project-skill-bootstrap

name

project-skill-bootstrap

description

Create project-local skills populated with actual project data, stack-specific conventions, and domain-specific procedures. Use after scaffolding to replace generic skill placeholders with real project-aware guidance that helps agents work effectively in this specific repo. Do not use before the project has established patterns (scaffold first) or for universal skills that belong in a skill library.

Project Skill Bootstrap

Use this skill to create the repo-local skills layer with actual project content.

Goals

• Make project context easy to reload for any agent or session
• Encode actual stack and domain standards, not generic placeholders
• Reduce ambiguity for weaker models
• Keep prompts short by moving stable procedure into local skills

Modes

• foundation: Populate baseline skills with actual project data
• synthesis: Add stack- or domain-specific skills based on project evidence

Foundation mode procedure

1. Read the canonical brief

Read the project brief for project facts, stack decisions, and constraints.

2. Gather project signals

Extract concrete facts from the repo:

# Stack evidence
cat package.json | jq '{name, scripts, dependencies}' 2>/dev/null
cat Cargo.toml 2>/dev/null | head -30
cat pyproject.toml 2>/dev/null | head -30

# Structure
tree -L 2 -I 'node_modules|dist|target|.git'

# Patterns (sample real code)
head -50 src/index.* src/main.* 2>/dev/null
ls tests/**/* 2>/dev/null | head -10

3. Populate baseline skills

For each baseline skill, rewrite with actual project content:

Skill	What to populate
project-context	Actual project summary, canonical doc paths, key architectural decisions
repo-navigation	Actual directory layout, key files, common agent queries
stack-standards	Real framework/language conventions, testing commands, linting rules
ticket-execution	Standard lifecycle + any project-specific stage requirements
docs-and-handoff	Project-specific doc paths and conventions

Keep generic skills (workflow-observability, research-delegation, isolation-guidance) as-is unless the project has specific requirements.

4. Write updated skills

Write each skill with YAML frontmatter (name and description required). Every skill must reference real files, real commands, and have no placeholder text.

Synthesis mode procedure

1. Analyze the stack

From the canonical brief, identify:

• Framework/runtime in use
• Database/ORM in use
• API patterns in use
• Deployment target
• Domain-specific workflow requirements

2. Research external patterns (reference only)

Look for relevant patterns from:

• Anthropic skills repo, awesome-copilot collection
• Framework documentation and best practices
• Community skill registries

CRITICAL: Do NOT install external skills directly. Use them as REFERENCE ONLY for synthesizing project-specific skills.

3. Synthesize project-specific skills

Create skills specific to THIS project. Examples:

• React project → component-patterns with project's component conventions
• API project → api-contracts with project's schema format
• Database project → migration-safety with project's DB engine specifics
• Testing-heavy → test-patterns with project's test runner and fixture conventions

4. Quality rules for synthesized skills

• Each skill must be repo-specific, not generic paraphrasing of docs
• Prefer procedure over reference dumping
• Each skill must justify its existence — don't create skills just because you can
• Keep total skill count manageable (weaker models struggle with >12-15 skills)
• Every skill must have proper YAML frontmatter

Output contract

Skills written to the project-local skills directory with:

• Project name in metadata
• Real file references (verified to exist)
• Real commands (verified to work)
• No placeholder text ([...], TODO, etc.)

Rules

• Use this after the main scaffold exists — it refines, not replaces
• Follow agent-prompt-engineering rules when creating skills to avoid anti-patterns
• Start with the smallest useful pack; add only when project evidence justifies it
• Never auto-install external skills — synthesis from reference only

Failure handling

• No stack profile exists: Run stack-profile-detector first
• Referenced file doesn't exist: Remove reference or note "planned: [file]"
• Command doesn't exist: Check package.json/Makefile; if truly missing, note gap
• Domain too complex to summarize: Create multiple domain skills (domain-glossary, domain-rules, domain-workflows)

References

• This is step 6 of the scaffold-kickoff flow — continue to ../agent-prompt-engineering/SKILL.md