Contributing

How to contribute to Make_Skills. The four-question discipline, where to add what, and what gets rejected at review.

The contributor-facing version of CONTRIBUTING.md at the repo root.

Quick start

git clone https://github.com/Lizo-RoadTown/Make_Skills.git
cd Make_Skills
# Self-host the platform (full instructions in /docs/installation)
cd platform/deploy && cp .env.template .env  # fill in ANTHROPIC_API_KEY
docker compose up -d --build
# In another terminal: run the web UI
cd ../../web && npm install && npm run dev

Open http://localhost:3000.

Where to add what

You're adding...	It goes...
A SKILL.md (markdown wisdom)	`skills/<name>/SKILL.md`
A subagent persona	`subagents/<name>/AGENTS.md` + `deepagents.toml`
A FastAPI endpoint	`platform/api/main.py` (or a sub-module)
An agent tool (`@tool` function)	`platform/api/<area>/tools.py` then wire into `builtin_tools`
A web UI page	`web/app/<route>/page.tsx`
Public docs	`web/content/docs/<name>.mdx` (or `concepts/`, `tutorials/`, etc.)
An ADR or design proposal	`docs/decisions/` or `docs/proposals/` (audit trail, separate from `web/content/docs`)

The four-question PR template

Every PR is reviewed against:

What changes for a self-host user? Do they need to update env vars, rebuild, run a migration?
What changes for the hosted-multitenant deployment? Same questions, plus: does it touch tenant scoping correctly?
Tests for both modes? At minimum: a test with tenant_id="default" (self-host) AND a test with a synthetic non-default tenant_id (hosted) verifying isolation.
Docs for both modes? The relevant doc explains how the change appears in both modes.

A PR missing any of those four is incomplete.

Anti-patterns rejected at review

Hardcoded tenant_id="default" outside the NoAuthBackend self-host path
Direct filesystem reads bypassing the config_loader abstraction (when added)
"Multi-tenancy later" — once data is being written, retrofitting is a migration nightmare
Tests that only cover one mode
A new feature without docs in the same PR
Premature tool extraction — promoting a skill to a tool before it's been used 3+ times the same way (see agentic-upskilling)
ADR amended in place after acceptance — write a superseding ADR instead, mark old one as Superseded by

Patterns we like

Use Depends(current_tenant) in any new FastAPI endpoint that touches tenant data
Filter every database query by tenant_id (helper functions encouraged)
Document the data type's tenant scope in a comment at the top of the model definition
Provide both a self-host smoke test (docker compose up && curl /healthz) and a multi-tenant test (mock auth, verify isolation)
New skills follow the agentic-skill-design PROBE → DECIDE → ACT → REPORT pattern

License

Apache 2.0 (LICENSE). Inbound = outbound: by contributing, you agree your contributions are licensed under the same terms.

The license decision is captured as ADR-001 — patent grant matters in AI, highest adoption in the AI/ML ecosystem, dual-license-able later if commercial defense becomes priority.

Code of Conduct

TBD. Defaulting to the Contributor Covenant until adopted explicitly. Be kind, give credit, assume good faith, no harassment.

Questions

Architecture questions — open a Discussion (preferred) or a GitHub issue with the architecture label
Bugs — GitHub issue with reproduction steps in self-host mode (most accessible to reproduce)
Feature ideas — open a Discussion first; major features should start as a one-page design proposal in docs/proposals/ that addresses the four discipline questions

Contributing

On this page