Create Agent Definition

Canonical Instructions

Read and follow repository-root AGENTS.md; this skill narrows that contract and never overrides it.

Description And Trigger

Use when adding or updating agent definitions under .claude/agents/ or .codex/agents/.

Inputs

• approved role list
• agent responsibilities
• repository governance rules

Preconditions

• the role is approved or clearly justified by the current repository workflow

Ordered Workflow

1. Define role and purpose.
2. Record trigger conditions, inputs, outputs, and tools.
3. Record permissions, scope, PR authority, approval authority, escalation rules, and evidence requirements.
4. Validate that the agent cannot approve its own work or exceed Phase 0 scope.

Required Outputs

• agent definition
• permission and scope summary

Evidence

• mapping to bootstrap documents
• explicit permission boundaries

Validation Checklist

• all mandatory sections exist
• permissions are least-privilege
• review agents are read-only

Failure Conditions

• agent permissions exceed role needs

Escalation Conditions

Escalate if the requested agent duplicates another role or requires broader permissions than approved.

Forbidden Behavior

• granting self-approval
• leaving repository scope implicit