Skip to content

HANDOFF.md

Latest commit

 

History

History
90 lines (75 loc) · 3.96 KB

HANDOFF.md

File metadata and controls

90 lines (75 loc) · 3.96 KB

HANDOFF -- OpenClaw Foundry

Session 2026-03-26. Job Packs v2 DONE + Skill Curation v4 DONE + Pipeline Automation DONE.

Commits This Session (10)

# Hash Summary
1 c344510 Job Packs v2: 10 packs, 4 lines, 50 static files
2 4d6d849 Full closure: API v5, D1 migrated, combos removed, CI prebuild
3 7fa4846 Skill Curation P0: dedup + bell-curve ratings + 2000 icons
4 d5fb846 Skill Curation P1-5: 2000 tags + 400 taglines + 0% Other
5 9e7c11b Pipeline automation: CI curation + incremental LLM + git commit
6 098e32e Timestamp cleanup
7 ad29744 CI timeout: 25 → 45 min for LLM enrichment
8 45e29da Carry-forward fix: restore tags from git HEAD, avoid LLM full retag
9 be0ddf8 seed-db: add npm ci for pinned wrangler
10 (CI) sync-data auto-commit: curated skills data

What Was Built

1. Job Packs v2 (COMPLETE)

  • 10 packs across 4 function lines (Engineering 5 / Data&AI 2 / Product 2 / Business 1)
  • 4-layer inheritance: L0 finance-tax (all inherit) → L1 line → L2 role → L3 project (optional)
  • 2-table D1 model: pack_layers (15) + config_packs (10)
  • Question tree UX: 2-step → 1 recommendation + Tab browse
  • E2E verified: VPS curl install.sh → 4 files installed

2. Skill Curation v4 (COMPLETE)

  • Phase 1 (Dedup): 89 generic names renamed, 0 same-author duplicates
  • Phase 2 (Ratings): Percentile bell-curve S:100 A:300 B:800 C:600 D:200
  • Phase 3 (Tags): 2000/2000 tagged (tech_stack + scenario + platform), Gemini Flash
  • Phase 4 (Editorial): 400 S/A taglines + 2000 Material Symbol icons
  • Phase 5 (Reclassify): 61 "Other" → 0 (0.0%)

3. Pipeline Automation (COMPLETE)

  • CI sync-data: scrape → merge → prebuild → curate → LLM enrich → seed SQL → git commit
  • Carry-forward: Phase 0 restores tags/taglines from git HEAD, LLM only processes new skills
  • Incremental LLM: tag-skills-llm.mjs skips already-tagged skills
  • Graceful degradation: no API key = skip LLM (doesn't block pipeline)
  • GOOGLE_API_KEY secret added to GitHub Actions

Known Issue: seed-db D1 Permission

Status: CF API Token (CLOUDFLARE_API_TOKEN) lacks D1 query permission. Impact: LOW — frontend uses static JSON, Worker API uses manual seed data (37310 skills). Fix: Create new API token in CF Dashboard with D1 Edit permission, then:

gh secret set CLOUDFLARE_API_TOKEN
# paste new token value

Key Files

File Purpose
scripts/curate-skills.mjs Dedup + rating + icons + carry-forward (Phase 0-2, 4)
scripts/tag-skills-llm.mjs LLM tags + taglines + reclassify (Phase 3-5, incremental)
scripts/generate-packs.mjs Job Pack static file generator (50 files)
scripts/generate-seed-sql.mjs Curated data → D1 seed SQL (with tags/tagline columns)
.github/workflows/deploy.yml Full pipeline DAG documented at top
doc/.../SKILL_CURATION_V4_PLAN.md Curation upgrade plan (6 phases)
doc/.../SKILL_CURATION_V4_PLAN-zh.html McKinsey Blue HTML version
doc/.../JOB_PACKS_ARCHITECTURE-zh.html Job Packs architecture doc

Pipeline DAG

Daily sync (cron 06:00 UTC):
  scrape-clawhub-api.mjs → data/clawhub-skills-raw.json
  scrape-mcp-registry.mjs → data/mcp-registry-servers.json
  merge-all-sources.mjs → data/unified-index.json
  prebuild-static.mjs → web/public/data/skills.json (raw)
  curate-skills.mjs → skills.json (carry-forward + dedup + rating + icons)
  tag-skills-llm.mjs → skills.json (+ tags + taglines + reclassify) [incremental]
  generate-seed-sql.mjs → worker/src/seed.sql (from curated data)
  git commit → persist curated data

Push to main:
  generate-packs.mjs (npm prebuild) → web/public/packs/ (50 files)
  next build → web/out/ (static export)
  wrangler pages deploy → CF Pages
  wrangler deploy → CF Worker

Git State

  • Branch: main
  • Latest: be0ddf8
  • CI: sync-data/deploy-frontend/deploy-worker GREEN, seed-db KNOWN ISSUE
  • Production: openclaw-foundry.pages.dev LIVE