Skip to content

docs(airmcp): correct starter tool count to ~111 + ignore Codex config#286

Merged
heznpc merged 1 commit into
mainfrom
chore/heznpc-app-session-2026-06-10
Jun 9, 2026
Merged

docs(airmcp): correct starter tool count to ~111 + ignore Codex config#286
heznpc merged 1 commit into
mainfrom
chore/heznpc-app-session-2026-06-10

Conversation

@heznpc

@heznpc heznpc commented Jun 9, 2026

Copy link
Copy Markdown
Owner

The "~125 starter tools" figure was a polluted local measurement that had
spread to 5 places — README, docs/ios-architecture, the smoke-mcp and
verify-published-package comments, and the tool-count-drift test comment —
while the repo's own dump-tool-manifest.mjs already said ~111.

Clean-room measurement (env -i, fresh HOME, no AIRMCP_/GEMINI_) confirms
the default npx -y airmcp STARTER surface is 111 tools, four ways:
runtime tools/list (env-clean) = 111; runtime with AIRMCP_TEST_MODE=1 = 111
(TEST_MODE only gates shutdown + test-only error helpers, not module
registration); starter set composition coherent (the 162 excluded tools are
exactly the opt-in modules); and dump-tool-manifest.mjs already says ~111.

--full is unchanged: 273 runtime / 272 registerTool headline / 285 manifest
superset, still guarded by count-stats + tool-count-drift. No gate ever
asserted 125 (smoke/verify floor is >=100), so this was a stale-comment
honesty gap surfaced by the RFC 0014 capability-surface work, not a test
failure. Tilde kept — the starter subset is a curated soft value with the

=100 smoke floor as its only hard gate.

Also gitignore the root AGENTS.md (the Codex twin of the already-ignored
CLAUDE.md, carrying the same "Not published" design notes) and .codex/ (the
Codex analog of .claude/*). The published agent guide stays tracked at
.github/AGENTS.md; the root-anchored /AGENTS.md pattern does not shadow it.
Keeps internal design notes out of the public tree.

Tests: 123 suites / 1923 pass. stats:check clean.

The "~125 starter tools" figure was a polluted local measurement that had
spread to 5 places — README, docs/ios-architecture, the smoke-mcp and
verify-published-package comments, and the tool-count-drift test comment —
while the repo's own dump-tool-manifest.mjs already said ~111.

Clean-room measurement (env -i, fresh HOME, no AIRMCP_*/GEMINI_*) confirms
the default `npx -y airmcp` STARTER surface is 111 tools, four ways:
runtime tools/list (env-clean) = 111; runtime with AIRMCP_TEST_MODE=1 = 111
(TEST_MODE only gates shutdown + test-only error helpers, not module
registration); starter set composition coherent (the 162 excluded tools are
exactly the opt-in modules); and dump-tool-manifest.mjs already says ~111.

--full is unchanged: 273 runtime / 272 registerTool headline / 285 manifest
superset, still guarded by count-stats + tool-count-drift. No gate ever
asserted 125 (smoke/verify floor is >=100), so this was a stale-comment
honesty gap surfaced by the RFC 0014 capability-surface work, not a test
failure. Tilde kept — the starter subset is a curated soft value with the
>=100 smoke floor as its only hard gate.

Also gitignore the root AGENTS.md (the Codex twin of the already-ignored
CLAUDE.md, carrying the same "Not published" design notes) and .codex/ (the
Codex analog of .claude/*). The published agent guide stays tracked at
.github/AGENTS.md; the root-anchored /AGENTS.md pattern does not shadow it.
Keeps internal design notes out of the public tree.

Tests: 123 suites / 1923 pass. stats:check clean.
@heznpc heznpc enabled auto-merge (squash) June 9, 2026 19:02
@heznpc heznpc merged commit b7e8b80 into main Jun 9, 2026
4 checks passed
@heznpc heznpc deleted the chore/heznpc-app-session-2026-06-10 branch June 9, 2026 19:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant