Skip to main content

Agents and Skills

The .claude/ directory contains specific agent definitions and skills designed to orchestrate work within acme-vue-encore.

Agent Personas​

The repository defines several specialized agent personas in .claude/agents/:

  • architect: A read-only planning agent that analyzes requirements, decomposes work, and validates approaches against the spec spine.
  • explorer: A read-only investigation agent that traces dependencies and gathers context.
  • implementer: A read-write execution agent that applies code changes according to a plan. It does not design; it only executes.
  • reviewer: A read-only review agent that examines changes for correctness, security, and spec compliance.
  • encore-expert: A read-only domain specialist for the Encore.ts framework and the specific security primitives in apps/api/lib.

By explicitly switching between these personas (e.g., using /agent architect), you ensure that Claude operates with the appropriate mindset and constraints for the task at hand.

Claude Skills​

Skills are modular, executable workflows located in .claude/skills/.

Key skills include:

  • /init: Executes the session bootstrap protocol defined in AGENTS.md.
  • /scaffold-feature: Automates the creation of a full-stack vertical slice (endpoint, migration, frontend component).
  • /validate-and-fix: Runs the CI checks (type-checking, testing, linting) locally and automatically attempts to fix any errors it finds.
  • /implement-plan: Instructs the implementer agent to execute a previously generated plan step-by-step.

To use a skill in Claude Code, simply type the slash command (e.g., /scaffold-feature).