Skills, MCP & plugins
The full skill catalog by category, the MCP servers and how to register them, and the Claude Code plugins. The authoritative skill registry is .claude/rules/skill-router.md.
How skills are invoked
A skill is a folder under .claude/skills/<name>/SKILL.md with a description, trigger phrases, and a body. You invoke a skill two ways:
- Explicitly — type the slash command, e.g.
/osint Acme Corp.
- By intent — the always-on skill router matches your natural-language message to the right skill and announces it before running.
Skills marked 🔒 are explicit-invocation only (disable-model-invocation: true in frontmatter) — the model never auto-routes them; only a typed slash command fires them. This protects high-blast-radius actions (backups, audits, corporate publishing) from accidental routing.
Skill catalog
Intel
| Command | What it does |
|---|
/osint | Deep multi-source intelligence brief on a company, person, market, or tech — the foundation other intel skills consume. |
/osint-advanced 🔒 | Heavier OSINT variant; never auto-triggers. |
/competitor-intel | Competitive analysis of a named vendor versus your own product. |
/market-brief | Market / regional / sector intelligence and TAM sizing. |
/ceo-intel | Global / geopolitical situational-awareness brief. |
/intel-briefing-newsletter | An externally published intelligence newsletter. |
/yt-pulse | YouTube trend scanner + transcript deep-analysis brief. |
/x-pulse | X / Twitter pulse and account monitoring. |
/notebooklm | NotebookLM audio overviews / podcasts from sources. |
/docparse | Parse PDF / DOCX / PPTX / XLSX with spatial bounding boxes + visual citations. |
/deep-research-advance | One-shot deep web research with adversarial verification and a cited report. |
Communication
| Command | What it does |
|---|
/email-draft | New outbound email in the operator's voice (draft only — never auto-sends). |
/email-respond | Reply to an existing email thread (draft only). |
/email-intel | Daily inbox intelligence: scans mail, groups threads, proposes CRM / task actions. |
/follow-up | Personalized follow-up email after a meeting or event. |
/ceo-to-ceo | Peer CEO-to-CEO correspondence (draft only). |
/corporate-letter | Branded external letter, PDF + DOCX, locked template. |
/telegram | Live Telegram client — send / read / search / forward. |
/tribe-message, /tribe-monday | Internal all-hands message; the recurring Monday cadence variant. |
/translate | Translation between languages. |
Content & design
| Command | What it does |
|---|
/linkedin-post, /linkedin-series | A single LinkedIn post; a multi-post content series plan. |
/linkedin-archive | Files a published LinkedIn post + images into the datastore archive. |
/keynote-deck | Conference / event speaking deck. |
/image-prompt, /flux-image | A text-to-image prompt; actual image generation via Replicate. |
/design | Design system — illustrations, mockups, social graphics, logos, infographics. |
/pptx-generator | Editable PowerPoint decks and LinkedIn carousels. |
/marp | Markdown-to-slides pipeline (PDF + HTML). |
CRM
| Command | What it does |
|---|
/crm | Personal CRM — add, log, radar, find, update contacts, contact health. |
/viraid | Reads + triages a dedicated Telegram intake channel into actions. |
/google-contacts | Google People API contact lookup / add / edit (standalone, separate from the CRM). |
Strategy
| Command | What it does |
|---|
/deep-think | Structured sequential reasoning engine for hard, multi-variable calls. |
/council | Parallel second opinions from independent third-party models. |
/deal-strategy | Full deal strategy: positioning, pricing, objections, next steps. |
/investor-pitch, /investor-update | A fundraising pitch deck; an investor / board update. |
/proposal, /partnership-doc, /official-doc, /xpager, /rfp-response | The locked external-document templates (commercial proposal, MOU/LOI/term sheet, board resolution/notice, OnePager, RFP/tender). |
/data-room | Investor data room + due-diligence materials. |
/voss | Negotiation prep — tactical empathy, accusation audit, calibrated questions. |
/state-check | Operational-state / function-health diagnostic. |
/meeting-prep | Counterpart briefing (composes osint + voss + CRM + comms). |
/odin | Personal AI advisor with a persistent knowledge brain — see Memory & ODIN. |
