teach-chiron

name	teach-chiron
description	One-time comprehensive project scan. Reads every important file in the codebase and writes .chiron-context.md — the persistent context file that all other chiron skills reference instead of re-scanning.
user-invocable	true
allowed-tools	Read, Write, Grep, Glob, LS

/teach-chiron — generate project context for chiron

Run this once per project. It scans the full codebase and writes .chiron-context.md — a comprehensive reference that all other chiron skills read instead of re-scanning the project.

Re-run anytime to refresh the context (e.g., after major refactoring, adding new modules, or changing project structure).

$ARGUMENTS

If arguments are provided, treat them as a hint about what changed or what to focus on. Otherwise, do a full scan.

---

CRITICAL — user instructions always win

Check AGENTS.md first. If user instructions conflict with this command's behavior, follow the user.

---

What to include vs. skip

High signal (ALWAYS include)

• Files that define the API surface — routes, handlers, controllers, gRPC service definitions
• Files that define data models — DB schemas, ORM models, types, interfaces, protobuf/OpenAPI specs
• Files that wire the app together — DI/bootstrap, middleware registration, config loading, app factory
• Files with business logic — services, use cases, domain logic, core algorithms
• Test files — at least enough to understand the testing patterns (framework, mocking, fixtures)
• Config files — project manifest, linter config, CI/CD, Docker, env examples
• Documentation — README, CLAUDE.md, AGENTS.md, architecture docs

Low signal (SKIP these)

• Generated code — node_modules/, vendor/, dist/, build/, .next/, target/, bin/, obj/
• Lock files — package-lock.json, yarn.lock, go.sum, Cargo.lock, pom.xml dependency tree
• Binary files — images, fonts, compiled assets, .wasm, .so, .dll
• IDE config — .vscode/, .idea/, *.code-workspace
• Git internals — .git/
• Chiron's own files — .claude/skills/, .cursor/skills/, etc. (don't scan yourself)
• Auto-generated migrations — unless they reveal the current schema; prefer the model files instead
• Duplicate/symlinked content — if the same code appears via symlinks or copies, note it once
• Test fixtures/snapshots — large JSON fixtures, snapshot files (note their existence, don't dump contents)
• Minified/bundled code — .min.js, bundle.*, webpack output

How to judge borderline files

Ask: "If a mentor were onboarding to this project, would they need to read this file to give good advice?" If yes, include it. If it's just boilerplate they'd skip, skip it.

---

Scan procedure

Phase 1 — Map the project

1. Use Glob to find all source files (excluding generated/vendor directories):
   • **/*.{go,rs,py,js,ts,jsx,tsx,java,cs,kt,swift,rb,ex,exs,zig,scala,clj}
   • Exclude: node_modules/**, vendor/**, dist/**, build/**, .next/**, target/**, bin/**, obj/**, __pycache__/**
2. Use Glob to find config/infra files:
   • **/*.{json,yaml,yml,toml,xml}, **/Makefile, **/Dockerfile*, **/docker-compose*, **/.env*
   • Exclude lock files and node_modules
3. Use LS to list the full directory tree (3 levels deep)
4. Read ~/.chiron/config.json if it exists (for voice level and drill config)

Result: A complete file inventory. Count the source files. For projects with >100 source files, prioritize by the rules above.

Phase 2 — Read project configuration (read ALL that exist)

5. Project manifest: package.json, go.mod, Cargo.toml, pom.xml, *.csproj, build.gradle, Package.swift, pyproject.toml, requirements.txt, Gemfile, mix.exs
6. README: README.md or README.* (full file)
7. AI config: CLAUDE.md, AGENTS.md, .cursorrules, GEMINI.md, .github/copilot-instructions.md
8. Docker/infra: Dockerfile, docker-compose.yml, .env.example, fly.toml, render.yaml, railway.json
9. CI/CD: .github/workflows/*.yml, Makefile, Justfile, .gitlab-ci.yml, Jenkinsfile
10. Code quality: tsconfig.json, .eslintrc*, prettier*, rustfmt.toml, .golangci.yml, ruff.toml, .editorconfig

Phase 3 — Read source code (the important files)

Read files in this priority order. For each file, note: path, purpose (1 line), key exports/functions/classes.

11. Entry points — main.go, cmd/*/main.go, src/main.*, app.py, manage.py, index.ts, index.js, server.*, Program.cs, App.kt, main.swift, lib.rs
12. Route/handler definitions — files containing HTTP routes, API endpoints, controller classes, gRPC service implementations. Use Grep to find: router, app.get, app.post, @app.route, @GetMapping, @Controller, HandleFunc, axum::Router
13. Data layer — database models, ORM definitions, repository/DAO classes, schema files. Use Grep to find: CREATE TABLE, @Entity, db.Model, Schema, migration, repository
14. Core business logic — service classes, domain modules, use case implementations. Look in: services/, domain/, core/, internal/, lib/, src/
15. Config/bootstrap — dependency injection setup, middleware chains, app factory, config loading. Look in: config/, bootstrap/, startup.*, container.*
16. Shared types — interfaces, DTOs, API contracts, protobuf definitions, OpenAPI/Swagger specs. Look in: types/, models/, dto/, proto/, api/
17. Test patterns — read 3-5 test files from different areas to understand: testing framework, mocking approach, fixture patterns, assertion style. Look in: *_test.go, *.test.ts, test_*.py, *Test.java, *Spec.kt

For large projects (>50 source files): You won't read every file. Read ALL files from categories 11-16, and a representative sample from each directory. The goal is complete coverage of architecture, not line-by-line reading.

