name	codebase-review-swarm
audience	swarm-plugin
description	Codex adapter for running a rigorous, quote-grounded codebase review or security/QA/accessibility/performance/AI-slop/enhancement audit. Use for full-repo or large-subsystem review reports; not for normal implementation.

Codebase Review Swarm

Read and follow ../../../.opencode/skills/codebase-review-swarm/SKILL.md as the canonical workflow.

Treat this as a read-only review workflow. Do not modify source files.
Write review artifacts only under .swarm/review-v8/runs/<run_id>/.
Run Phase 0 inventory first and stop for review-mode selection unless the user already selected tracks and explicitly authorized continuing.
Preserve the skill's evidence rule: no quote, no claim.
For dispatch lanes, treat output as a preview and call retrieve_lane_output for full output_ref artifacts before consuming ledgers, counting candidates, routing validation, or declaring coverage clean; degraded or incomplete lane output is an explicit coverage limitation.
Use rg, focused file reads, targeted shell commands, and independent validation to close coverage units before writing the final report.
If the user asks for fixes after the report, start a separate implementation workflow rather than editing during the review run.

name	codebase-review-swarm
audience	swarm-plugin
description	Codex adapter for running a rigorous, quote-grounded codebase review or security/QA/accessibility/performance/AI-slop/enhancement audit. Use for full-repo or large-subsystem review reports; not for normal implementation.

Read and follow ../../../.opencode/skills/codebase-review-swarm/SKILL.md as the canonical workflow.

Treat this as a read-only review workflow. Do not modify source files.
Write review artifacts only under .swarm/review-v8/runs/<run_id>/.
Run Phase 0 inventory first and stop for review-mode selection unless the user already selected tracks and explicitly authorized continuing.
Preserve the skill's evidence rule: no quote, no claim.
For dispatch lanes, treat output as a preview and call retrieve_lane_output for full output_ref artifacts before consuming ledgers, counting candidates, routing validation, or declaring coverage clean; degraded or incomplete lane output is an explicit coverage limitation.
Use rg, focused file reads, targeted shell commands, and independent validation to close coverage units before writing the final report.
If the user asks for fixes after the report, start a separate implementation workflow rather than editing during the review run.

codebase-review-swarm