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 developer who uses Claude Code and deliberately avoids GitHub Copilotcan adopt gh-aw — the Quick Start supports --engine claude and every tool is engine-agnostic. But the docs treat Copilot as the privileged default: it's the fallback engine, gh aw init scaffolds Copilot artifacts, and the README's hidden agent-onboarding block hardcodes Copilot with no Claude equivalent. Key finding: a Claude user who follows the README (rather than the Quick Start) provisions Copilot and never sets ANTHROPIC_API_KEY.
Severity Findings (Critical → Major → Minor)
🔴 Critical Blockers
README onboarding is Copilot-only. The hidden agent-onboarding block (README.md:13-41) hardcodes engine: copilot and copilot-requests: write with no Claude path or secret guidance. A Claude user (or an agent) following the README verbatim sets up Copilot and never configures ANTHROPIC_API_KEY. Fix: add a Claude branch (engine: claude + ANTHROPIC_API_KEY) or link to the Quick Start engine selector.
🟠 Major Obstacles
gh aw init defaults to Copilot artifacts (cli.md:147); --engine claude is needed to skip them (cli.md:153) but isn't surfaced in Quick Start. Fix: show gh aw init --engine claude in the first-run path.
Silent revert to Copilot. Removing/omitting engine: reverts the runtime to Copilot (quick-start.mdx:75); 33 of 256 example workflows declare no engine and therefore run on Copilot. A Claude user copying an example silently gets Copilot. Fix: warn prominently and/or set an explicit engine in all examples.
Claude auth stops at the API key. Core docs name only ANTHROPIC_API_KEY (quick-start.mdx:81); Bedrock/Vertex, model selection, and enterprise auth are deferred to /reference/auth/ and /reference/engines/ (not in the core pages). Fix: cross-link these from Quick Start's Claude tab.
Example parity gap. 124 copilot vs 61 claude worked examples (≈157 vs 61 counting the 33 defaults). Custom engine has 0 examples. Fix: add Claude equivalents for flagship examples.
🟡 Minor Confusion
Copilot documents two auth paths — org billing (no PAT) or COPILOT_GITHUB_TOKEN PAT (cli.md:176) — while Claude/Codex/Gemini show only the single API-key path; no "org billing" note for non-Copilot engines.
web-search is engine-dependent/ambiguous (tools.md:62-67): Codex disables it by default; some engines need third-party MCP servers. A Claude user can't tell at a glance whether it "just works."
Secrets bootstrap (cli.md:265) accepts --engine {copilot,claude,codex,gemini} but doesn't restate the engine→secret-name mapping.
Engine & Tool Matrix
Engine
Setup
Examples
Auth
Score
Copilot (default)
gh aw init scaffolds it by default (cli.md:147)
124 (+33 defaults)
2 paths: org billing (no PAT) / COPILOT_GITHUB_TOKEN PAT (cli.md:176)
★★★★★
Claude
--engine claude / engine: claude (quick-start.mdx:69,163); init --engine claude skips Copilot (cli.md:153)
engine: codex; must declare web-search: to enable web (tools.md:67)
14
OPENAI_API_KEY / CODEX_API_KEY
★★☆☆☆
Custom
via mcp-servers (tools.md:184)
0 worked examples
n/a
★☆☆☆☆
Tools (all engine-agnostic — good news for Claude users):edit, github, bash, web-fetch, playwright, cache-memory, repo-memory, qmd, agentic-workflows, cli-proxy, custom mcp-servers — 11 engine-agnostic, 0 Copilot-only, 1 unclear (web-search, tools.md:62). No tool is gated to Copilot; Copilot's only privilege is being the default engine.
Parity: Copilot has ~2× Claude's example coverage; the default-engine bias widens the effective gap to ~157 vs 61. An experimental pi engine appears in 21 workflows (more than Codex's 14). Custom-engine users have no worked examples.
Auth Gaps
Core how-they-work/architecture pages contain zero secret/API-key setup; auth is entirely offloaded to Quick Start + linked /reference/auth/ (architecture.mdx, how-they-work.mdx).
Only Copilot gets a no-key/org-billing option documented; Claude/Codex/Gemini show a single API-key path only (cli.md:176).
gh auth login --scopes repo,workflow appears only in quick-start.mdx:34; cli.md:403 separately notes workflows:write — no consolidated auth-scope reference.
README onboarding block gives no secret-setup steps for any engine (README.md:13-41).
Recommended Actions
Priority 1 (Critical): Add a Claude branch to the README onboarding block (engine: claude + ANTHROPIC_API_KEY) so README-only readers aren't funneled into Copilot.
Priority 2 (Major): Surface gh aw init --engine claude in Quick Start; add an explicit engine to all 33 engine-unspecified example workflows and warn about the silent Copilot fallback; cross-link Claude enterprise auth (Bedrock/Vertex, model selection) from the Claude tab.
