菜单
Show the state of the benchmark — pinned projects, which of the six tasks each has, and how many evaluation reports exist per task. Use when the user says "list the bench", "what projects do we have", "show benchmark coverage", "what's been evaluated".
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
| name | list-bench |
| description | Show the state of the benchmark — pinned projects, which of the six tasks each has, and how many evaluation reports exist per task. Use when the user says "list the bench", "what projects do we have", "show benchmark coverage", "what's been evaluated". |
| version | 1.0.0 |
Inventory the benchmark: projects, task coverage, and report counts.
Usage: /list-bench
Run the inventory helper:
lib/inventory.py
It prints each project (with its pinned short commit), the N/6 tasks
authored, a per-task ✓/· grid, and how many reports exist per task.
Summarize for the user: which projects are fully authored (6/6), which tasks
are missing, and which (project × task) cells still have no reports.
Suggest the next action — /create-task for missing tasks or
/evaluate-agent for un-evaluated ones.
If asked to aggregate scores (e.g. "average correctness per model"), read the
relevant reports/*/metrics.json directly and roll up along
run.model / run.project / run.difficulty / run.kind.
Author a benchmark task (prompt + expected solution) for a pinned project. Use when the user says "create a task", "make a hard task on <project>", "add a scout/coder challenge", or asks to design an easy/medium/hard evaluation. Studies the repo, designs a difficulty-calibrated task, and writes task.md, solution.md, and task.json.
Run a coding agent blind against a benchmark task, review its work, score it on every metric, and write a report. Also runs stateful, resumable multi-task campaigns ("fully eval <model>") that tick a durable checklist case-by-case in serial order. Use when the user says "evaluate <model> on <task>", "fully eval <model>", "run the benchmark", "score an agent", "test <model> against <project>", "resume the eval". Spins up a throwaway git worktree in /tmp/work, dispatches the agent (pi-as-mcp by default), waits serially, then produces reports/<timestamp>-<model>/ with report.md, metrics.json, frozen request+solution, and diff.zip.
Import and pin a codebase into the benchmark. Use when the user says "import repo", "add a project to the bench", "pin <repo>", or gives a git URL/path to benchmark against. Clones the repo at a frozen commit, writes its SUMMARY.md orientation doc and project.json, and registers it in manifest.json.