docs(issues): own gap inventory, split into fix vs add

[ManSio]

What

A dry, self-referential gap inventory appended to ISSUE.md. No external attribution: every referent is internal (file:line), every number comes from a command run in this repo.

Twenty items, split into F — fix what is ours and A — add what we do not have at all, plus an explicit list of things not to adopt.

Why these twenty

The list came from measuring this repo against itself rather than against a wish list: which gates are wired into nothing, which numbers in the docs cannot be produced by today's command, which guards are red, and which affordances exist only as prose.

The seven that matter most

	Item	Evidence
1	F5 — the protocol guard is red: exit 1, 7 findings (6 rate tools without an empty-population refusal, 1 frozen artifact without sha256), and it is in neither CI nor pre-commit. check_t10/check_t11 have no test assertion at all	scripts/audit_protocol_guards.py:66,87-98
2	F6 — three confirmed doc/code drifts: 31/64 vs actual 32/65 tools; 8 documented pre-commit checks vs 11 invoked; a published "8 reported, 62.5% false positives" against a current 7	server_tools.py:6,10 · .githooks/pre-commit:8-16 · run_all.py:71-73
3	F1 — the only item that hits the agent user rather than us: no typed degraded states, so "0 results" and "could not check" return the same value	search_tools.py:197,607
4	F2a — no exact runtime guard on the MCP surface; the only exact one sits behind importorskip, the other counts text	test_mcp_schema_flat.py:38 · auto_doc_updater.py:459
5	F4 — run_all.py, the "prove every guard can fail" suite, is opt-in	tools/verification/run_all.py
6	F8 — no CI job fails when a lane is flaky or cancelled	.github/workflows/ci.yml
7	A3 — the frozen-manifest mechanism exists; the requirement for hard negatives does not	frozen_overlap_check.py

Deliberately not adopted

Recorded in the file so nobody re-litigates it: shrinking 65 tools to 8 mega-tools (size is not protection without a discriminator guard), a JSON vector store with a lexical prefilter (we are already on LanceDB with an IVF guard), and an offline memory harness with no production caller.

Verification

• 57 insertions, 0 deletions — purely additive.
• Full pre-commit set green (fast mode, gate-zero deferred to CI as documented).

Not verified: nothing in this PR is executable code, so there is nothing to run. The claims are read-level claims about the current tree and are falsifiable by the referenced file:line.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: d2801469-dd78-4682-b302-87999656225e

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}