メニュー
Run a CvC game and capture LLM-Python communication trace for analysis
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Diagnose the biggest CvC policy weakness from eval metrics and LLM-Python trace data
Run multi-episode CvC evaluation and produce structured JSON metrics
Implement one targeted CvC code change from analysis, verify improvement, submit if better
Install softmax-cli, authenticate via softmax login, and find or create a cogames player
| name | cogamer.play |
| description | Run a CvC game and capture LLM-Python communication trace for analysis |
Run a CvC game and capture trace data for analysis.
Announce at start: "I'm using the play skill to run a CvC game and capture the LLM-Python trace."
softmax cogames play -m <mission> -p class=cvc_policy.cogamer_policy.CvCPolicy --render=log --save-replay-file /tmp/cvc-replay.json.z
Defaults: mission=machina_1, steps=1000. Override via arguments passed to the skill.
The CvCPolicy writes an LLM-Python communication trace to /tmp/cvc-trace/. Each file is a JSON with:
agents: per-agent step count, LLM call count, final resource biasllm_trace: chronological list of every LLM call with prompt, raw response, parsed fields, latencyRead the trace after play to understand how the LLM and Python code interacted.
--mission <name> or -m (run softmax cogames play --help for options)--steps <n> or -s--render gui for visual mode, log for headless--seed <n> for reproducibility--save-replay-file <path> for replay dataRead /tmp/cvc-trace/*.json for the LLM communication trace. Use /cogamer.evaluate for multi-episode scoring or /cogamer.analyze to diagnose issues from this run.