Skip to Content
Cheat Sheet/bootstrap-agents

/bootstrap-agents

The third phase of project setup — injects project-specific context from the PRD and architecture doc into every agent’s configuration so they reason about your project, not a generic template.

When to use it

Run this after /bootstrap-product and /bootstrap-architecture have shipped. The command reads both documents and updates each agent’s .claude/agents/<name>.md file with project-specific context: which tech stack to assume, which conventions to follow, which boundaries not to cross. Without this step, every agent invocation starts from generic engineering defaults rather than the project’s actual stance.

The difference matters more than it sounds. The ticket-worker that knows “this project uses Astro + Starlight, deployed to Render” makes different scaffolding choices than one that doesn’t. The PR-reviewer that knows “every Ready ticket must carry a safety:* label” can enforce it; one that doesn’t can’t. Specialization is the difference between “agents that work for you” and “agents that work in general.”

Don’t skip this even if the project feels too small to warrant specialization. Specialized agents waste less time asking questions the project already answered.

How it fits

Inputs are the PRD and architecture doc. Output is updated agent config files that every later agent-invoking command consumes.

What it does (quick)

  • Verifies the PRD and architecture doc exist (pre-flight)
  • For each agent in .claude/agents/, identifies the project-specific context that should be injected
  • Updates the agent’s .md file with the project’s stack, conventions, safety taxonomy, and architectural boundaries
  • Preserves the agent’s role and authority — only the project-context section is rewritten
  • Outputs a summary of which agents were updated and what changed, so the operator can spot-check the specialization
  • /bootstrap-architecture — predecessor; the architecture doc is the primary specialization input
  • /bootstrap-workflow — successor; activates the workflow the now-specialized agents will participate in
  • /upgrade — when the framework updates upstream, this command re-runs to re-specialize against the new templates

Canonical spec: .claude/commands/bootstrap-agents.md

Last updated on