← ClaudeAtlas

review-looplisted

Codex-native Stage 1 review-loop skill. Orchestrates planning and execution with a Codex Executor and Claude/Codex reviewer backends while sharing the .review-loop protocol with Claude Code.
NYTC69/review-loop · ★ 3 · Code & Development · score 76
Install: claude install-skill NYTC69/review-loop
# review-loop — codex orchestration Read `docs/protocol/loading.md`, then run: ```bash python3 <support-root>/scripts/read_protocol.py --runtime codex --stage entry-review-loop --output .review-loop/tmp/protocol-entry-review-loop.md ``` Resolve <support-root> to this plugin/repository, not the task workspace. Keep cwd in the user's workspace. After exit 0, read the complete output file in bounded chunks before acting; stdout contains only a compact hash/size receipt, not the instruction body. A missing, unreadable, or incompletely read file blocks the action. `docs/protocol/loading.json` is the action/prerequisite map for both runtimes. Detect fresh / plan-exists / code-exists / explicit-resume via the entry procedures. For fresh work run planning → planning-review; after approval continue execution → execution-review → gate → polish → docs → security → delivery. Do not stop after exec alone. Before initialization load `session-init` AND `review-loop-init`; before a resume load `resume` AND `review-loop-init`. Before any work-agent dispatch load the matching `planning` or `execution` bundle; before review load `planning-review` or `execution-review`. On planning approval, only the plan-only entry loads `plan-exit`. The umbrella retains its session/lock and continues directly into execution. Before a stage transition, retry, stop or error exit, load its applicable bundle per `loading.md`. Use `parallel-review` only in the Codex orchestrator for N>1 Claude-CLI reviewer job