/recall | Workspace-wide semantic recall over the local memory index. |
Operations
| Command | What it does |
|---|
/prime 🔒 | Session boot context load + parallel health checks. |
/dashboard, /next, /radar, /queue | Daily brief; "what's next" recommender; overdue-action detector; the terminal-native Action Queue. |
/weekly-review | End-of-week review. |
/dream 🔒, /memory-hygiene | Memory consolidation; memory rot scan. See Memory & ODIN. |
/backup 🔒, /sync 🔒 | Commit + push both repos; pull updates (git ff-only + corporate refresh). |
/push-updates 🔒, /publish-corporate 🔒, /promote-corporate 🔒, /rollback-corporate 🔒 | Multi-user fleet publishing pipeline (corporate content down to executives). |
/create-plan, /pre-impl 🔒, /implement | Plan → pre-implementation gate → execute. |
/evaluate, /scrutinize, /validate, /editorial-review | Artifact grading; principal-level audit; fact-check; document-structure editor. |
/workspace-deep-audit 🔒 | Full 8-section comprehensive workspace audit. |
/calibrate 🔒, /checkpoint 🔒 | End-of-session self-improvement; manual session-handoff save. |
/align 🔒, /devil 🔒, /burst 🔒 | Forced scope clarification; contrarian critique; N content variants. |
/sentinel 🔒, /bridge-health 🔒, /cold-sweep | Comms-monitor control; daemon fleet health; CRM "red debt" drain. |
/thread | Operational thread registry — open / log / close / find. |
/playwright, /ast-grep, /context7, /setup-browser-cookies | Headless browser; structural code search; library docs; cookie import. |
/skill-creator 🔒, /request-skill 🔒, /setup-wizard 🔒 | Author / request a skill; configure a fresh workspace. |
/event-debrief, /interview-prep, /zk, /mullvad | Post-event recap; hiring framework; Zettelkasten capture; Mullvad VPN control. |
MCP servers
The engine ships zero baked-in MCP config — all MCP servers are registered at user scope in the machine-local ~/.claude.json (gitignored, never in either repo). A fresh clone has no MCP servers until you re-register them with claude mcp add ... -s user.
| Server | Provides | Notes |
|---|
| playwright | Browser-automation MCP — navigate, click, type, snapshot, screenshot, evaluate JS, network inspection. | WSL-critical: launch through a native Linux Node absolute path, never npx — on WSL, npx/node on PATH resolve to the Windows install and the JSON-RPC stdio does not survive the WSL↔cmd.exe boundary. Needs the bundled Chromium + headless shell and apt libs libnss3 libnspr4 libasound2t64. |
| pencil | Editor for encrypted .pen design files — read / generate / validate designs. .pen files must only be accessed via these tools. | WSL interop: a native Windows .exe MCP server launched over WSL interop has working stdio and connects fine — unlike npx .cmd shims. New tools load on the next Claude Code restart. |
| Gmail / Google Calendar / Google Drive | Managed Claude-platform connectors — mail threads & drafts, calendar events, Drive files. | Bound to the operator's Google account; no engine config. Gmail is draft-only by design (aligns with the human-gated send rule). |
Plugins
Plugins are enabled via enabledPlugins in .claude/settings.json (travels in git) but installed per project path (machine-local, never synced — see Prerequisites §8). All come from the claude-plugins-official marketplace.
| Plugin | Provides |
|---|
| superpowers (v5.1.0) | 14 process skills: brainstorming, writing-plans, executing-plans, subagent-driven-development, using-git-worktrees, test-driven-development, systematic-debugging, verification-before-completion, code-review send/receive, finishing-a-development-branch, writing-skills, using-superpowers, dispatching-parallel-agents. Bootstraps at session start via its own hook. |
| skill-creator | Scaffolds / improves skills. A local /skill-creator also exists; bare-name lookup resolves to the local skill, the namespaced form forces the plugin. |
| claude-md-management | revise-claude-md and claude-md-improver — maintain CLAUDE.md. |
| frontend-design | Distinctive UI / visual-design guidance. |
| security-guidance | Security-guidance prompts (e.g. GitHub Actions injection warnings). |
Plugin skills are never auto-routed
Plugin-namespaced skills (e.g. superpowers:brainstorming) require an explicit slash command, the plugin's own metadata firing it, or another skill invoking them. Local skills win on bare-name collisions; use the namespaced form to force the plugin variant.