sequenceDiagram
    participant CI as CI Pipeline
    participant Catalog as Catalog Validator
    participant Files as README/AGENTS
    participant Repo as Repository (agents/, skills/, commands/)
    participant Report as Validation Report

    CI->>Catalog: run catalog.js --text
    Catalog->>Files: read README.md & AGENTS.md
    Files-->>Catalog: return documented counts & structure lines
    Catalog->>Repo: enumerate actual files/directories
    Repo-->>Catalog: return actual counts
    Catalog->>Catalog: build expectations (tables + project-structure)
    Catalog->>Catalog: compare expected vs actual
    alt match
        Catalog->>Report: success message
    else mismatch
        Catalog->>Report: failure details
        Report-->>CI: mark validation failed
    end

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.

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

Followed up on the review notes with one additional commit:

• hardened catalog.js AGENTS structure parsing to tolerate extra whitespace and - / – / — separators
• added a regression test covering those formatting variations
• opened translation drift follow-up issue #527 so the localized-doc scope exclusion is tracked explicitly

Validation rerun: pnpm test OK

Greptile Summary

This PR fixes catalog count drift by bumping stale documentation counts (agents: 16→21, skills: 65→102, commands: 40→52 in README.md; 20/65+/51→21/102/52 in AGENTS.md) and enforcing the existing catalog.js validator in both the pnpm test script and the CI validate job. The structure-section parsing added to parseAgentsDocExpectations is a meaningful new enforcement layer, and the regression tests give good baseline coverage.

Key observations:

• The second README.md comparison table (| **Agents** | 21 |) uses plain numbers without ✅ and is not matched by any tablePatterns regex — not the old one and not the new one with (?:\*\*)?. This table's counts will not be caught by the CI check if they drift in the future. The (?:\*\*)? addition appears targeted at this table but doesn't reach it because the ✅ requirement also needs to be relaxed or a separate pattern added.
• The skills structure-section mode ternary is a no-op (always resolves to 'minimum'), meaning a skills count lower than documented will not be caught via the structure block — this was flagged in a prior review thread.
• The test helper patches catalog.js source via a .*? regex, which is brittle to multi-line declarations — also previously flagged.

Confidence Score: 3/5

• Safe to merge for documentation and pipeline wiring; the logic gaps leave some count-drift paths undetected.
• The count updates are correct and confirmed by counting actual filesystem entries (21 agents, 52 commands, 102 skills). The CI wiring is straightforward. However, two gaps reduce confidence: (1) the second README comparison table is updated but not validated — a future drift there would pass CI silently; (2) the skills structure-section mode bug (always 'minimum', previously flagged) means a reduction in skill count goes undetected through that check path. These are functional blind spots in the enforcement mechanism that this PR introduces.
• scripts/ci/catalog.js (tablePatterns regex coverage) and tests/ci/validators.test.js (README fixture completeness and structure-section isolation)

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[catalog.js --text] --> B[buildCatalog\nCount agents, commands, skills\nfrom filesystem]
    A --> C[parseReadmeExpectations\nREADME.md]
    A --> D[parseAgentsDocExpectations\nAGENTS.md]

    C --> C1[Quick-start line\naccess to N agents M skills P commands\nmode: exact x3]
    C --> C2[First table rows with checkmark\nAgents N agents row\nmode: exact x3]
    C --> C3[Second bold table rows\nAgents N row - no checkmark\nNOT matched by any pattern]

    D --> D1[Summary line\nproviding N specialized agents M skills P commands\nmode: exact or minimum for skills]
    D --> D2[Structure block lines\nagents slash N subagents\nskills slash N workflow skills\ncommands slash N slash commands\nWARN: skills always minimum mode]

    B --> E[evaluateExpectations\n12 total checks]
    C1 --> E
    C2 --> E
    D1 --> E
    D2 --> E

    E --> F{Any check fails?}
    F -- yes --> G[Print mismatches\nprocess.exit 1]
    F -- no --> H[Documentation counts match\nprocess.exit 0]

_{Last reviewed commit: 3fe0389}