菜单
IntelligentDesign improvement iteration. Directly analyzes code and prompts, identifies a specific improvement, implements it, tests, and submits. Use when asked to "review the code", "improve the prompts", or "intelligent design".
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
Generate an HTML training dashboard from .coach/ state showing experiments, scores, learnings, TODOs. Pulls live tournament status. Opens in browser. Use for "dashboard", "training status", "show progress".
Save current coaching session state. Writes log, updates state.json and todos.md, commits and pushes if git is configured. Use at the end of each improvement cycle.
Run one full coaching step — start session, choose approach (PCO or IntelligentDesign), improve, save session. This is the skill that /loop calls every 30 minutes. Use when asked to "run a coaching step" or "coach step".
PCO improvement iteration. Plays a game, collects experience, runs CvCLearner to propose program patches, tests locally, and submits if improved. Use when asked to "run PCO" or "evolve programs".
Run one improvement iteration. Chooses between PCO and IntelligentDesign based on recent effectiveness, runs the chosen approach, falls back to the other if it fails. Use when asked to "improve the agent", "run a coaching session", or "coach improve".
Start the coaching loop. Initializes state, then runs /coach.step every 30 minutes. Use when asked to "start coaching", "coach the agent", or "start the coaching loop".
| name | coach.improve.design |
| description | IntelligentDesign improvement iteration. Directly analyzes code and prompts, identifies a specific improvement, implements it, tests, and submits. Use when asked to "review the code", "improve the prompts", or "intelligent design". |
Directly read the code and prompts, identify a specific improvement, implement it, test, submit.
Reads architecture and alpha.0 reference from IMPROVE.md.
Run eval from IMPROVE.md on seed 42 to establish baseline score.
Pick ONE focus area based on IMPROVE.md strategies, .coach/todos.md, and what hasn't been tried:
agent/main.py, roles.py, targeting.py, pressure.py). Look for bugs, inefficiencies, or gaps vs the alpha.0 reference in IMPROVE.md_build_analysis_prompt() and _parse_analysis() in programs.py. Is the LLM seeing the right info? Could it return more than just resource_bias? Could it detect stagnation like alpha.0 does?helpers/targeting.py. Are aligner_target_score and scramble_target_score well-tuned? Compare weights vs alpha.0helpers/types.py and pressure.py against alpha.0 (e.g. RETREAT_MARGIN 15 vs 20, enemy AOE radius 4 vs 20)cvc_policy.py. Is the LLM feedback loop working? Could the analyze program influence more than mining? Could it adjust role allocation or targeting?Make a focused, isolated change. Write the code directly.
_build_analysis_prompt() or _parse_analysis() in programs.pyagent/helpers/types.py or the relevant mixinprograms.py and wire into all_programs()Run eval across 5+ seeds. If average score drops vs baseline, revert.
Use the submit command from IMPROVE.md. Log the submission version.
Report back with:
improved: whether scores improvedscore_before: baseline averagescore_after: post-change average (or null if reverted)focus: which area was analyzed (code/prompt/scoring/parameter/architecture)description: what changed and why