Priority 3 (Minor): Note whether non-Copilot engines have org-billing options; clarify web-search behavior per engine in tools.md; restate engine→secret mapping in cli.md secrets bootstrap; add Claude equivalents for flagship examples to close the parity gap.
Data source: automated doc-reader + engine-example-counter scans of README + 5 core docs and 256 workflow files (2026-07-15). Note: this run could not persist to cache-memory (sandbox blocked /tmp/gh-aw/cache-memory), so trend history was not updated.
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 developer who uses Claude Code and deliberately avoids GitHub Copilot can adopt gh-aw — the Quick Start supports
--engine claudeand every tool is engine-agnostic. But the docs treat Copilot as the privileged default: it's the fallback engine,gh aw initscaffolds Copilot artifacts, and the README's hidden agent-onboarding block hardcodes Copilot with no Claude equivalent. Key finding: a Claude user who follows the README (rather than the Quick Start) provisions Copilot and never setsANTHROPIC_API_KEY.Severity Findings (Critical → Major → Minor)
🔴 Critical Blockers
README.md:13-41) hardcodesengine: copilotandcopilot-requests: writewith no Claude path or secret guidance. A Claude user (or an agent) following the README verbatim sets up Copilot and never configuresANTHROPIC_API_KEY. Fix: add a Claude branch (engine: claude+ANTHROPIC_API_KEY) or link to the Quick Start engine selector.🟠 Major Obstacles
gh aw initdefaults to Copilot artifacts (cli.md:147);--engine claudeis needed to skip them (cli.md:153) but isn't surfaced in Quick Start. Fix: showgh aw init --engine claudein the first-run path.engine:reverts the runtime to Copilot (quick-start.mdx:75); 33 of 256 example workflows declare no engine and therefore run on Copilot. A Claude user copying an example silently gets Copilot. Fix: warn prominently and/or set an explicit engine in all examples.ANTHROPIC_API_KEY(quick-start.mdx:81); Bedrock/Vertex, model selection, and enterprise auth are deferred to/reference/auth/and/reference/engines/(not in the core pages). Fix: cross-link these from Quick Start's Claude tab.copilotvs 61claudeworked examples (≈157 vs 61 counting the 33 defaults). Custom engine has 0 examples. Fix: add Claude equivalents for flagship examples.🟡 Minor Confusion
COPILOT_GITHUB_TOKENPAT (cli.md:176) — while Claude/Codex/Gemini show only the single API-key path; no "org billing" note for non-Copilot engines.tools.md:62-67): Codex disables it by default; some engines need third-party MCP servers. A Claude user can't tell at a glance whether it "just works."cli.md:265) accepts--engine {copilot,claude,codex,gemini}but doesn't restate the engine→secret-name mapping.Engine & Tool Matrix
gh aw initscaffolds it by default (cli.md:147)COPILOT_GITHUB_TOKENPAT (cli.md:176)--engine claude/engine: claude(quick-start.mdx:69,163);init --engine claudeskips Copilot (cli.md:153)ANTHROPIC_API_KEYonly; enterprise deferred (quick-start.mdx:81)engine: codex; must declareweb-search:to enable web (tools.md:67)OPENAI_API_KEY/CODEX_API_KEYmcp-servers(tools.md:184)Tools (all engine-agnostic — good news for Claude users):
edit,github,bash,web-fetch,playwright,cache-memory,repo-memory,qmd,agentic-workflows,cli-proxy, custommcp-servers— 11 engine-agnostic, 0 Copilot-only, 1 unclear (web-search,tools.md:62). No tool is gated to Copilot; Copilot's only privilege is being the default engine.Parity: Copilot has ~2× Claude's example coverage; the default-engine bias widens the effective gap to ~157 vs 61. An experimental
piengine appears in 21 workflows (more than Codex's 14). Custom-engine users have no worked examples.Auth Gaps
/reference/auth/(architecture.mdx,how-they-work.mdx).cli.md:176).gh auth login --scopes repo,workflowappears only inquick-start.mdx:34;cli.md:403separately notesworkflows:write— no consolidated auth-scope reference.README.md:13-41).Recommended Actions
engine: claude+ANTHROPIC_API_KEY) so README-only readers aren't funneled into Copilot.gh aw init --engine claudein Quick Start; add an explicit engine to all 33 engine-unspecified example workflows and warn about the silent Copilot fallback; cross-link Claude enterprise auth (Bedrock/Vertex, model selection) from the Claude tab.web-searchbehavior per engine intools.md; restate engine→secret mapping incli.mdsecrets bootstrap; add Claude equivalents for flagship examples to close the parity gap.Data source: automated
doc-reader+engine-example-counterscans of README + 5 core docs and 256 workflow files (2026-07-15). Note: this run could not persist tocache-memory(sandbox blocked/tmp/gh-aw/cache-memory), so trend history was not updated.Beta Was this translation helpful? Give feedback.
All reactions