Swarm

Swarm

Read .opencode/skills/swarm/SKILL.md for the source behavior model, then adapt it to Codex.

Codex-specific execution notes:

• Treat this as a workflow posture, not an instruction to call OpenCode /swarm commands.
• Use parallel local reads/searches where safe.
• Use update_plan for visible task tracking on substantial work.
• Add independent review pressure through $qa-sweep, $swarm-pr-review, or self-critic checks when subagents are unavailable.
• Keep the user looped in with short progress updates during longer work.
• For any task that edits code, tests, docs, package metadata, release notes, or skill files, final completion requires:
  1. objective validation evidence,
  2. independent implementation reviewer approval on the actual latest diff,
  3. separate critic approval after reviewer approval,
  4. no unresolved NEEDS_REVISION/BLOCKED items, and
  5. no edits after those approvals.
• Record reviewer and critic verdicts in durable task artifacts. For issue-tracer work, use 08b-implementation-review.md and 09-final-critic.md; otherwise use task-local review artifacts unless the repo forbids artifacts.
• Explorer subagents, a plan critic, passing tests, or self-review do not satisfy the final implementation-review gate when subagent delegation is available and swarm work is authorized.

For implementation tasks, prefer $swarm-implement; for fresh PR review, prefer $swarm-pr-review; for known PR feedback closure, prefer $swarm-pr-feedback; for bug tracing, prefer $issue-tracer.