skill-trace-debugger

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Skill Trace Debugger

Surface trace and eval coverage gaps for repository skills before live behavioral runs.

Scope: Static trace readiness checks only. Does not execute live evals or harvest session logs.

在技能进行实时行为运行前，发现仓库中Skill的追踪和评估覆盖缺口。

范围： 仅进行静态追踪就绪性检查。不执行实时评估或收集会话日志。

Dispatch

调度

$ARGUMENTS	Action
`<name>`	Trace readiness report for one skill
`repo` / `--all`	Fleet trace readiness summary
`--format json`	Machine-readable report
Empty	Show trace signal checklist

$ARGUMENTS	操作
`<name>`	单个Skill的追踪就绪性报告
`repo` / `--all`	全仓库Skill的追踪就绪性汇总
`--format json`	机器可读格式的报告
Empty	显示追踪信号检查清单

Trace Readiness Checklist

追踪就绪性检查清单

Signal	Meaning
`evals/evals.json`	Behavioral cases exist for post-hoc grading
`scripts/check.py`	Portable validator contract is wired
explicit-invocation case	Eval includes a slash-command or named dispatch prompt
NOT-for boundary	Description names the correct alternate skill

Signal	含义
`evals/evals.json`	存在用于事后评分的行为测试用例
`scripts/check.py`	已连接可移植验证器契约
explicit-invocation case	评估包含斜杠命令或命名调度提示
NOT-for boundary	描述中指定了正确的替代Skill

Workflow

工作流程

bash

uv run python skills/skill-trace-debugger/scripts/trace_report.py <name>
uv run python skills/skill-trace-debugger/scripts/trace_report.py --all --format json

Route live trace grading to agent-eval-runner or maintainer CI workflows.

bash

uv run python skills/skill-trace-debugger/scripts/trace_report.py <name>
uv run python skills/skill-trace-debugger/scripts/trace_report.py --all --format json

将实时追踪评分路由至agent-eval-runner或维护者CI工作流。

Validation Contract

验证契约

bash

uv run python skills/skill-trace-debugger/scripts/check.py

bash

uv run python skills/skill-trace-debugger/scripts/check.py

Critical Rules

关键规则

Never claim live trace capture from static reports alone.
Treat missing explicit-invocation eval cases as a coverage gap, not a hard failure for internal skills.
Do not mutate eval manifests without maintainer approval.
Pair trace reports with skill-eval-scaffolder when adding new cases.
Keep prompts grounded in each skill's dispatch table.

切勿仅通过静态报告声称支持实时追踪捕获。
将缺失的显式调用评估用例视为覆盖缺口，而非内部Skill的严重故障。
未经维护者批准，不得修改评估清单。
添加新测试用例时，需将追踪报告与skill-eval-scaffolder配合使用。
确保提示与每个Skill的调度表保持一致。