Propagate the real library size everywhere and unify the machine catalogs
behind one generator so they stop drifting independently.
- build_manifest.py is now single-source: one run regenerates cli/skills.json,
registry.json, .gemini/skills-index.json (was 203) AND the website catalog
skills.json (was 294). Per-skill entries gained license/category/subdomain/
tools_count.
- website rebuilt from the fresh catalog: 339 skill pages, 17 domain pages,
llms.txt, agents/commands, sitemap (counts now 339/17/774).
- prose updated: README, CLAUDE.md, CHANGELOG (4.10.0), mkdocs.yml, and the
docs/ install + architecture pages.
- gh repo description updated to 339/17/774.
Note: the legacy committed site doc-pages (landing, getting-started, guides)
use an older custom design the current generators no longer emit; their
embedded aggregates are left for a separate site-tooling decision (mkdocs
Material migration vs. keep custom).
Align every assistant-facing config with the real library (was 204 skills /
13 domains / 559 tools) and adopt newer platform features.
- configs: correct counts to 339/17/787 and drop dead dir refs
(engineering-team/, marketing-skill/) across .cursorrules, .windsurfrules,
.clinerules, .goosehints, AGENTS.md, GEMINI.md
- cursor: add modern .cursor/rules/*.mdc Project Rules (always-on overview,
SKILL.md authoring, scripts/*.py conventions); keep .cursorrules as fallback
- claude marketplace: rewrite marketplace.json to v3.0.0 with all 17 domains
as plugins and live counts; add the 9 missing per-domain plugin.json files
and fix the 8 stale ones
- manifest: build_manifest.py now also emits .gemini/skills-index.json (was
stuck at 203) from the same source as cli/skills.json + registry.json, so the
three machine indexes can't drift; regenerated all three
New cross-domain `workflow/` domain with two process meta-skills (orthogonal
to the 16 content domains, they multiply their value):
- workflow/handoff: package in-progress work into a self-contained context
doc so a teammate or fresh agent can continue; includes handoff_context.py
(captures git state) and a fill-in template.
- workflow/skill-router: match a vague/cross-domain intent against the whole
catalog and recommend the best-fit skill(s). route_skill.py ranks by token
+ multi-word phrase overlap over cli/skills.json — deterministic, no ML.
Both carry their own Clarify-First gate (Pattern 11).
Also, per a quality audit of the 297-skill gate rollout (verdict: 100% pass,
zero defects), gate two skills that were initially mis-skipped as advisory but
are in fact generative (consistent with already-gated peers):
- c-level-advisor/cfo-advisor (builds 3-year financial models)
- marketing/growth-marketer (designs A/B experiments, computes sample size)
- scripts/build_manifest.py: register the `workflow` domain
- regenerated cli/skills.json and registry.json (339 skills, 17 domains)
Move deep content from oversized SKILL.md bodies into load-on-demand
references/ files, so each skill's per-trigger cost drops sharply while
full knowledge stays available when a task needs it.
- 20 SKILL.md slimmed to lean maps (all now 1.0K-1.8K body tokens,
down from 5.6K-18.6K); combined per-trigger cost ~155K -> ~28K tokens
- 70 new references/ files; content relocated verbatim (zero info loss)
- add scripts/skill_token_audit.py: stdlib tool scoring every skill on
Tier-1/2/3 token cost and progressive-disclosure usage
- regenerate cli/skills.json + registry.json for bumped versions
Deepens the 54 PM skills from "documentation" to "tools you can actually
use today." Three coordinated additions:
1. WORKED EXAMPLES (~13,000 lines across 54 files)
Every PM skill now has examples/<scenario>.md with a realistic
company, the workflow applied, and the FULL generated artifact PMs
can copy. Recurring fictional cast (Acme Analytics, Wayfinder,
Northwind, Helix Platform, Pylon) makes scenarios cross-reference
naturally. Highlights: full 8-section PRD for "Shared Dashboards",
AI PRD with eval/guardrails/model-fallback, blameless RCA for a
47-min payment outage, weekly Yellow-status exec update, 8-interview
synthesis to opportunity tree, full Stripe CIRCLES answer.
2. LIVE DATA ADAPTERS (tools/adapters/)
Three stdlib-only scripts that pull from real APIs and emit JSON
in the shape PM Python tools expect. Removes the "first get the
data" friction.
- jira_to_json.py -> raw / status-update / cycle-time / dependency-map
- linear_to_json.py -> raw / status-update / cycle-time / dependency-map
- notion_to_json.py -> raw / prds / okrs / roadmap / feedback
Auth via env vars (JIRA_TOKEN, LINEAR_API_KEY, NOTION_TOKEN).
Pipe directly into the consuming PM tool:
jira_to_json.py --format cycle-time | flow_metrics.py --input -
3. MCP TOOLS (scripts/mcp_server.py)
15 PM skills wrapped as Claude Code MCP tools, callable from inside
any AI conversation without manually running Python. Each accepts
`format` and `input`. Names: pm_create_prd, pm_status_update,
pm_funnel_analyze, pm_flow_metrics, pm_dependency_map,
pm_feedback_triage, pm_nsm_tree, pm_refinement_score,
pm_interview_synthesize, pm_prioritize, pm_okr_validate,
pm_roadmap_transform, pm_pre_mortem, pm_release_notes,
pm_stakeholder_map. Smoke-tested end-to-end through JSON-RPC.
PM README updated with new sections for adapters and MCP. CHANGELOG
[4.5.0] entry added.
