docs: document ingest/ingest-connect and gate the advertised method count

[devin-ai-integration]

Summary

SKILL.md routes ingest and ingest-connect (both have references, both are advertised in the README's "Pull from tools" row), but neither appeared in docs/skills.md or docs/skills-reference.md, and every doc claimed "35 methods". A number a reader cannot verify is worse than no number — especially for a tool whose pitch is dated, checkable claims.

Both methods are now documented, and the count is 37 everywhere (docs/skills.md, docs/skills-reference.md, docs/REPO_LAYOUT.md, adapters/README.md, adapters/LOCAL-LLM.md).

New check.js gate, so this cannot drift again. docs/skills-reference.md's six domain tables are the canonical per-method list, and the gate asserts:

routed(SKILL.md routing table)  ⊆  documented(skills-reference domain tables)
documented                      ⊆  names mentioned in docs/skills.md
claimed count (both docs)        =  |documented|

Verified by injecting each drift and confirming it fails:

injected	caught
bump one doc's count to 36	docs/skills-reference.md claims 36 methods; 37 are documented
drop the two new reference rows	SKILL.md routes skill(s) missing from docs/skills-reference.md: ingest, ingest-connect
drop the two names from docs/skills.md	docs/skills.md claims 37 methods; 35 are documented

Routed is 36 vs 37 documented because debrief is reached through the daily layer rather than the routing table, so the gate requires containment rather than equality there. 78/78 tests + gates pass.

Link to Devin session: https://app.devin.ai/sessions/f135381c4682413bae73dff38eb6d1a3
Requested by: @suboss87

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

Runtime verification — claims gate

Adversarially tested as one-at-a-time mutations in a throwaway copy, each with printed proof the mutation applied. Baseline: exit 0, 37 documented, 37 routed, 78/78 tests.

What the gate now catches

mutation	verdict	caught by
markdown-link routing row, method documented nowhere	exit 1	must name its reference as `references/<name>.md`
HTML <code> / HTML <a href> reference	exit 1	reference-shape hard-fail
``references/x.md`` (double backticks)	exit 1	accepted as routed → set equality (38 routed)
reference in the method cell only	exit 1	cannot read the method name
routing row naming a method but no reference	exit 1	routes 'ghost-method' without naming a reference
routing row split across two lines	exit 1	same, on the orphaned first line
[land](.../does-not-exist.md)	exit 1	target does not exist
[land](.../close.md)	exit 1	link text/target mismatch
documented target: #fragment / directory / trailing slash / ./references/ / absolute / wrong casing	exit 1 (all 6)	row unparsed → set + count checks
a method name removed from docs/skills.md while a hyphenated sibling remains (ingest behind ingest-connect)	exit 1	does not list method(s): ingest
duplicate documented row · phantom documented method · wrong count in a non-headline position · reworded headline	exit 1	—
untouched control · intentional duplicate routing row (review has two triggers)	exit 0	correctly permissive

The last three rows are the round-6 findings: \b treats - as a boundary, so \bingest\b matched inside ingest-connect and a method could disappear from the docs behind a hyphenated sibling (it affected audit, build, ingest — 3 of 37); and rows the reference-shape check never saw could route a method that nothing then required anyone to document. Both are closed in e7cec39: presence is matched with (?<![\w-])name(?![\w-]), and any routing row whose method cell names a method must name a reference — a - placeholder (CLI-only rows like fde prep) and the table separators are correctly exempt.

Four-way merge with #28+#30+#31 on Main, all four verified via merge-base --is-ancestor: 52 gates, 83/83. Only conflict is a one-hunk union in test/fde-cli.test.js with #30 (both append at the same EOF anchor, no logic overlap).