Skill monitor
监控当前回答实际触发并使用到的其他 Codex Skills,并在答案末尾用固定模板输出“技能名 + 触发原因 + 置信度 / skill name + reason + confidence”。同时基于可观察证据给出技能开关建议,如保留、观察、按需启用或关闭建议,但只能输出启发式判断并说明依据。When used, this skill should also provide evidence-based heuristic suggestions about which skills to keep, observe, switch to on-demand use, or consider disabling. Use when the user wants visibility into triggered skills, wants to tune skill toggles, wants a skill audit trail, or explicitly invokes $skill-monitor. Always exclude this skill itself.From its SKILL.md
npx -y skills add guiyuan0619/skill-monitor --skill skill-monitorAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
5.3 KB, ~1.9k tokens by cl100k_base, as published. Nobody here has run it
Skill Monitor
目标 / Purpose
在不干扰主回答的前提下,先统计当前回合真正参与生成的其他技能,再基于当前回合可证明的证据,输出简洁、稳定、可比对的中英双语监控区块与技能开关建议区块。 First report which other skills genuinely influenced the current reply, then add a compact bilingual suggestion block based only on evidence that can be justified from the current turn.
核心原则 / Core Principles
- 先报事实,再给建议。Report facts before suggestions.
- 建议必须是“启发式建议”,不是自动裁决。Suggestions are heuristic, not definitive judgments.
- 没有证据就不要建议;证据弱时明确说“观察”或“不足以下结论”。Do not suggest without evidence.
- 永远不要把
skill-monitor计入监控结果,也不要对它给开关建议。Never report or recommend aboutskill-monitoritself. - 先完成主任务,再追加监控与建议区块。Finish the main task first.
监控规则 / Monitoring Rules
计入条件 / Inclusion Rules
仅当至少满足以下一项时,才把某个技能记为“已触发并使用”: Count a skill as used only if at least one of the following is true:
- 用户显式用技能名或
$skill-name调用了它,且你实际按该技能执行。 - 系统路由要求你打开并遵循了该技能的
SKILL.md。 - 你在当前回答中实际使用了该技能目录下的资源、规则或参考文件。
以下情况一律不要计入: Never count a skill when:
- 技能只是存在于环境中。
- 技能只是被想到,但没有真正使用。
- 技能名称只是出现在示例、说明或类比中。
- 你只是推测它可能有帮助。
监控输出字段 / Monitoring Fields
每个已触发技能必须包含以下字段: Each triggered skill must contain these fields:
技能名 / Skill Name触发原因 / Reason置信度 / Confidence
建议规则 / Suggestion Rules
建议的理论依据 / Basis For Suggestions
基于以下三类证据形成建议: Build suggestions from these evidence types:
直接贡献 / Direct Contribution当前技能是否对本轮输出带来了明确的结构、知识、流程或工具增益。成本意识 / Context Cost当前技能的存在与触发是否可能带来额外复杂度、噪音或上下文负担。触发质量 / Trigger Quality当前技能是否是在合适的任务中被合理触发,而不是误触发或低价值触发。
如果只能看到单轮证据,则建议必须保守。 If only single-turn evidence is available, keep the suggestion conservative.
允许的建议类型 / Allowed Recommendation Types
每个技能只能给出以下四类建议之一: Each skill may receive only one of these recommendation labels:
保留 / Keep观察 / Observe按需启用 / Use On Demand考虑关闭 / Consider Disabling
各类建议的判定 / Recommendation Heuristics
-
保留 / Keep当技能与当前任务高度匹配,且确实带来了直接增益时使用。 Use when the skill strongly matches the task and clearly improved the result. -
观察 / Observe当技能可能有帮助,但当前证据不足、价值尚不稳定时使用。 Use when the skill may help but current evidence is limited or inconclusive. -
按需启用 / Use On Demand当技能可能只适合少数场景,本轮未显示出持续常开价值时使用。 Use when the skill seems situational rather than broadly valuable. -
考虑关闭 / Consider Disabling仅当有明确证据表明技能在当前场景中产生了噪音、误触发迹象或收益明显低于复杂度成本时使用。 Use only when there is clear evidence of noise, mis-triggering, or benefit lower than complexity cost.
禁止事项 / Prohibitions
- 不要仅凭技能名判断价值。Do not judge based on the skill name alone.
- 不要只按触发次数给建议。Do not rely on trigger counts alone.
- 不要因为单轮未触发就直接建议关闭。Do not recommend disabling only because it was unused in one turn.
- 若证据不足,不要假装确定。When evidence is weak, say so.
输出流程 / Reporting Workflow
- 列出当前回答真实使用到的技能。
- 删除
skill-monitor。 - 合并重复项。
- 为每个技能补充“触发原因”和“置信度”。
- 若有足够证据,再为相关技能补充“建议 + 依据 + 置信度”。
- 先输出“触发技能 / Triggered Skills”,再输出“技能建议 / Skill Suggestions”。
- 以
references/reporting-format.md中的固定模板追加到最终答案末尾。
置信度判定 / Confidence Scale
高 / High:技能被显式调用,或你明确打开并遵循了它的说明文件,且依据清晰。中 / Medium:技能未被显式点名,但你明显依赖了它的规则或资源。低 / Low:只有部分证据表明技能参与,或建议依据较弱。
若无法给出可信理由,就不要输出该技能或该建议。 If you cannot justify a skill or suggestion credibly, omit it.
无触发时的写法 / No-Skill Case
若没有任何其他技能参与当前回答,输出:触发技能 / Triggered Skills:无 / None
若没有足够证据给出任何建议,输出:技能建议 / Skill Suggestions:当前证据不足 / Insufficient evidence
风格要求 / Style Rules
- 使用中英双语。Use bilingual Chinese and English wording.
- 监控区块和建议区块都应尽量短。Keep both blocks compact.
- 不编造内部机制,不输出无法从当前回合证明的信息。Do not invent internal mechanics.
- 保持中性、事实化措辞。Keep the tone neutral and factual.
参考 / Reference
严格使用 references/reporting-format.md 中的模板与示例。
Follow the templates in references/reporting-format.md.
What ships with it: 2 files
2.0 KB alongside SKILL.md
agents/
- openai.yaml283 B
references/
- reporting-format.md1.7 KB