Agentic Skills Submodules

🗒️ Description

Two git submodules feeding the agent system in agentic-ai-system and agentic-ai-private. One of three repos in the Agentic AI Repos cluster.

RepoVisibilityLicenseURLPlugin name
shared-skillsPublicApache 2.0github.com/200iqlabs/shared-skills200iqlabs-agent-skills
private-skillsPrivateUNLICENSED (proprietary)github.com/200iqlabs/private-skills200iqlabs-private-skills

Both built on the Agent Skills standard (agentskills.io). Each skill = skills/<name>/ with SKILL.md (YAML frontmatter + body). Auto-discovery via plugin manifest. Distribution: plugin marketplace (/plugin marketplace add 200iqlabs/shared-skills) or git submodule.

🧩 shared-skills (public, Apache 2.0)

Modular library of business advisory agents for founders and small teams. Forkable, installable as plugin or submodule.

Business advisory skills (13)

SkillWhat it doesTrigger
cfoCash flow, runway, P&L, budget, unit economics”ile kasy”, “runway”, “ile zarabiamy”
tax-advisorPolish tax (JDG/PSA): CIT, VAT, ZUS, IP Box, B+R, estoński CIT, ryczałt”ile netto z faktury”, JPK, /analiza /optymalizacja
legalPL legal: NDA, B2B, RODO, IP, KSH, AI Act, OWU, founders agreement/analiza /draft /brief /owu /checklist
business-consultantMeeting notes, solution design, estimation, offers, SWOT, process mapping, Make vs n8n vs Zapier vs custom”notatki”, “oferta”, “wycena”, “discovery”
linkedin-contentLinkedIn posts, hooks, content calendar, personal brand”post na LinkedIn”, “ghostwriter”
process-mappingExcalidraw / Mermaid diagrams (sequential business processes only)“zmapuj proces”, “AS-IS TO-BE”
slidesMarkdown → Marp PDF (LinkedIn carousels, talk slides)/slides:init :explore :new :draft :build :tweak :archive
vibe-codingModern UI without UX/UI knowledge (React + Vite + Tailwind 4 + shadcn/ui)“vibe”, design briefs
ingestInbox processing → data/ + archive + catalog/index/ingest, /ingest NAME, /ingest NAME <text>
environment-setupWizard: configure shared-skills + create context files”set up environment”
find-skillsDiscovery + install + security audit from open ecosystem”find skill”, “is there a skill that…“
marketingTODO Phase 2 (placeholder)
product-managerTODO Phase 2 (placeholder)

Developer-workflow skills (added 2026-07, commit 6f12340)

Migrated from loose global ~/.claude/skills/ + ~/.claude/commands/ into the plugin so the same personal coding-agent tools install + update across machines via /plugin update. Placed verbatim (battle-tested, not re-authored via skill-creator).

SkillWhat it doesTrigger
review-fixFetch PR review comments → fix valid issues → commit → push → reply on GitHub/review-fix [PR]
review-loopAutomated Claude↔Copilot review cycle on a PR until stable/review-loop <PR> <change>
prepare-openspec-goalTranscript-checkable completion condition for /goal implementing an OpenSpec change — OpenSpec-specialized sibling of the generic prepare-goal/prepare-openspec-goal <change>

Slash commands (commands/*.md, flat files → plugin-global only; sync-skills.sh symlinks command directories, not flat files, so these are NOT mirrored into consumer repos):

  • /decisions — surface open decisions one at a time, each with a recommendation (PL)
  • /explain-diff — walk a diff (PR / branch vs main) file-by-file (PL)
  • /explain-design — walk an OpenSpec design.md heading-by-heading (PL)

Other skills added since seed

  • prd — agent-ready PRD skill for a 7-day MVP (Startup Builders W3).
  • prepare-goal — generic /goal completion-condition formulator (parent of prepare-openspec-goal).
  • research-en — research outline → deep-research fan-out → report pipeline.

Tools (CLI integrations)

  • tools/clickup/ — taski, daily, projekty
  • tools/revolut/ — transactions for CFO close (OAuth + private cert)
  • tools/common/helpers.sh — shared bash utilities

In agentic-ai-system we use repo-level tools/ (Airtable, Stripe, finances, tech-stack, scheduler, qamera-mcp). The submodule tools/ are used by skills directly. inFakt was refactored from CLI → MCP.

Philosophy

  • Progressive disclosure — metadata → body → references on-demand.
  • Pushy descriptions — undertriggering > overtriggering. Better one trigger too many.
  • Phase 0 placeholders — marketing + product-manager flagged “PLACEHOLDER”.
  • Polish-first — most agents (tax, legal, business consulting) target PL users.
  • CLI > MCP — saves context window.

Context layer (2 layers)

  1. references/ in skill — domain knowledge, distributed.
  2. context/ in user repo — user-specific data, gitignored, created manually or via environment-setup.

Each skill using context files MUST have ## Context Dependencies section with required/recommended files + missing-file message.

