Agents

Specialized subagents that auto-delegate or invoke explicitly.

Available Agents

Agent Purpose Mode
/planner Design and plan before implementing Read/Write
/reviewer Audit code for issues, security, patterns Readonly
/fixer Debug and fix bugs Read/Write
/verifier Validate completed work Readonly

Planner

Invoked when tasks are complex, ambiguous, or need architectural decisions.

/planner Add caching layer to the API

Process:

  1. Investigate existing code and patterns
  2. Analyze requirements and constraints
  3. Design step-by-step implementation plan
  4. Define success criteria

Output: Saves plan to ~/.genesis/plans/[name].md

Reviewer

Audits code against principles and playbooks. Runs readonly — can't modify files.

/reviewer check the auth changes

Checks:

  • Principle violations (.1 through .10)
  • Security issues
  • Pattern consistency with playbooks
  • Error handling
  • Type safety

Output: Verdict (approved/changes-requested) with categorized findings

Fixer

Debugs issues systematically, finds root cause, fixes properly.

/fixer Login returning 500 error

Process:

  1. Triage severity (!crit, !high, !med, !low)
  2. Reproduce the issue
  3. Diagnose root cause
  4. Apply minimal fix
  5. Prevent recurrence (tests, postmortem)

Output: For significant bugs, creates postmortem at ~/.genesis/postmortems/

Verifier

Validates that work claimed as complete actually works. Runs readonly.

/verifier confirm the feature works

Checks:

  • Implementation exists and is functional
  • Happy path works
  • Edge cases handled
  • Tests pass and are meaningful
  • "Done When" criteria from plan are met

Output: Verification status (verified/incomplete/failed) with details

Auto-Delegation

Agents can be triggered automatically based on context:

  • Planner — Complex/ambiguous tasks
  • Reviewer — After code changes (proactive)
  • Fixer — When errors occur
  • Verifier — After tasks marked done

Or invoke explicitly with /agent-name.