{"id":"187c9ba8-308a-4a6d-905e-da9979da90d3","slug":"clawhub-graysilver-llm-workflow-diagnoser","name":"Llm Workflow Diagnoser","description":"Use this skill whenever a user wants to evaluate whether an existing offline / reusable workflow is worth converting into an LLM-driven workflow. Triggers on...","canonicalUrl":"https://www.xpersona.co/agent/clawhub-graysilver-llm-workflow-diagnoser","sourceUrl":"https://clawhub.ai/graysilver/llm-workflow-diagnoser","homepage":"https://clawhub.ai/graysilver/skills/llm-workflow-diagnoser","source":"CLAWHUB","vendor":{"slug":"clawhub","label":"Clawhub","url":"https://clawhub.ai/graysilver/skills/llm-workflow-diagnoser"},"protocols":["OPENCLEW"],"capabilities":[],"trustScore":null,"trustConfidence":"unknown","artifactCount":0,"benchmarkCount":0,"lastRelease":"0.1.0","freshnessAt":"2026-10-09T09:47:39.182Z","freshnessLabel":"Oct 9, 2026","securityReviewed":true,"openapiReady":false,"stats":[{"label":"Trust score","value":"Unknown"},{"label":"Compatibility","value":"OpenClaw"},{"label":"Freshness","value":"Oct 9, 2026"},{"label":"Vendor","value":"Clawhub"},{"label":"Artifacts","value":"0"},{"label":"Benchmarks","value":"0"},{"label":"Last release","value":"0.1.0"}],"factsPreview":[{"factKey":"vendor","category":"vendor","label":"Vendor","value":"Clawhub","href":"https://clawhub.ai/graysilver/skills/llm-workflow-diagnoser","sourceUrl":"https://clawhub.ai/graysilver/skills/llm-workflow-diagnoser","sourceType":"profile","confidence":"medium","observedAt":"2026-10-09T09:47:39.182Z","isPublic":true},{"factKey":"protocols","category":"compatibility","label":"Protocol compatibility","value":"OpenClaw","href":"https://www.xpersona.co/api/v1/agents/clawhub-graysilver-llm-workflow-diagnoser/contract","sourceUrl":"https://www.xpersona.co/api/v1/agents/clawhub-graysilver-llm-workflow-diagnoser/contract","sourceType":"contract","confidence":"medium","observedAt":"2026-10-09T09:47:39.182Z","isPublic":true},{"factKey":"traction","category":"adoption","label":"Adoption signal","value":"3.1K downloads","href":"https://clawhub.ai/graysilver/llm-workflow-diagnoser","sourceUrl":"https://clawhub.ai/graysilver/llm-workflow-diagnoser","sourceType":"profile","confidence":"medium","observedAt":"2026-10-09T09:47:39.182Z","isPublic":true},{"factKey":"latest_release","category":"release","label":"Latest release","value":"0.1.0","href":"https://clawhub.ai/graysilver/llm-workflow-diagnoser","sourceUrl":"https://clawhub.ai/graysilver/llm-workflow-diagnoser","sourceType":"release","confidence":"medium","observedAt":"2026-07-23T15:18:18.015Z","isPublic":true},{"factKey":"handshake_status","category":"security","label":"Handshake status","value":"UNKNOWN","href":"https://www.xpersona.co/api/v1/agents/clawhub-graysilver-llm-workflow-diagnoser/trust","sourceUrl":"https://www.xpersona.co/api/v1/agents/clawhub-graysilver-llm-workflow-diagnoser/trust","sourceType":"trust","confidence":"medium","observedAt":null,"isPublic":true}],"highlights":["3.1K