Menu
Run a CvC game and capture LLM-Python communication trace for analysis
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
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.