:45)Unclear/Undocumented:
tools.md:155. Full table appears inengines.md:381-388buttools.mddoesn't cross-reference it.Authentication Requirements
Current Documentation
Quick Start guide covers authentication for:
quick-start.mdx:75-79quick-start.mdx:82-86(shorter than Copilot's)Missing for Claude Users
auth.mdx:109link is wrong (see Obstacle 1)CLAUDE_CODE_OAUTH_TOKENsupported?" cross-link fromquick-start.mdx— onlyauth.mdx:121-123answers itquick-start.mdx:82-86NOTE box (Copilot's NOTE box at line 75-79 is similarly thin, but Copilot also has a troubleshooting section inauth.mdx:97-99)Secret Names
COPILOT_GITHUB_TOKEN✅ANTHROPIC_API_KEY✅ (CLAUDE_CODE_OAUTH_TOKENexplicitly unsupported)OPENAI_API_KEYorCODEX_API_KEY✅GEMINI_API_KEY✅COPILOT_GITHUB_TOKEN(perengines.md:20-21) — surprising and not inauth.mdxtableCOPILOT_GITHUB_TOKENdefault, switches to provider-specific whenmodel:usesprovider/modelformat — not inauth.mdxtableExample Workflow Analysis
Workflow Count by Engine in Docs (
engine: <name>literal occurrences)Quality of Examples
Copilot Examples: Saturate the docs; every architectural and config example defaults to Copilot. This is consistent and expected given the default-engine choice.
Claude Examples: Sufficient to learn the engine —
engines.md:322-325, 344-352, 408-415, 447-451all show idiomatic Claude config — but no end-to-end Claude workflow file is featured the way Copilot examples are inarchitecture.mdx. A Claude user has to mentally translate.Recommended Actions
Priority 1: Critical Documentation Fixes
auth.mdx:109Anthropic URL — Change tohttps://console.anthropic.com/settings/keysto matchquick-start.mdx:84. File:docs/src/content/docs/reference/auth.mdxauth.mdx:17-23table — Add Crush, OpenCode, Pi rows matchingengines.md:20-22. File:docs/src/content/docs/reference/auth.mdxgh aw initdispatcher agent file is Copilot-only — One sentence atcli.md:134. File:docs/src/content/docs/setup/cli.mddocs/src/content/docs/setup/quick-start.mdxPriority 2: Major Improvements
engine:is omitted. File:docs/src/content/docs/setup/quick-start.mdxtools.md:155should link to the full per-engine table inengines.md:381-388. File:docs/src/content/docs/reference/tools.mdarchitecture.mdx— Mirror the Copilot example atarchitecture.mdx:228. File:docs/src/content/docs/introduction/architecture.mdxengines.md:26-28earlier in the funnel. File:docs/src/content/docs/introduction/how-they-work.mdxPriority 3: Nice-to-Have Enhancements
engine.agent→ no Claude equivalent (use prompt-only);max-continuations→ usemax-turnsfor in-run iteration only; etc.engines.md:26-28with concrete decision criteria (token cost, max-turns control, ecosystem fit).engine: customreferences from any audit/prompt templates — it is not a real engine value.Positive Findings
What Works Well
README.md:40puts all four engines on equal footing in the overview paragraphquick-start.mdx:29lists all four engines as equally valid prerequisitesquick-start.mdx:71inline-cites all four secret docs anchors — strong wayfindingadd-wizardinteractively prompts for engine — no manual frontmatter editing required for first runengines.md:34-45engine-feature comparison table is exemplary — honest, visible asymmetryengines.md:439-498provides a substantial Claude-specific security/permission sectionauth.mdx:121-123explicitly documentsCLAUDE_CODE_OAUTH_TOKENis unsupported — saves Claude Code Max/Teams subscribers from confusiongh aw secrets bootstrap --engine claudeexists for Claude-specific secret validation (cli.md:240-245)playwright,cache-memory,safe-outputs,agentic-workflows, etc.gh aw new --engine claude(cli.md:202) injectsengine: claudeinto generated frontmatterConclusion
Can Claude Code Users Successfully Adopt gh-aw?
Answer: Yes, with moderate friction.
Reasoning: No critical blocker prevents adoption. A Claude user who reads
engines.mdand the auth reference, and runsgh aw add-wizard, will get a working Claude workflow on the first try. The pain is cumulative: a wrong URL here, a missing NOTE box there, a Copilot-only file silently created during init, and a sea of Copilot-flavored examples to mentally translate. Each is small; together they signal "Copilot is the first-class citizen and Claude is a supported alternative."The good news: every issue identified is a docs fix, not an engineering gap. The product itself genuinely supports Claude at parity for most authoring workflows. The same four major obstacles have been flagged in five consecutive audits — they are tractable, low-risk PRs.
Overall Assessment Score: 6/10
Breakdown:
Next Steps
architecture.mdxand the Quick Start.engines.md:34-45as a first-class navigational entry.Appendix: Files Reviewed
Complete List of Documentation Files Analyzed
README.mddocs/src/content/docs/setup/quick-start.mdxdocs/src/content/docs/setup/cli.mddocs/src/content/docs/introduction/how-they-work.mdxdocs/src/content/docs/introduction/architecture.mdxdocs/src/content/docs/reference/tools.mddocs/src/content/docs/reference/engines.mddocs/src/content/docs/reference/auth.mdxdocs/src/content/docs/examples/,docs/src/content/docs/guides/, andengine:literal occurrence counts across all docsReferences:
Report Generated: 26450723204
Workflow: claude-code-user-docs-review
Engine Used: claude (eating our own dog food 🐕)
Beta Was this translation helpful? Give feedback.
All reactions