Reusable Claude Code skills library - trading signals, sales automation, RunPod deployment
Overview
99 production-ready Claude Code skills (3 stable, 96 active), 1 archived, plus 1 Cowork plugin carrying 10 SDR skills. AI-native engineering, GTM, sales automation, BDR pipeline automation, and trading. Counts come from ./scripts/doc-counts.sh — run it before quoting a number anywhere. Then Settings → Skills → upload the zip. claude.ai never writes back; the repo is the source of truth. Skills are . scripts/deploy.sh symlinks every active/ and stable/ skill into ~/.claude/skills/, so any repo on this machine uses them by name — no per-project symlink, no CLAUDE.md path reference, no copied zip. Editing a deployed skill edits this repo (it is the same directory), and check-drift.sh fails if anything in ~/.claude/skills/ is a copy instead of a link. Zips exist only for the claude.ai / Cowork upload path. 41 BDR skills carry customer_facing: true in config.json and a block that points at stable/epiphan-house-rules-skill.
README
Skills Library
99 production-ready Claude Code skills (3 stable, 96 active), 1 archived, plus 1 Cowork plugin carrying 10 SDR skills. AI-native engineering, GTM, sales automation, BDR pipeline automation, and trading.
Counts come from
./scripts/doc-counts.sh— run it before quoting a number anywhere.
Quick Start
Claude Code CLI (the only install you need)
./scripts/deploy.sh # symlink every active/ + stable/ skill into ~/.claude/skills/
./scripts/check-drift.sh # confirm ~/.claude/skills is exactly this repo
claude.ai / Claude Desktop / Cowork (push-only mirror)
./scripts/rebuild-zips.sh # one zip per skill in dist/ (gitignored)
Then Settings → Skills → upload the zip. claude.ai never writes back; the repo is the source of truth.
How skills reach other projects
Skills are global. scripts/deploy.sh symlinks every active/ and stable/ skill into ~/.claude/skills/, so any repo on this machine uses them by name — no per-project symlink, no CLAUDE.md path reference, no copied zip. Editing a deployed skill edits this repo (it is the same directory), and check-drift.sh fails if anything in ~/.claude/skills/ is a copy instead of a link. Zips exist only for the claude.ai / Cowork upload path.
House rules (customer-facing skills)
41 BDR skills carry customer_facing: true in config.json and a `` block that points at stable/epiphan-house-rules-skill. That one skill holds the four gates every prospect-facing output passes, in order:
| Gate | Tool | Rule |
|---|---|---|
| Q qualify | qualify_lead |
Run right after the candidate pull, before tiering, enrichment, or drafts |
| S spec | search_product_catalog or Epiphan Knowledge search |
Before any product claim; never from memory |
| B brand | check_my_copy |
On every customer-facing string |
| D draft | Gmail create_draft |
Outbound text becomes a draft, never a send |
It also owns the lead-suppression rule, the self-healing contract (retry → degrade → alert → halt + run log), and the single declaration of identifiers (HubSpot portal, AE owner IDs, SDR seats, canonical emails). Test T18 fails any customer-facing skill missing the block.