downloads","Trust evidence available"],"agentCard":{"name":"Llm Workflow Diagnoser","description":"Use this skill whenever a user wants to evaluate whether an existing offline / reusable workflow is worth converting into an LLM-driven workflow. Triggers on...","source":"CLAWHUB","sourceId":"clawhub:s17c6q86pekf2pm4rzv7z5wrsd8b36w8:llm-workflow-diagnoser","homepage":"https://clawhub.ai/graysilver/skills/llm-workflow-diagnoser","repository":"https://clawhub.ai/graysilver/llm-workflow-diagnoser","documentation":"https://www.xpersona.co/agent/clawhub-graysilver-llm-workflow-diagnoser","protocols":["OPENCLEW"],"examples":[{"kind":"example","language":"text","snippet":"# 诊断说明\n- 一句话说明：跨平台文案生成是高容错 + 多样化输出的典型 LLM 全量改造场景，可压到月级回收。\n- 改造步骤：\n  - 步骤 1：固化「产品要点 + 受众标签 → 三平台文案」的 prompt 模板\n  - 步骤 2：用最近 4 周历史素材做对照验证，统计一次通过率\n  - 步骤 3：灰度 50% 流量到 LLM 路径，老流程并行兜底\n  - 步骤 4：监控互动率后下线老流程\n- ROI 分数：8/10（10为满分）\n- 改造关键：\n  - 关键 1：受众多样化 + 文案允许多版本，LLM 能直接打主力\n  - 关键 2：互动率可量化衡量改造效果\n  - 关键 3：人工兜底成本可接受\n- 下一步建议：\n  - 建议 1：先跑 1 周端到端验证\n  - 建议 2：建立「一次通过率 / 互动率」两道监控\n  - 建议 3：法务相关文案保留人工复核\n\n# LLM 介入边界\n- 硬护栏护住：无\n- 仅部分场景适用：法务 / 行业合规相关文案，需人工复核\n- 全量适用：跨平台改写、文案初稿、配图建议、CTA 优化\n- AI 可护住：生成（多平台文案）、推理（CTA 排序）、设计（多版本对比）、检索（历史高互动素材）\n\n# 改造难度评级\n\n维度\t｜ 难度值（xx/10，10为满分）｜解释\n输入多样性\t｜ 3/10（10为满分）｜ 产品要点半结构化，受众标签可控\n输出一致性要求\t｜ 2/10（10为满分）｜ 多平台风格本身要求多样化\n容错率\t｜ 2/10（10为满分）｜ 文案可人工润色，可回滚\n决策密度\t｜ 4/10（10为满分）｜ 需判断\"哪个角度最能打\"\n可验证性\t｜ 3/10（10为满分）｜ 有 CTR / 互动率数据\n\n# 落地 MVP 路线图\n- MVP 步骤 1：拿最近 1 周产品更新，让 LLM 生成 3 平台 × 5 版文案，人工对比\n- MVP 步骤 2：运营在新流程上跑 2 周，老流程并行作对照\n- MVP 步骤 3：下线老流程，保留人工兜底通道与互动率监控\n\n# 风险清单\n- 风险 1：LLM 文案风格漂移 → 缓解：固化 prompt + 风格样例库\n- 风险 2：互动率不升反降 → 缓解：保留老流程做 4 周 A/B 对照\n- 风险 3：法务合规文案误用 → 缓解：合规相关字段强制人工审核"},{"kind":"example","language":"text","snippet":"# 诊断说明\n- 一句话说明：高频交易信号同时命中实时 + 高精度 + 高合规三个硬护栏，LLM 不能进生产路径。\n- 改造步骤：\n  - 步骤 1：保持原实时信号生成脚本不变\n  - 步骤 2：把 LLM 隔离在盘后复盘场景，不进生产路径\n  - 步骤 3：用 LLM 做异常信号归因与策略回顾\n- ROI 分数：1/10（10为满分）\n- 改造关键：\n  - 关键 1：实时性 + 零容错 + 高合规三大硬护栏同时命中\n  - 关键 2：LLM 推理延迟不可能压到毫秒级\n  - 关键 3：错信号直接进生产路径，无法批量回滚\n- 下一步建议：\n  - 建议 1：放弃「让 LLM 直接出信号」的想法\n  - 建议 2：把 LLM 用在盘后复盘 / 异常归因\n  - 建议 3：保留当前脚本与回放框架，不引入 LLM 链路\n\n# LLM 介入边界\n- 硬护栏护住：实时信号生成、毫秒级决策、生产路径写入\n- 仅部分场景适用：盘后复盘报告、异常信号归因\n- 全量适用：无\n- AI 可护住：检索（复盘历史）、推理（归因假设），但都不可直接进生产路径\n\n# 改造难度评级\n\n维度\t｜ 难度值（xx/10，10为满分）｜解释\n输入多样性\t｜ 2/10（10为满分）｜ 行情字段固定\n输出一致性要求\t｜ 9/10（10为满分）｜ 信号必须严格一致\n容错率\t｜ 10/10（10为满分）｜ 错信号直接进生产路径\n决策密度\t｜ 5/10（10为满分）｜ 信号生成有规则\n可验证性\t｜ 2/10（10为满分）｜ 有完整回测与回放框架\n\n# 落地 MVP 路线图\n- MVP 步骤 1：明确把 LLM 隔离在盘后复盘场景\n- MVP 步骤 2：用 LLM 做异常信号归因，输出归因假设给人工审核\n- MVP 步骤 3：不引入任何进生产路径的 LLM 链路\n\n# 风险清单\n- 风险 1：把 LLM 引入实时链路造成延迟 → 缓解：明确隔离边界\n- 风险 2：复盘归因被误用为生产信号 → 缓解：复盘输出必须经人工审核才可入决策\n- 风险 3：误判 ROI 偏高 → 缓解：以实时性 / 合规为否决项"}]}}