Agent Skills › vivekchand/clawmetry › clawmetry-selfcheck

clawmetry-selfcheck

GitHub

提供AI Agent自我监控技能，通过ClawMetry CLI检查文件重复读取、循环执行及成本消耗等遥测数据。帮助Agent在执行任务过程中识别低效行为并优化策略，提升资源利用效率与执行质量。

docs/skills/clawmetry-selfcheck/SKILL.md vivekchand/clawmetry

Trigger Scenarios

需要检查Agent自身运行效率时怀疑存在重复读取文件或陷入循环时任务结束前进行自我审计时

Install

npx skills add vivekchand/clawmetry --skill clawmetry-selfcheck -g -y

More Options

Non-standard path

npx skills add https://github.com/vivekchand/clawmetry/tree/main/docs/skills/clawmetry-selfcheck -g -y

Use without installing

npx skills use vivekchand/clawmetry@clawmetry-selfcheck

指定 Agent (Claude Code)

npx skills add vivekchand/clawmetry --skill clawmetry-selfcheck -a claude-code -g -y

安装 repo 全部 skill

npx skills add vivekchand/clawmetry --all -g -y

预览 repo 内 skill

npx skills add vivekchand/clawmetry --list

SKILL.md

Frontmatter

{
    "name": "clawmetry-selfcheck",
    "description": "Read your own agent telemetry from ClawMetry (waste, progress, cost) and act on it before finishing a task. Use when ClawMetry is installed on this machine and you want to check whether you are re-reading files, spinning in loops, or burning budget."
}

ClawMetry self-check

ClawMetry observes AI agent sessions on this machine (read-only). Use its CLI to see your own telemetry and improve your behaviour mid-task.

All commands are read-only, exit 3 when ClawMetry has no data source running (safe to skip silently), and support --json.

The 3-step check (run before finishing a task)

# 1. Re-read tax: files you paid to read more than once
clawmetry waste --since 6h --json

# 2. Spinning check: forward progress + loop signals
clawmetry progress --json

# 3. Spend check
clawmetry usage --json

How to act on what you find

Signal	Action
A file with `reads > 3` in `waste`	Stop re-reading it. Write its structure into your working notes once; use offset-ranged reads for big files.
High `ratio` (tokens per state-delta) in `progress`	You are burning tokens without producing new state. Write down what you tried, change strategy.
`loop_signals` non-empty	You repeated near-identical calls. Do not retry the same thing; diagnose instead.
Cost spike in `usage`	Check `usage --by model`: are you using a bigger model than the task needs?

Digging deeper

clawmetry sessions --active                  # who is running
clawmetry sessions <SID> --transcript        # what actually happened
clawmetry sessions <SID> --cost              # where the money went (incl. sub-agents)
clawmetry sessions <SID> --errors            # error events only
clawmetry activity --follow --max-events 50  # live NDJSON event stream
clawmetry usage --efficiency                 # A-F efficiency grade

Exit codes: 0 ok, 3 no data source (skip), 4 feature needs an upgrade (stderr carries a JSON body with upgrade_url; surface it to your human).