Skills (99)
Stable (3) — always loaded
| Skill | What It Does |
|---|---|
| workflow-enforcer | Automatic agent routing + discipline enforcement |
| project-context | Session continuity across conversations |
| epiphan-house-rules | Shared policy: four gates (Q/S/B/D), suppression, self-healing, identifiers |
Core (3)
| Skill | What It Does |
|---|---|
| workflow-orchestrator | Dual-team workflow + cost tracking + agent orchestration |
| cost-metering | API cost tracking, budget alerts, model routing |
| portfolio-artifact | Engineering metrics capture, report templates |
Dev Tools (15)
| Skill | What It Does |
|---|---|
| extension-authoring | Author skills, hooks, slash commands, subagents |
| debug-like-expert | Systematic debugging with hypothesis testing |
| planning-prompts | Project planning and meta-prompt creation |
| worktree-manager | Git worktree automation for parallel dev |
| agent-teams | Orchestrate parallel Claude Code sessions |
| subagent-teams | In-session Task tool subagent orchestration |
| agent-capability-matrix | Task→agent mapping, 70+ agents cataloged |
| git-workflow | Conventional commits, PR templates, merge strategies |
| testing | TDD, test pyramid, mocking, coverage strategies |
| api-design | REST/GraphQL API design patterns |
| security | Auth, secrets, OWASP, RLS, security audit |
| api-testing | API testing with Postman and Bruno |
| docker-compose | Local dev environments with Docker Compose |
| frontend-ui | Enterprise SaaS frontend: Tailwind v4, shadcn/ui, Next.js |
| heal-skill | Auto-diagnose and repair broken skills |
Infrastructure (8)
| Skill | What It Does |
|---|---|
| langgraph-agents | Multi-agent LangGraph systems |
| groq-inference | Ultra-fast GROQ API: chat, vision, audio, tools |
| openrouter | Chinese LLMs (DeepSeek, Qwen) via OpenRouter |
| voice-ai | Production voice agents: Deepgram + Cartesia + Twilio |
| unsloth-training | Fine-tune LLMs with GRPO/SFT |
| runpod-deployment | GPU serverless deployment patterns |
| supabase-sql | Clean SQL migrations for Supabase |
| stripe-stack | Stripe payments + webhooks for Next.js |
Business — GTM & Sales (10)
| Skill | What It Does |
|---|---|
| gtm-pricing | GTM strategy, pricing, opportunity evaluation |
| research | Market and technical research |
| sales-revenue | Sales outreach, discovery, RevOps |
| crm-integration | Close CRM, HubSpot, Salesforce patterns |
| hubspot-revops | HubSpot SQL analytics, lead scoring, pipeline forecasting |
| content-marketing | B2B content strategy |
| data-analysis | Executive data analysis and dashboards |
| miro | Miro boards for strategy, architecture, sprints |
| prospect-research-to-cadence | Apollo research → outreach → sequence automation |
| phone-verification-waterfall | Golden Rules filtering + Clay waterfall enrichment |
Business — Sales Workflow Automation (7)
| Skill | What It Does |
|---|---|
| meddic-call-prep-auto | Auto-generate MEDDIC call prep from CRM + Apollo data |
| deal-momentum-analyzer | Score deal velocity, predict close vs stall |
| portfolio-deal-linker | Link engineering output to revenue impact |
| trading-alert-scheduler | Daily pre-market digest with regime detection |
| ae-handoff-brief | BDR→AE handoff briefs with MEDDIC summary |
| call-recording-analyzer | Score call recordings against discovery rubric |
| dead-deal-recovery | Resurrect closed-lost deals with re-engagement plays |
BDR Automation — Daily Pipeline (16)
| Skill | What It Does |
|---|---|
| prospect-enrich | Mon-Fri 6:00 AM — phoneless contact enrichment (DEMO REQUEST first) |
| prospect-refresh | Monday 6:30 AM — net-new ICP prospect search + Gmail drafts |
| sequence-load | Monday 7:15 AM — auto-load into Apollo sequences |
| callable-lead-count | M-F 7:25 AM — ATL/BTL inventory + runway metrics |
| morning-brief | M-F 7:30 AM — calendar + dials + deals + drafts briefing |
| nooks-autopilot | M-F 8:05 AM — autonomous lead hunt → Nooks sequence enrollment (shadow-first) |
| vertical-list-builder | Daily callable lists by vertical × territory (USA/Canada/UK/LatAm); owns the SDR seat roster (references/seats.md) |
| sdr-dial-lists | M-F 7:30 AM — per-SDR ranked dial queues, roster resolved from vertical-list-builder |
| sdr-call-coaching | Friday 11 AM — score SDR calls, per-rep coaching cards |
| he-dial-queue | Higher-ed callable dial inventory, tiered by ATL/BTL |
| inbound-speed-to-lead | Sub-5-minute first touch on HIGH-intent inbounds: qualify, draft, route to AE, alert, log |
