CtrlK
BlogDocsLog inGet started
Tessl Logo

code-as-harness

检测摩擦信号或陌生任务 → 搜证据确认重复 → 诊断根因 → 用代码修已有 harness 或建新 harness。 两种模式:Fix(确认重复后 → 写 hook/lint/guard)+ Build(反复出现的新任务类型 → Agent Team Leadership 规划新 skill/tool/流程)。 Use when: operator表达不满且搜证据确认历史上确实重复出现过同类问题(不是字面匹配"又")、 连续 cancel 工具调用、收到反复出现的陌生任务类型且无对应 skill、 自己撞到工具/runtime 摩擦需要上报(雨刮器条款细则见正文"猫侧主动上报"节)。 Not for: 一次性批评(搜证据未发现重复)、玩笑式"笨猫"(后跟哈哈哈)、 有明确 error message 的首次代码 bug(用 debugging)、reviewer P1/P2 反馈(用 receive-review)、 一次性新任务(直接做,不建 harness)。 Output: Rich block 诊断卡(根因 + 证据 + 建议)+ 可选 F128 新 thread 提议(平行修复不打断当前任务)。 GOTCHA: 不是每次被批评都弹诊断卡——必须先搜证据确认重复,才进入诊断流程。 过度触发 = 猫在逃避批评。一次性陌生任务直接做不建 harness,只有反复出现才沉淀。

67

Quality

81%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Low

Low-risk findings worth noting

SKILL.md
Quality
Evals
Security

Quality

Content

77%Weight 40%Scale 1-3

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A highly actionable body with concrete tools, copy-paste templates, and a well-sequenced phased workflow. It loses points on conciseness (heavy rule repetition) and progressive disclosure (no reference files despite sizable self-contained sub-topics).

Suggestions

Deduplicate the "先搜证据确认重复" and "一次性新任务直接做不建 harness" rules — state each once in 核心原则 and let tables/Phase steps reference rather than re-explain it.

Extract the 雨刮器条款细则 and the taste-signal path into reference files (e.g. references/friction-reporting.md, references/taste-signals.md) and link to them one level deep, keeping SKILL.md as an overview.

Consolidate the 不触发/灰区/Common Mistakes tables — several rows restate the same distinction (one-off vs repeated), which could be merged to save tokens.

DimensionReasoningScore

Conciseness

The content is non-fluffy and assumes Claude's competence, but the same core rules are restated many times — "先搜证据确认重复" recurs in 核心原则, the 触发判定铁律, 判定流程, Phase 1, and two Common Mistakes rows; "一次性新任务直接做不建 harness" repeats across 核心原则, 灰区, Build前置闸门, and Common Mistakes — so it could be tightened.

2 / 3

Actionability

Provides concrete tool calls (search_evidence, list_threads, cat_cafe_create_rich_block, cat_cafe_propose_thread), exact file paths (docs/taste/vignettes/, private/taste/), and copy-paste-ready YAML rich-block and F128 initialMessage templates — fully executable guidance.

3 / 3

Workflow Clarity

A clear Phase 1→5 sequence with an explicit pre-gate (搜证据 → YES/NO), a quantification step, a root-cause classification table, and a Common Mistakes table serving as error-recovery feedback, plus downstream routing to request-review/merge-gate.

3 / 3

Progressive Disclosure

Sections are well-organized with headers, but at ~260 lines with no bundle files and zero external references, self-contained sub-flows (雨刮器条款细则, the taste-signal path, the F128 template) are inline when they would benefit from being split into one-level-deep reference files.

2 / 3

Total

10

/

12

Passed

Description

85%Weight 40%Scale 1-3

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong, specific description with explicit Use-when/Not-for/Output routing and good conflict avoidance. Its only weakness is that the most natural verbatim trigger phrases sit in the triggers list rather than the description field itself.

Suggestions

Fold 2–3 of the most natural verbatim trigger phrases (e.g. "又忘了", "每次都", "怎么还") directly into the description's Use-when clause so the trigger_term_quality is fully self-contained.

Trim the parenthetical routing details (e.g. "雨刮器条款细则见正文") from the description — they add tokens without aiding triggering.

DimensionReasoningScore

Specificity

Lists multiple concrete actions — "搜证据确认重复 → 诊断根因", "写 hook/lint/guard", "Agent Team Leadership 规划新 skill/tool/流程", and a concrete output ("Rich block 诊断卡(根因 + 证据 + 建议)+ 可选 F128 新 thread 提议"), matching the multiple-specific-actions anchor.

3 / 3

Completeness

Explicitly answers what (the action chain), when (a multi-condition "Use when:" clause), when not ("Not for:" with examples), and output ("Output:"), plus a GOTCHA caveat — clearly both what and when.

3 / 3

Trigger Term Quality

The description frames signals indirectly ("operator表达不满", "连续 cancel 工具调用", "反复出现的陌生任务类型") and quotes a few natural fragments ("又", "笨猫"), but the richest verbatim user phrases ("又忘了", "每次都") live in the separate triggers list, so coverage is partial rather than comprehensive.

2 / 3

Distinctiveness Conflict Risk

The "Not for:" clause routes away from sibling skills (debugging for first-time code bugs, receive-review for reviewer P1/P2, one-off new tasks), carving a clear niche of evidence-confirmed repeated friction unlikely to trigger the wrong skill.

3 / 3

Total

11

/

12

Passed

Validation

93%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

frontmatter_unknown_keys

Unknown frontmatter key(s) found; consider removing or moving to metadata

Warning

Total

15

/

16

Passed

Repository
zts212653/clowder-ai
Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.