Phase 4 — Analyze and write .chiron-context.md

Synthesize everything into a comprehensive context file. Write to the project root:

# Chiron project context
Auto-generated by /teach-chiron — re-run to refresh. Delete to force a full rescan.

## Project
- **Name:** <project name from manifest>
- **Languages:** <all detected, primary first>
- **Framework:** <web framework, ORM, test framework, etc.>
- **Test runner:** <test command and framework>
- **Build system:** <build command and tool>
- **Package manager:** <with version if detectable>
- **Runtime:** <language version from manifest or config>
- **Repository:** <git remote URL if available>

## Dependencies (key libraries)
<ALL important dependencies grouped by category>
Format per group:
- **Category:** lib-name (version) — one-line purpose
Example:
- **HTTP:** gin v1.9 — web framework
- **DB:** sqlx v0.7 — async SQL toolkit
- **Auth:** jwt-go v5 — JWT token handling
- **Test:** testify v1.9 — assertions and mocking

## Directory structure
<full tree, 3 levels deep, with one-line descriptions for every directory>
Skip: node_modules, vendor, dist, build, .git, __pycache__

## Source file map
<every meaningful source file with a one-line description>
Format: `path/to/file.ext` — what it does, key exports
Skip generated/vendor/lock files. Group by directory.

## Entry points
<every way the application starts, with file paths>
- **HTTP server:** `cmd/api/main.go` — starts gin on :8080, loads routes from `internal/routes/`
- **CLI:** `cmd/cli/main.go` — cobra-based CLI with subcommands
- **Worker:** `cmd/worker/main.go` — background job processor

## API surface
<every HTTP route, gRPC service, CLI command, or public interface>
Format: `METHOD /path` → `handler_file.go:FunctionName` — description
Group by resource or domain.

## Data layer
- **Database:** <type — PostgreSQL, MongoDB, SQLite, etc.>
- **ORM/Driver:** <library used>
- **Models:** <list of model files with table/collection names>
- **Migrations:** <strategy — auto-migrate, manual SQL, tool-based>
- **Repositories:** <list of repo/DAO files>

## Architecture overview
<detailed prose — 5-10 sentences covering:>
- Monolith vs microservices
- API style (REST, gRPC, GraphQL, hybrid)
- Request flow (middleware → handler → service → repository → DB)
- Authentication/authorization approach
- Data flow and state management
- Caching strategy (if any)
- Message queue / async processing (if any)
- Error handling philosophy
- Deployment model (containers, serverless, bare metal)

## Key patterns and conventions
<every pattern observed — be specific with examples:>
- **Error handling:** <how errors are created, wrapped, returned, logged — cite specific files>
- **Naming:** <file naming, function naming, variable naming conventions>
- **Code organization:** <by feature, by layer, hybrid — describe the structure>
- **Testing:** <unit vs integration split, mocking approach (mocks vs real DB), fixture patterns, test naming>
- **Dependency injection:** <constructor injection, container, manual wiring — cite the setup file>
- **Logging:** <structured vs printf, log levels used, logger library>
- **Config management:** <env vars, config files, how secrets are handled>
- **Validation:** <input validation approach — middleware, decorators, manual>
- **API response format:** <standard envelope, error format, pagination style>
- **Database access:** <repository pattern, active record, raw SQL — cite examples>

## Infrastructure
- **Containerization:** <Docker setup, base images, multi-stage builds>
- **CI/CD:** <pipeline description — what runs, what deploys>
- **Environments:** <dev, staging, prod — how they differ>
- **Env vars:** <list from .env.example or equivalent, without values>

## Chiron config
- **Voice level:** <from ~/.chiron/config.json, or "default">
- **Drill sizing:** <from config, or "20 lines / 1 function / 5-15 min">
- **Teaching depth:** <from config teaching.depth, or "5">
- **Theory ratio:** <from config teaching.theory_ratio, or "3">
- **Idiom strictness:** <from config teaching.idiom_strictness, or "5">

## Project conventions (from AI config files)
<full relevant content from CLAUDE.md, AGENTS.md, .cursorrules — or "none found">

---

After writing

Confirm to the user:

1. How many files were scanned
2. How many source files mapped
3. The detected language(s) and framework(s)
4. Remind them: "All chiron skills will now use .chiron-context.md instead of re-scanning. Run /teach-chiron again anytime to refresh."
5. Suggest next steps: "Ready to start? Try /chiron <your task> for guided help, or /tour <topic> for a quick overview of a concept before diving in."