菜单
Run cross-verification on an existing evaluation run and generate recheck_summary.md plus recheck_comparison.json.
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
Analyze game source code and write a concise implementation summary for downstream testing. Use when keypoint verification needs implementation-aware controls, physics, and timing mapping.
Run the full game-gen-eval pipeline in normal mode (state injection + visual interaction verification) from a generated game. Use when you want an end-to-end evaluation that produces summary_report.md and evaluation_report.md.
Recheck an existing normal keypoint evaluation artifact without rerunning the game.
Extract discrete, verifiable, specification-first keypoints from a game description for subsequent evaluation of game logic correctness. Use when preparing `games/{game_name}/keypoints.md` from a description file.
Generate a runnable TypeScript game from a game description file, plus data.md and a state injection API for evaluation. Use when converting `descriptions_example/*.md` into a playable game project under `games/`.
Convert fuzzy keypoint text into concrete short interaction data for executable verification. Use as the normal-mode `keypoint-tester` pre-step, not as a standalone top-level workflow.
| name | recheck-orchestrator |
| description | Run cross-verification on an existing evaluation run and generate recheck_summary.md plus recheck_comparison.json. |
This skill is the .codex entrypoint for rechecking a completed run.
It should use the stable direct script instead of rebuilding orchestration logic inline:
python3 scripts/run_recheck_direct.py \
--workspace "$(pwd)" \
--game-name "{game_name}" \
--run-id "{run_id}" \
--max-concurrent "{max_concurrent}"
The direct script launches fresh Codex sessions for each keypoint recheck, normalizes recheck_result.json, and writes the final comparison reports.
Parse parameters from the user request:
game_name (required)run_id (required)max_concurrent (optional, current direct runner still executes sequentially for stability)model (optional)reasoning_effort (optional, default low)step_timeout (optional, default 600)runs/{game_name}/{run_id}/summary_report.md exists.runs/{game_name}/{run_id}/recheck_summary.mdruns/{game_name}/{run_id}/recheck_comparison.jsonresult.json, test_result.json, or screenshots.python3 scripts/run_recheck_direct.py \
--workspace "$(pwd)" \
--game-name tetris \
--run-id run_tetris_normal_t300 \
--max-concurrent 1