Context fileSkills that consume it
company.mdlegal, tax-advisor, cfo
consultant-profile.mdbusiness-consultant
projects-portfolio.mdbusiness-consultant
author-profile.mdlinkedin-content
finances.mdcfo
legal-entities.mdlegal, tax-advisor

MANDATORY skill-creation workflow

For new skills or significant edits: use /skill-creator. Do NOT hand-write SKILL.md.

  1. /skill-creator — capture intent, interview, draft.
  2. Generate min. 5 test prompts per agent.
  3. Run evals + review.
  4. Iterate on feedback + benchmarks.
  5. Optimize description (target ≥ 80% triggering accuracy).

Can skip for typo fixes, reference file additions, date updates, reference-only changes.

Recent commits

  • 6f12340 (2026-07-13) feat(skills): add developer-workflow skills + slash commands (review-fix, review-loop, prepare-openspec-goal + /decisions, /explain-diff, /explain-design)
  • 06e784bef30e79 feat(prd): agent-ready PRD skill; refactor(linkedin-content): plain-Polish, context-first
  • f6fff6f feat(prepare-goal): formulate /goal completion conditions
  • 3a06818 refactor(cfo): route inFakt through MCP, drop CLI scripts
  • 405ce5d fix(cfo): use OAuth-refreshed Revolut tokens
  • 61ae0fb feat(find-skills): add skill discovery with security audit
  • 9c04da1 feat(ingest): drop on-disk summary file — chat report + git są audit trail

🧩 private-skills (proprietary, UNLICENSED)

Skills not suitable for open-source distribution (confidential know-how, internal platform integrations).

Skills

1. coach-the-five

Personal business coach based on Tomasz Karwatka’s methodology from “The Five: Pierwsze pięć lat prowadzenia firmy”.

  • Author: Paweł Lipowczan. Source: “The Five” — Tomasz Karwatka.
  • Role: mentor (40%) + consultant + coach.
  • Triggers: startup strategy, PMF, scaling readiness, founder burnout, co-founder agreements/conflicts, equity splits, fundraising, IP+services model, exit strategy, EBITDA. PL: wypalenie, wspólnik, udziały, sprzedaż firmy, kryzys.
  • Frameworks: tornado framework, “PE yourself”, founders agreement.
  • References: 7 modules + checklists per phase of first 5 years.

2. platforma-tworzenie-sesji-zdjeciowych

Operator skill for Qamera AI platform via MCP server qamera-ai-hetzner.

  • Goal: agent (not human) registers + monitors photo sessions in Qamera AI SaaS.
  • MCP transport: Streamable HTTP (fix dbf9a55 in main repo).
  • Tools: get_products, get_models, get_sceneries, get_presets, register_photo_shoot, get_images.
  • Use cases: create session from existing records, validate IDs, run generation, check image status.

Recent commits (2026-05-10)

  • 6f59ac9 feat: add platforma-tworzenie-sesji-zdjeciowych skill (Qamera MCP)
  • f5a6c7f chore: remove skill workspaces from repo to avoid submodule pollution
  • 8ca6169 feat: initial private-skills repo with coach-the-five skill

🧩 How they compose

agentic-ai-system  /  agentic-ai-private          ← consumer repos
├── shared-skills/    (submodule)
├── private-skills/   (submodule)
└── tools/sync-skills.sh                          ← creates symlinks:
                                                    skills/<name>/ → .claude/skills/<name>/
                                                                     .github/skills/<name>/
                                                                     .cursor/skills/<name>/
                                                                     .agent/skills/<name>/

Lifecycle

  1. Read — IDE scans its own skill dir, loads SKILL.md, uses frontmatter description for auto-routing.
  2. Updategit submodule update --remote./tools/sync-skills.sh → commit new submodule SHAs.
  3. New skill — create in submodule via /skill-creatorgit submodule update --remote → add name to .gitignore (Synced agent skills section) → ./tools/sync-skills.sh.

Global plugin vs submodule coexistence. The dev-workflow skills also install globally via the 200iqlabs-agent-skills plugin, so in a consumer repo (e.g. agentic-ai-private) a skill can exist at both scopes — project symlink (submodule) AND global plugin. Claude Code dedupes by name — no merge, no corruption — and both point at the same repo. The one real hazard is version skew: the plugin (refreshed via /plugin update) vs the submodule (pinned SHA, git submodule update --remote) can drift. Gotcha: /plugin update reinstalls from the local marketplace clone — run /plugin marketplace update shared-skills first or it silently keeps stale content.

Generic skills, local data

Skills NEVER hardcode paths to context/operations/tech-stack/.... They get data via canonical files (finances.md, company.md) declared in ## Context Dependencies. This makes the skills cleanly forkable for users outside 200IQ LABS.

🧩 Glossary

  • Agent Skill — agentskills.io standard: dir with SKILL.md (frontmatter + body), optional references/ and scripts/.
  • Plugin manifest.claude-plugin/plugin.json.
  • Progressive disclosure — metadata → body on activation → references on-demand.
  • Triggering accuracy — % cases skill activates correctly. Target ≥ 80%.
  • MCP server — Model Context Protocol; alternative to CLI tools.
  • Skill workspace — local skill instance with runtime state; removed from private-skills (commit f5a6c7f).