claude-toolkit-catalog

Concierge index for the Claude Code toolkit on this machine — router-first rule, by-intent map, workflow recipes, and rich family descriptions for the Mundi Princeps workflows. Curated head is hand-maintained; auto-generated inventory lives below the marker and is refreshed by ~/.claude/scripts/build-toolkit-scout.sh.

META-DELIVERABLE — user drops a URL (article/YouTube/tweet/repo/tool), agent autonomously classifies + extracts + decides which of {new skill, agent, command, KB wiki entry, MEMORY append, CLAUDE.md update, registry refresh, autoresearch propagation} apply, writes additively, commits once, rsyncs to Achilles. Replaces monthly R&D marathons.

Generate board-discussion materials end-to-end from a data room — ingest (Drive/SharePoint) → Gemini-powered summarization → pitch-deck generation → Q&A prep → artifact bundle. Use when prepping for a board meeting, IC discussion, or investor update where you have a raw data room and need polished materials.

Reusable orchestrator that enriches a counterparty (company, investor, LP, advisor, or person) via entity resolution → parallel fan-out across PitchBook + CapIQ + A-Leads + Orbis + Perplexity → merged dossier with provenance. Use when you need a complete profile of a counterparty before outreach, IC discussion, or deal evaluation. Invoked by /mundi:counterparty-enrich and chained by /mundi:origination-run.

End-to-end deal-origination pipeline orchestrator — converts a thesis / sector / theme into screened targets with IC-ready one-pagers. Chain: thesis lock → PB screener → signal enrichment → scoring → counterparty-enrich fan-out → one-pager generation → deal-tracker append. Use when running origination sprints, building a pipeline for a specific theme, or refreshing an existing pipeline with new entrants.

End-to-end intent-signal pipeline orchestrator — scrape/classify/score intent signals from web + PB + CapIQ, dedupe, and route to downstream (outreach, CRM, scoring). Use when monitoring a theme for new signals, ingesting a fresh signal batch, or wiring a signal → outreach handoff. Paired with /mundi:intent-signal-run.

Investor-outreach pipeline — PB investor screener → Apollo/A-Leads contact enrichment → Gemini message personalization → HeyReach campaign push → Supabase CRM log. Use when building an investor outreach campaign from a thesis or list of target firms, refreshing contacts on existing campaigns, or wiring signal-to-outreach handoff.

Cross-model consensus — spawn parallel calls to Claude + Gemini + GPT on the same question, aggregate via vote or merge, return a decision with divergence flagged. Use when a high-stakes decision needs genuine model-level diversity (not just framing variation within one provider). Distinct from /mundi:consensus which is Claude-only framing variation.

Adversarial cross-model debate — Claude vs Gemini vs GPT argue opposing sides of a decision, structured multi-round, escalating specificity. Use when genuine adversarial diversity is needed (not Claude-vs-Claude debate which has correlated priors). Distinct from /mundi:debate which is Claude-only chat.json.

Task-type → best-LLM-per-task router. Classifies incoming task (code, long-context research, creative, math, vision, agentic) then dispatches to the right provider (Claude for code, Gemini for long-context, GPT for creative, etc.) per published benchmark + cost profile. Use when composing an orchestrator that needs to call the "right" LLM automatically — or when budget optimization matters across many calls.

Centralized proactive auth/session refresh across all Mundi auth-gated providers (PitchBook, CapIQ, Supabase, Unipile, Cognito/OKTA-SNL). Monitor token TTL, refresh before 401s, codify the shared headful-Chrome-under-Xvfb + cookie-dump + systemctl-restart + circuit-breaker-close pattern. Use when setting up new auth-gated scrapers, debugging 401 loops, or scheduling preventive refresh crons.

Mundi-specific integration layer for tob-fp-check — wires false-positive checking into Kadenwood / PitchBook / CapIQ / signal-validation review workflows so "fixed" findings don't retrigger across sessions. Use when running /mundi:security-audit, /mundi:refusal-test, validating intent-signal hits, or reviewing bug reports where recurrence needs suppression.

Audit MCP intent → tool coverage — cross-reference MCP tool catalogs against intent buckets to find gaps (intents with zero tools, tools unclaimed by any intent, consuming-app data needs not wired). Applies to PitchBook, CapIQ, and any future MCPs. Use when onboarding a new MCP, adding new intents, or investigating "why doesn't my query route to anything."

Runbook for wiring the lead-enricher's SupabaseWriter (112 lines, batch-of-100 inserts to enriched_contacts) into the enrichment waterfall as the terminal step, plus handling provider-swap away from PeopleDataLabs (100/month free cap blocks production). Use when shipping lead-enricher to production, scaling beyond free tier, or adding a new enrichment provider to the waterfall.

Install + configure gitleaks (or trufflehog) as a mandatory pre-commit hook across Mundi Princeps repos, with Mundi-specific regex bundle (AIza, sk-ant-, gho_, Bearer, Twilio AC SID, "cookie" JSON field). Backfills patterns from ~/.claude/scripts/phase3/check-secrets.sh. Use when onboarding a new repo, a secret leak just happened, or pre-commit isn't yet wired.

Provider-aware SSH deploy skill — 7-step template (verify-dir → git-ff-pull → pip-install → systemctl-restart → wait-active → tool-count-sanity → CLI-entrypoint-check) generalized across PB, CapIQ, KadenVerify, investor-outbound, signal-pipeline. Use when deploying code to Achilles / mundi-ralph, setting up deploy for a new app, or debugging a stalled deploy.

One-page quick-maths for agency / AI-automation economics — retainer pricing, margin targets, break-even per hire, token cost per lead, CAC vs LTV. Saraev's framing that most agency operators fail on the math, not the service. Use when pricing a retainer, evaluating a client's profitability, deciding whether to hire, or estimating token cost per outbound.

Generate blog post hero images and marketing visuals using Google Gemini's image generation model. Use for content marketing, social media assets, or presentation graphics.

Pointer to hesreallyhim/awesome-claude-code — the canonical community curation of Claude Code skills, agents, hooks, slash-commands, plugins, statuslines, and MCP servers (39,800+ stars, actively maintained). Use to discover community solutions before building new Claude Code tooling from scratch.

Pointer to Andrej Karpathy's "Neural Networks Zero to Hero" — the canonical lecture series that builds micrograd (autograd engine) and a working MLP from scratch in pure Python + math. Use when onboarding to ML fundamentals, before fine-tuning work, or when framework abstractions hide a bug you need to debug at the gradient level.

18 AI personas (Aristotle, Socrates, Sun Tzu, Feynman, Kahneman, Karpathy, Sutskever, Taleb, etc.) deliberate hard decisions via structured multi-round debate across Claude/OpenAI/Gemini/Ollama. Use when making a high-stakes call where genuine multi-model disagreement beats single-model consensus.

Pre-landing PR review. Analyzes diff against the base branch for SQL safety, LLM trust boundary violations, conditional side effects, and other structural issues. Use when asked to "review this PR", "code review", "pre-landing review", or "check my diff". Proactively suggest when the user is about to merge or land code changes.

Use when reviewing code changes, PRs, architecture, security, plans, prompts, or performance. 7 modes: code, architecture, security, plan, prompt, performance, full. Ensemble consensus, auto-fix, escalation, scope drift detection.

Universal SOP for building and shipping software with agentic AI. Clarity-first, build fast, iterate from usage. 3 tracks, parallel-by-default orchestration, review loops in every phase.