CNC智能报价Workflow - 多Agent协作闭环。从黑盒24h到白盒10min。主赛道Workflow Hacker + 副赛道Agent Worker。

Summary of what to consider before installing/running: - Capability mismatch: The README/SKILL.md promises a 3-agent, fast workflow, but config.json sets production mode and the engine's logic will trigger Rule-Only fallback (pure rule math) — you will likely not get the advertised multi-agent behavior unless you change config or provide missing modules. - Missing components: agent2_rag expects case_retriever and risk_control modules and agent3 tries to optionally load UniSkill from an absolute local path; none of those are included. If you enable them, inspect their code before use. - Optional local path insertion: agent3_meta inserts '/home/admin/.openclaw/workspace/skills/uniskill-v4' into sys.path. If that path exists on your system it could cause the workflow to import and execute local code. Only enable/unpack third-party components you trust. - Bugs and robustness: There are small code issues (e.g., agent3 references Path without importing it) and version mismatches (config.json v2.0.1 vs skill v2.0.2). Test in an isolated environment first. Recommendations: 1. Run this in a sandbox or isolated environment (no sensitive data) to confirm actual behavior. 2. If you need the multi-agent pipeline, obtain and review the missing modules (case_retriever, risk_control, uniskill-v4) and verify their provenance and code before placing them on the filesystem where the skill can import them. 3. If you do not want the Rule-Only fallback, edit config.json (stability.mode) and understand the implications; changing it may cause the engine to attempt imports that are currently missing. 4. Contact the author/maintainer to clarify why production mode forces fallback and to request the missing components or an explanation of intended deployment. Given these inconsistencies, exercise caution and verify the runtime behavior before trusting this skill for production quoting.

Type: OpenClaw Skill Name: cnc-quote-workflow Version: 2.0.2 The bundle implements a multi-agent CNC pricing workflow consisting of input parsing, RAG-based retrieval, and meta-cognitive review. The code logic is consistent with the stated purpose of automating manufacturing quotes, using standard libraries like re, logging, and yaml. No evidence of data exfiltration, malicious execution, or harmful prompt injection was found; the hardcoded path in agent3_meta.py appears to be a localized integration attempt for a secondary skill rather than a malicious backdoor.