Skip to content

Agent Delegation

Framework for delegating tasks to agents with a durable audit trail and human review gates.

When to delegate

Delegate a task to an agent when: - The work is well-scoped with clear acceptance criteria - The output is a concrete artifact (code, document, research summary) - Human review of the output is feasible before it's acted on

How to delegate

  1. Copy a template from templates/ that matches the task type
  2. Fill in the scope, acceptance criteria, and output location
  3. Hand the filled template to the agent (paste into Claude Code or Cursor chat)
  4. When the agent completes, review the output against the acceptance criteria
  5. Append an entry to audit-log.md with the outcome

Templates

Template Use for
task-delegation.md General code, research, or documentation tasks

Audit log

All completed delegations are recorded in audit-log.md. This is an append-only log — never edit past entries.

Review gates

  • Sync review (default): You review output before it's saved or acted on
  • Async with spot-check: Agent writes output to a draft location; you review at next available time
  • Mark the review status in the audit log entry: approved, revised, or rejected