| install-base-expansion-engine | Monday 8 AM — land-and-expand triggers across owned Pearl + cloud accounts |
| linkedin-social-selling | Per-account LinkedIn warming kits + weekly thought-leadership drafts tied to the dial list |
| paid-marketing-lead-enrichment-bdr-list | Weekly paid-social/paid-search lead triage + Apollo/Clay swarm enrichment → dial list |
| weekly-kpi-report | Friday 5:30 PM — per-rep SDR scorecards (dials, connects, meetings, pipeline) |
| batch-send-drafts | Review staged Gmail drafts, tier via qualify_lead, one-click send links (never sends) |
Sales Intelligence (17)
| Skill | What It Does |
|---|---|
| champion-identifier | Find internal champions at target accounts |
| cold-email-sequence-generator | 7-14 email sequences with A/B subject lines |
| contact-hunter | Multi-source contact info extraction |
| email-template-generator | Professional email templates by scenario |
| inbound-lead-qualifier | Score inbound leads by ICP + intent + urgency |
| intent-signal-aggregator | Monitor buyer intent signals (jobs, funding, tech changes) |
| linkedin-sales-navigator-alt | LinkedIn prospect list building + tracking |
| lookalike-customer-finder | Find 100+ companies matching best customers |
| meeting-intelligence-system | Extract decisions/actions from meeting transcripts |
| personalization-at-scale | Personalized first lines for 100s of prospects |
| pipeline-health-analyzer | Forecast accuracy + stall prediction |
| sales-methodology-implementer | MEDDIC/BANT/Sandler/Challenger/SPIN frameworks |
| social-selling-content-generator | 30+ LinkedIn posts for prospect attraction |
| competitive-displacement-finder | Competitor EOL (Extron SMP, Matrox) + rival footprints → ranked swap-pitch targets |
| pipeline-yield-analytics | Closed-loop funnel conversion by source, vertical, buyer power, rep; Leak Map |
| epiphan-weekly-sales-tracker | One-call weekly_brief digest appended to the FY2026 Weekly Sales Tracker sheet |
| customer-success-meetings | Pre-call Slack briefs + post-call recaps for CS handoffs (#team-csr) |
Sales Enablement — Epiphan GTM (11)
| Skill | What It Does |
|---|---|
| epiphan-ai-mcp-guide | Day-one reference for Epiphan AI + CRM/Clari MCP toolset |
| epiphan-call-playbook | Canonical cold/discovery call playbook, verified spec bank |
| business-pulse | Live firm-wide sales pulse with coaching takeaways |
| close-plan-generator | MEDDIC close plans with confidence score + mutual action plan |
| demo-execution-playbook | Vertical demo flows, objection matrix, post-demo scorecard |
| cost-discovery-coach | Discovery questions mapped to cost-calculator inputs |
| post-demo-automation | Demo recap emails, CRM updates, next-step automation |
| greenfield-pearl-tracker | Track greenfield Pearl cohort deals + buying signals |
| orlob-discovery-framework | Chris Orlob 5-step discovery applied to Epiphan verticals; 0-3 rubric used by sdr-call-coaching |
| sdr-email-sms-playbook | 5-touch email/SMS cadence by vertical, staged as Gmail drafts; phone companion is epiphan-call-playbook |
Marketing & Brand (3)
| Skill | What It Does |
|---|---|
| epiphan-brand-assets | Brand gate: wraps the Epiphan Brand MCP (asset kit, writing style, check_my_copy) before and after drafting |
| agent-ready-av-campaign | Campaign rulebook for the agent-ready AV content cluster: keyword map, AEO/GEO structure, cleared stats |
| live-event-survey-analysis | Test messaging claims against the 2026 Live Event Survey data (branching + multi-select aware) |
Business — Trading & Brokerage (2)
| Skill | What It Does |
|---|---|
| trading-signals | 5 TA methodologies, 25+ options strategies, regime detection |
| ibkr-api | Interactive Brokers API: portfolio, trades, multi-account |
Strategy (5)
| Skill | What It Does |
|---|---|
| business-model-canvas | 9-block business model design |
| blue-ocean-strategy | Blue ocean market differentiation |
| jobs-to-be-done | JTBD + ODI analysis for sales discovery and strategy |
| challenger-sale | Teach-Tailor-Take Control framework for B2B sales |
| never-split-the-difference | FBI negotiation: tactical empathy, calibrated questions, Ackerman model |
Archived (1)
| Skill | Status |
|---|---|
| gtm-brain | Archived 2026-07-18 to archived/gtm-brain-skill (Neo4j relationship graph; not deployed, not counted) |
Naming exceptions (no -skill suffix, live-automation names): nooks-autopilot, sdr-dial-lists, sdr-call-coaching, epiphan-call-playbook, vertical-list-builder.
