recheck-orchestrator

name	recheck-orchestrator
description	Run cross-verification on an existing evaluation run and generate recheck_summary.md plus recheck_comparison.json.

Overview

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.

Input

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)

Required Behavior

1. Verify runs/{game_name}/{run_id}/summary_report.md exists.
2. Run the direct recheck script.
3. Report the generated artifact paths:
   • runs/{game_name}/{run_id}/recheck_summary.md
   • runs/{game_name}/{run_id}/recheck_comparison.json
4. Do not rerun Playwright.
5. Do not overwrite original result.json, test_result.json, or screenshots.

Example

python3 scripts/run_recheck_direct.py \
  --workspace "$(pwd)" \
  --game-name tetris \
  --run-id run_tetris_normal_t300 \
  --max-concurrent 1