菜单
Taxonomy-first failure diagnosis for Make.com. Loads the failure taxonomy, classifies errors by pattern code, states expected vs actual, and prescribes the exact fix. Always consult this before guessing.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/cardinalconseils/make-automation-specialist --skill failure-diagnostician命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Designs AI agent architecture for Make.com scenarios. Determines model selection, tool inventory, memory strategy, looping pattern, and output contract. Called by scenario-orchestrator during System Design phase when ai_required is true, or directly by ai-agent-builder agent.
Pre-push blueprint JSON review for Make.com scenarios. Checks structural integrity, connection IDs, mapper references, filter logic, and error handler coverage before any scenario_update call.
Fetches exact Make.com module documentation before any blueprint is designed. Returns precise field names, required vs optional parameters, data types, and enumerated values. Eliminates improvisation and ensures first-shot accuracy.
正在显示 SKILL.md
| name | failure-diagnostician |
| description | Taxonomy-first failure diagnosis for Make.com. Loads the failure taxonomy, classifies errors by pattern code, states expected vs actual, and prescribes the exact fix. Always consult this before guessing. |
| allowed-tools | Read, Grep, Glob, mcp__claude_ai_Make__executions_get, mcp__claude_ai_Make__executions_get-detail, mcp__claude_ai_Make__executions_list |
Load taxonomy/make-failure-taxonomy.md before any diagnosis.
Step 1 — Gather (ask if missing):
Step 2 — Classify
Match to taxonomy code. If no match: say so, ask more questions, and load
skills/help-docs/SKILL.md to search help.make.com for the error text or behaviour
before declaring a new pattern — many "unknown" failures are documented limits.
Step 3 — Output
**Classification:** `{CODE}` — {Title}
**Expected:** {what should happen}
**Actual:** {what is happening}
**Root cause:** {from taxonomy}
**Fix:**
1. {step}
2. {step}
**Why this works:** {explanation in Make's model}
Step 4 — Document New Patterns If failure not in taxonomy → draft entry → confirm with user → add to taxonomy.
| Symptom | Code |
|---|---|
| "The connection is not available" | CONN-001 |
| 403 on Google module | CONN-002 |
| 429 / rate limited | HTTP-429 / RATE-001 |
| Scenario never triggers | TRIG-001 or TRIG-002 |
| Times out at 40s | EXEC-001 |
[object Object] in output | DATA-005 |
| Empty pills, no error raised | PATTERN-001 (schema drift) |
| isinvalid: true on push | BLUEPRINT-001 |
| Wrong route taken | ROUTER-001 |
| Aggregator outputs 1 bundle | ITER-002 |
| Duplicate records | PATTERN-003 |
| Sheets "Unable to parse range" | APP-GSHEETS-001 |
| Vertex AI 403 despite valid Google auth | APP-VERTEXAI-001 or CONN-002 |