Epiphan SDR Toolkit (Cowork plugin)
plugins/epiphan-sdr-toolkit/ is a Cowork plugin (v0.10.0) versioned here and installed in Cowork / claude.ai as epiphan-sdr-toolkit:*. Its 10 skills are plugin skills, not library skills, and are not counted in the 99:
| Plugin skill | What It Does |
|---|---|
| nooks-dial-prep | Battlecard per prospect on the rep’s Nooks dial list (account intel, opener, Orlob questions, verified specs) |
| live-objection-battlecard | Mid-call rebuttal + Challenger reframe + Voss question for an objection or competitor mention |
| nooks-call-wrap | Post-call MEDDIC summary, correct Nooks disposition, follow-up draft, AE handoff |
| sdr-ramp-cert | Product/competitive quiz, role-play, scored certification for a new rep |
| nooks-call-timing | Who to dial now by prospect local time; learns best windows from connect-rate-by-hour |
| nooks-followup-stager | Call transcript → human follow-up staged as a Gmail draft (single or batch) |
| demo-confirm-and-handoff | Route a booked demo to the right AE, draft confirmation, book the calendar event, stage the reminder |
| no-connect-multitouch | Voicemail script + Gmail draft + LinkedIn note on one hook for a no-answer dial |
| team-coaching-rollup | Per-rep scoreboard with coaching flags for Tim and Victor |
| nooks-segment-sync | Manual segment-sheet lookup and confirm-first Nooks sync/enrollment |
Install is one click when the plugin appears in chat. Required connectors: Nooks, Epiphan AI, HubSpot, Epiphan Brand, Gmail, Google Calendar. See plugins/epiphan-sdr-toolkit/README.md.
Project Structure
skills/
├── active/ # 96 trigger-activated skills
├── stable/ # 3 always-loaded skills (workflow-enforcer, project-context, epiphan-house-rules)
├── archived/ # 1 archived skill (gtm-brain) — not deployed, not counted
├── plugins/
│ └── epiphan-sdr-toolkit/ # Cowork plugin v0.10.0, 10 SDR skills
├── dist/ # One zip per skill for claude.ai upload (gitignored; rebuild-zips.sh)
├── scripts/
│ ├── deploy.sh # Symlink active/ + stable/ into ~/.claude/skills/ (+ hook scripts into ~/.claude/scripts/)
│ ├── check-drift.sh # Verify the deploy (H1-H4 hard, W1-W4 warn); --harvest pulls a stray copy back
│ ├── rebuild-zips.sh # Rebuild dist/*.zip
│ ├── doc-counts.sh # Authoritative counts for the docs
│ ├── test-skills.sh # Integration tests: T1-T9, T16-T18 per skill; T6 + T15 library-wide
│ ├── test-outcomes.sh # Infrastructure tests (T10-T14)
│ ├── normalize-analytics.sh # One-time backfill: canonicalize skill names in analytics logs
│ ├── normalize-house-rules.sh # Rewrite S-lines to the canonical text
│ ├── lib/canonical-skill-name.sh # Shared helper: raw/plugin name → directory name
│ ├── log-skill-usage.sh # PostToolUse hook target (activations)
│ ├── log-outcome.sh # PostToolUse hook target (outcomes)
│ ├── skill-analytics-report.sh # Usage reporting
│ ├── skill-health-report.sh # Skill health CLI ($0 cost)
│ └── hooks/ # SessionStart + workflow hooks; user-settings.snippet.json for ~/.claude/settings.json
├── templates/ # Skill starter templates
├── .claude/ # Project config, observers, test matrix
├── SKILLS_INDEX.md # Detailed skill documentation
├── DEPENDENCY_GRAPH.md # Visual skill relationships
├── PLANNING.md # Current sprint
├── BACKLOG.md # Future work
├── ARCHIVE.md # Completed sprints
└── README.md # This file
Architecture
Skills use progressive disclosure to minimize token usage:
- Level 1: YAML frontmatter — loads at startup for activation matching
