[iscekic]

What

.kilo/skills/kilo-mcp/SKILL.md is generated from services/kilo-mcp/catalog.json, and every pull request that changes the catalog regenerates it. The census carried totals, per-area counts and a second-segment breakdown, so two pull requests that each added a procedure produced different bytes for the same file. Whichever merged second conflicted.

The census now derives one thing: the sorted list of first path segments, one bullet per area. No totals, no counts, no per-area breakdown.

Why this removes the conflict

• Two pull requests that add or remove procedures inside existing areas generate byte-identical skill files, so the drift check stays clean and CI does not commit the file back to the branch at all.
• A new or removed area produces a one-line diff in an alphabetically stable position.
• Counts never informed a decision: the skill already tells the agent to run kilo_search and never to pick a path from this file. Area names still orient the reader.

Verification

• pnpm exec jest src/scripts/mcp-catalog/skill.test.ts on a work VM with Postgres (17 tests, all pass). New tests pin "procedures added inside existing areas leave the skill byte-identical" and "a new area adds exactly one line".
• Regenerated the committed skill from the real catalog and diffed it against the previous file.
• Added two procedures under existing areas to the real catalog, regenerated, diffed: identical output. Then added a new area: exactly one line added.
• pnpm --filter web typecheck, oxlint, and oxfmt --list-different clean on the changed files.

Deploy notes

None. The generator, its template, and its tests change together, and CI regenerates the file from the catalog as before.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Executive Summary

The generator, template, committed skill, and tests change coherently: the census is reduced to a deterministic sorted list of first path segments, the checked-in SKILL.md matches the committed catalog's 49 areas, and no other in-repo consumer used the removed censusCatalog/readCatalogRows/Census exports.

Assumptions: [INFERENCE] Validation commands were not run (read-only review); the drift guarantees were verified by reading the generator, template, committed skill, catalog keys, and the CI workflow rather than by executing tests.

Files Reviewed (4 files)

• .kilo/skills/kilo-mcp/SKILL.md
• apps/web/src/scripts/mcp-catalog/skill-template.md
• apps/web/src/scripts/mcp-catalog/skill.test.ts
• apps/web/src/scripts/mcp-catalog/skill.ts

---

_{Reviewed by deepseek-flash-latest · Input: 61.1K · Output: 9K · Cached: 847.7K}

_{Review guidance: REVIEW.md from base branch main}