Wide push of 17 new PM skills + skeleton across discovery, execution, career,
and integrations. PM is now the deepest and most-visited domain in the library.
New skills:
- discovery: interview-synthesis (transcript -> opportunity solution tree)
- execution: prfaq, backlog-refinement, status-update-generator,
north-star-metric, roadmap-communication, beta-program, launch-playbook,
cycle-time-analyzer, dependency-map, story-splitting
- career (new sub-domain): pm-interview-prep, pm-career-ladder,
pm-onboarding, pm-1on1s
- integrations: linear-expert (Linear GraphQL + Jira migration patterns),
notion-pm (Notion DB design for PRDs/OKRs/Roadmap/Decisions)
Foundation:
- SHARED_OUTPUT_SCHEMA.md defines --format json|markdown|mermaid|confluence|
notion|linear convention all PM Python tools follow
- project-management/examples/feature-end-to-end.md worked end-to-end scenario
- Role-based Quick Start in PM README (APM -> CPO, 7 role entry points)
- 4 new bundles in bundles.json: pm-starter-pack, pm-scrum-master,
pm-discovery-coach, pm-career-track
Fixes:
- scripts/build_manifest.py now recurses one level into container folders
(discovery/, execution/, career/) so nested skills appear in the registry.
Previously the registry undercounted PM by ~30 nested skills.
- .claude/settings.json hook commands use \$CLAUDE_PROJECT_DIR instead of
relative paths so they work regardless of CWD.
Bolded PM as the most-visited domain in README.md, CLAUDE.md, docs/SKILLS.md,
and project-management/{README,CLAUDE}.md. Reconciled skill count across all
index files: 282 total, 42 PM.
Adds registry.json at repo root as the public-facing index third-party
indexers can consume. Slimmer schema than cli/skills.json (no file
lists or sizes — those stay in the internal CLI manifest).
Also fixes a latent bug in scripts/build_manifest.py: the DOMAINS list
was missing documents, personal-productivity, and vertical-advisors,
silently dropping 21 skills from cli/skills.json. Total is now 250
skills across 15 domains.
Both manifests are now generated from the same walk in one pass, so
they stay in sync.
5 curated plugin bundles ready for submission to
anthropics/claude-plugins-official as external plugins.
Bundles:
- engineering-skills (8 flagship): fullstack, devops, security, docker,
terraform, mcp-server-builder, rag-architect, code-reviewer
- compliance-skills (9 flagship): ISO 27001, SOC 2, GDPR, EU AI Act,
NIS2, DORA, ISO 13485, MDR, FDA
- c-level-skills (7 flagship): CEO/CTO/CFO/CISO/CMO advisors plus
board-meeting and board-deck-builder
- marketing-skills (6 flagship): content-creator, seo-audit, demand
acquisition, campaign-analytics, landing-page-generator, x-twitter
- product-skills (6 flagship): product-manager-toolkit, strategist,
agile-product-owner, product-designer, ux-researcher, design-system
Each bundle has .claude-plugin/plugin.json per Anthropic's external
plugin schema, a README, and symlinks into the canonical skill
directories. Symlinks keep a single source of truth for skill content;
scripts/export_bundle.py dereferences them for self-contained exports
when submitting PRs to the Anthropic registry.
Total: 44 KB repo addition, 36 symlinks.
npx claude-skills add <skill> installs any skill from the library into
the detected AI assistant in one command. Auto-detects Claude Code,
Cursor, Codex, Gemini CLI, Copilot, Windsurf, Cline, Aider, Goose via
marker files. Pure Node.js (18+), no Python required for end users.
Commands: list, search, info, add, create, update, remove.
Adds:
- scripts/build_manifest.py: stdlib-only Python that walks domain
folders and emits cli/skills.json (229 skills, 12 domains indexed)
- cli/: npm package with commander + kleur, 10 passing tests
- .github/workflows/build-manifest.yml: regenerates manifest on
SKILL.md changes, auto-commits
- .github/workflows/publish-cli.yml: publishes to npm on cli-v* tag,
supports dry_run via workflow_dispatch
The prompt now explicitly instructs the AI to:
- Provide the complete skill definition in a ready-to-copy code block
- Never say "paste from above" or "scroll up"
- Auto-detect the platform (Claude.ai vs ChatGPT vs other)
- Give exact step-by-step instructions with the actual text to paste
- Include a real test prompt from the skill's workflows
Every skill page now has two copy-paste options:
- **Try in Chat** (Quick) -- condensed prompt for instant one-off use
in any AI chat. No setup needed.
- **Add to My AI** (Full Skill) -- complete SKILL.md content with
step-by-step instructions for the AI to help create a persistent
Claude Project or Custom GPT. The AI walks the user through setup
on their specific platform.
Both prompts are auto-generated from each SKILL.md. The "Add to My AI"
prompt includes the full skill definition so the created project has
complete expertise, workflows, templates, and frameworks.
Every skill page now has a "Quick Start" tab (default) with a one-click
"Copy Prompt" button. Users can paste the generated prompt directly into
claude.ai, ChatGPT, or any AI chat to get that skill's expertise
instantly -- no repo setup, no CLI, no configuration needed.
- Auto-generates condensed prompts from each SKILL.md
- Prompt includes key capabilities, frameworks, and guidance
- Copy button with visual feedback (turns green on copy)
- Scrollable preview of the full prompt
- "Quick Start" is the default tab on all 244 skill pages
Amin Borghei