[SeanShaffer88]

What this does

Grows the catalog from 2,202 → 2,937 entries (2,931 distinct URLs, 101 categories, 13 domains) and restructures the repo so an AI can be pointed at it and reach the right few files fast — plus so future additions are mechanical rather than manual.

1. Structure for AI consumption

Change	Why
index/catalog-stats.json (new, generated)	One canonical source for every count, rollup, and file size. Human docs quote it instead of hard-coding numbers.
scripts/catalog_stats.py (new)	--write regenerates the stats file and rewrites the stats block in README.md, AI_INDEX.md, llms.txt; --check fails on drift.
validate.py now gates counts	Fails if catalog-stats.json is stale or a generated stats block drifts. Count errors can no longer ship silently.
README.md rewritten	Reading-path table ("what you want → file → size"), corrected numbers, all six expansion catalogs, full regeneration order, evidence/provenance section.
AI_INDEX.md rewritten	One page: read order, all 13 domains with real counts, and a table of every machine-readable entry point.
AGENTS.md, index/README.md, CONTRIBUTING.md, generated nav	Consistent at 13 domains / 101 categories / ~35% compact ratio; the contribution guide now documents the discovery sweep and dedupe rules.

Entry points stay stable and small-first: AI_INDEX.md (3 KB) → index/by-category/<id>.json (1–30 KB) → index/resources.compact.json (1.2 MB, 35% of canonical) → resources.json (3.3 MB) → llms-full.txt (1.8 MB, one-shot).

2. Two dated expansions

expansion-2026-11-10 — 82 entries. Gap analysis of project entry points: SemVer, Keep a Changelog, Conventional Commits, Choose a License, SPDX, REUSE, ADRs, Diátaxis, Contributor Covenant, Twelve-Factor, C4 (new category Software engineering foundations and standards), plus Cypress/Selenium/WebdriverIO/Jest/Robot Framework, HTTP clients and mock servers, editors, package managers, and observability/BI tools.

expansion-2026-11-11 — 653 entries. The discovery channels the request named:

• 160 GitHub topics via gh api search/repositories (ai-agents, llm, llms, ai-prompt, ai-prompts, mcp, mcp-server, reverse-engineering, vibe-coding, ai-skills, skills, copilot, claude, codex, model-context-protocol, agent-skills, llm-evaluation, guardrails, llmops, plus 140 more), ranked by stars with a 100-star floor.
• gittrend.io/hidden-gems — today/week/month boards.
• awesomehub.js.org — topic indexes (cli-apps, selfhosted, go, python, rust, github-actions, …).

Candidates were deduplicated against the catalog by GitHub owner/repo identity, normalized URL, and repository name, then hand-reviewed per category; archived, forked, off-topic, and unlicensed-where-material repos were dropped (the few dual-use or copyleft entries carry the trade-off inline).

Families that previously had zero entries and now have coverage: AI agent skills/memory/MCP clients/guardrails, reverse-engineering tooling, ORMs and database servers, message queues and workflow engines, OpenTelemetry SDKs and log pipelines, i18n libraries, OCR/document extraction, rich-text and note-taking engines, editor/plugin ecosystems, .NET application blocks, Android libraries.

3. Evidence and limits

Per-entry evidence in docs/expansion-2026-11-11-checks.json: GitHub repository ID, detected license (SPDX), description, stars at check time, archived/fork flags, last push, and the catalog topics that surfaced it. Every new record carries source, checked_on, review scope, and verification_level.

This is metadata and reachability evidence only — no README review, release/maintenance audit, licensing opinion, security assessment, compatibility test, or adoption recommendation. No package was installed or executed.

Verification

python scripts/validate.py              → Validated 2937 records, 101 categories, stable IDs/facets/relationships,
                                          generated indexes, CSV/JSON agreement, and local Markdown links.
python scripts/catalog_stats.py --check → current
python -m unittest discover -s tests    → Ran 7 tests ... OK

The 2026-11-10 batch had been sitting uncommitted; this PR includes it. Existing records keep their IDs, facets, and related-link symmetry (verified against git show HEAD:resources.json).

Follow-ups (not in this PR)

• The Dev Resource Explorer snapshot still imports commit 9f4d7c8a (2,202 entries); it needs a re-import to show the new batch.
• Further sweeps of non-English ecosystems and awesome-* lists are possible with the recipe now documented in CONTRIBUTING.md.