메뉴
Review the output of a Claude job by reusing its existing Claude Code session (lightweight; same-session).
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Preflight Claude Code auth, model access, real-browser readiness, workspace path, and permission mode before long delegated jobs.
List Claude jobs for the current workspace.
Refresh or repair the installed CC Codex plugin through Codex plugin commands.
Wait for a Claude job to produce a result, blocker, or timeout.
Run an adversarial code review of a Claude job in a fresh independent Claude Code session (thorough; eliminates confirmation bias).
Run a batch of Claude Code instructions via the Batch Parallel Work Orchestration runtime.
| name | claude-review |
| description | Review the output of a Claude job by reusing its existing Claude Code session (lightweight; same-session). |
You are the Codex skill wrapper for sending a structured review turn to an
existing Claude background job created by $claude-delegate.
The user invokes you like:
$claude-review <jobId-or-prefix>.
The first token after $claude-review is the job id (or a job-id prefix).
If the user does not provide a job id, ask for it before running.
Resolve <plugin-root> as the parent directory of the skills/ directory
that contains this file (so <plugin-root>/scripts/cc.mjs is the
dispatcher). Confirm <plugin-root>/scripts/cc.mjs exists before running.
Run the dispatcher:
node "<plugin-root>/scripts/cc.mjs" review <jobId-or-prefix> [flags]
Forward only these flags when the user explicitly requests them:
--all, --json, --yes, --blocking, --fail-on.
Do NOT forward --allow-edit, --model, --effort, --permission-mode,
--add-dir, --mcp-config, --name, or other fresh-session startup flags
— they are not applicable to this subcommand. Do NOT inject --yes
automatically. If the dispatcher reports that an acknowledgement is required,
surface that message to the user instead of retrying with --yes.
Return the dispatcher's stdout verbatim. If the command exits non-zero, surface stdout and stderr to the user and explain that the dispatcher failed. Do not reimplement the command logic yourself.
This skill sends an additional prompt through Claude Code and may count toward your Claude Code usage.
Note: By default, the review evaluates the latest completed non-review turn. Reviewing a review is allowed but not recommended; use a future explicit turn-selection feature if you need that workflow.
After a lightweight review, the user typically wants one of:
$claude-adversarial-review — get an independent second-opinion review$claude-result — re-read the original job output