- Level 2: SKILL.md content — loads when skill triggers
- Level 3: reference/*.md — loads only when deep-dive needed
Scripts
# Deploy (symlink) all skills into ~/.claude/skills/
./scripts/deploy.sh [--prune] [--dry-run]
# Verify the deploy matches the repo
./scripts/check-drift.sh [--quiet] [--harvest ]
# Rebuild zips for claude.ai / Cowork upload
./scripts/rebuild-zips.sh
# Integration tests
./scripts/test-skills.sh [--verbose] [--skill ]
./scripts/test-outcomes.sh
# Print the counts the docs quote
./scripts/doc-counts.sh [--json]
# View skill usage analytics
./scripts/skill-analytics-report.sh [--days N] [--all]
Test IDs
| ID | Scope | Checks |
|---|---|---|
| T1 | per skill | SKILL.md + config.json exist |
| T2 | per skill | Frontmatter parses with name + description |
| T3 | per skill | config.json valid with required keys |
| T4 | per skill | Required XML sections in SKILL.md |
| T5 | per skill | SKILL.md ≤ 500 lines |
| T7 | per skill | Every integrates_with resolves to a directory |
| T8 | per skill | activation_triggers non-empty |
| T9 | per skill | A/B variants schema (when present) |
| T16 | per skill | Frontmatter name equals the directory name |
| T17 | per skill | Frontmatter keys in the Claude Code allowlist |
| T18 | per skill | `` block present when customer_facing: true (enforced with HOUSE_RULES_ENFORCED=1 until the default flips) |
| T6 | library | No circular depends_on |
| T15 | library | Deploy drift (runs check-drift.sh --quiet; full run only) |
| T10-T14 | infra | test-outcomes.sh: outcome logging, validator, variant assignment |
Latest full run (2026-10-04, ./scripts/test-skills.sh --verbose):
✅ All 1091 tests passed across 99 skills
Analytics
Skill activations and outcomes are logged from every repo via user-level hooks (scripts/hooks/user-settings.snippet.json → ~/.claude/settings.json), not just this one. Names are canonicalized to directory names (scripts/lib/canonical-skill-name.sh), so epiphan-sdr-toolkit:nooks-dial-prep and morning-brief roll up correctly. Data lives in ~/.claude/skill-analytics/.
MCP Integrations
Skills connect to these MCP servers for live data:
| Server | Used By |
|---|---|
Epiphan AI (HubSpot CRM, Clari, orders, qualify_lead, search_product_catalog, SQL) |
Nearly every BDR skill; epiphan-ai-mcp-guide is the reference |
| Epiphan Knowledge (public product docs) | Spec gate S alongside search_product_catalog |
Epiphan Brand (check_my_copy, asset kit, writing style) |
Brand gate B; epiphan-brand-assets wraps it |
| Apollo.io | prospect-research, phone-verification, enrichment skills |
| Clay | phone-verification, paid-marketing enrichment |
| Nooks | nooks-autopilot, sdr-dial-lists, sdr-call-coaching, weekly-kpi-report, plugin skills |
| Gmail (drafts only, never send) | Every skill that produces outbound text (gate D) |
| Google Calendar / Drive | meddic-call-prep, demo handoff, weekly-sales-tracker, survey analysis |
| IBKR | ibkr-api, trading-alert-scheduler |
Hardware
Developed and tested on a MacBook Air 15-inch M4 (24 GB RAM, macOS Sequoia). Parallel-agent guidance in PLANNING.md assumes this machine (5-6 concurrent agents).
Principles
- AI-native — Built for Claude, not ported from elsewhere
- Progressive loading — Lean context, maximum capability
- Battle-tested — Extracted from production BDR workflows
- One source of truth — The repo is deployed by symlink; nothing lives only in
~/.claude/skills/
License
MIT
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add scientiacapital/skills