菜单
Join Deep Code Review as a reviewer for an active review run.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | dcr-reviewer-init |
| description | Join Deep Code Review as a reviewer for an active review run. |
| invocation-args | {"task_name":"required","suffix":"required","domain":"optional"} |
Do not start this reviewer role implicitly.
Bind the reviewer run to an explicit task_name and suffix at invocation time.
Add domain when the human or lead assigns one.
.agent/skills/code-review/SKILL.md.agent/playbooks/deep-code-review/PLAYBOOK.md.agent/playbooks/deep-code-review/template/round_template/README.mdKeep these hot-loaded for the duration of the playbook:
.agent/skills/code-review/SKILL.md.agent/playbooks/deep-code-review/skills/dcr-reviewer-init/SKILL.mdFor every new round, begin by reading the active round README.md.
Then follow 00_round_context.md for the full round contract and read order.
If this is not the first round, also read the prior round outputs referenced in 10_previous_round_feedback.md.
Before writing findings, know:
Your output file should be:
reviewers/reviewer_<suffix>.mdDo not edit findings_ledger.md.
Finding ids in shared artifacts are maintained by the lead only.
You are responsible for:
.agent/skills/code-review/SKILL.mdThe active round packet is the live source of truth.
Do not infer scope, reviewed state, or reviewer target paths from chat if the round artifacts already define them.
Do not duplicate or override the active round packet with ad hoc reviewer instructions unless the human explicitly changes the round plan.
Do not drop from the review automatically.
If you reach No findings.:
Lightweight follow-up mode means:
Close a task cleanly at the end of the standard workflow. Use for summary, promotion decisions, project progress updates, session state updates, and session log updates.
Initialize a dotagent working session from within a runtime. Use as the native runtime entrypoint when the project is already initialized and the agent needs to start from the framework correctly.
Join Deep Co-Planning as a reviewer for an active planning run.
Start Deep Co-Planning as the lead for an active planning run.
Start Deep Code Review as the lead for an active review run.
Review code changes with production-grade rigor. Use for mandatory self-review before presenting implementation as complete, or when the user explicitly asks for a review. Focus on findings first: correctness, regressions, scale, architecture, best practices, extendability, and production readiness.