swarm-implement - SKILL.md Agent Skill

name: swarm-implement description: Codex adapter for complex implementation work using a swarm-like workflow. Use for multi-file features, bug fixes, refactors, risky code changes, or tasks that benefit from exploration, scoped planning, implementation, review, and validation.

Read .claude/skills/swarm-implement/SKILL.md for the source workflow, then apply Codex-native tooling.

Codex-specific execution notes:

Use update_plan for substantial multi-step work.
Use multi_tool_use.parallel for independent repo reads and searches.
Use apply_patch for manual edits.
Use focused shell validation after each meaningful change.
Bring in narrower skills as needed: $engineering-conventions, $writing-tests, $running-tests, $qa-sweep, or $issue-tracer.
Before calling work complete, run a no-unwired/no-deferred gate: verify every new behavior is wired through config/registration/docs/tests/generated artifacts as applicable, no work was deferred or declared out of scope without explicit user instruction, and scope decisions were not made silently.
For any worktree edit, also run the swarm final gate before declaring completion:
- objective validation is recorded,
- an independent implementation reviewer approved the actual latest diff,
- a separate critic approved after reviewer approval,
- every NEEDS_REVISION or BLOCKED item was fixed and re-reviewed,
- no edit occurred after the latest reviewer/critic approval.
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 a task-local review artifact that names the verdicts, evidence reviewed, and responses to every blocker.
Do not count explorer output, plan criticism, passing tests, or self-review as the final implementation reviewer gate when subagent delegation is available.

Do not invoke OpenCode /swarm commands from Codex unless the user explicitly asks to operate OpenCode Swarm itself.