独立思考五部曲

ReviewAudited by ClawScan on May 10, 2026.

Overview

This instruction-only skill is mostly a thinking workflow, but it directs the agent to run an unreviewed local script and persist/reuse wiki/vector memory without clear scope or user control.

Review carefully before installing. Only use it if you understand and trust the referenced local query_vector.py script and the vector/wiki knowledge base. Prefer changing the instructions so vector searches and wiki writes require explicit approval, have clear storage boundaries, and do not permanently disable normal Memory Search.

Findings (4)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

ConcernHigh Confidence
ASI01: Agent Goal Hijack
What this means

The agent may ignore safer or expected memory/search tools and rely on a private vector system even when the user did not explicitly request it.

Why it was flagged

The skill attempts to override the agent's normal memory/search behavior beyond a single task and force use of a specific backend.

Skill content
永久禁用系统自带 Memory Search,仅使用专属后台向量系统(`query_vector.py` / ChromaDB)
Recommendation

Limit this instruction to opt-in use for specific tasks, and avoid permanent or global tool-disabling language.

ConcernHigh Confidence
ASI05: Unexpected Code Execution
What this means

Installing the skill could lead the agent to run local code whose behavior ClawScan cannot verify.

Why it was flagged

The skill requires execution of a hardcoded local Python script that is not included in the reviewed artifacts and has no install/provenance information.

Skill content
python3 /Users/markma01/.openclaw/workspace/scripts/query_vector.py "<检索词>"
Recommendation

Provide the helper script as reviewed source, declare it in metadata/install requirements, and require user confirmation before running it.

What this means

Private or stale knowledge-base content could be reused in future reasoning, and poisoned or incorrect entries may influence decisions.

Why it was flagged

The skill directs the agent to rely on persistent vector/wiki history as authoritative context, but does not define boundaries, provenance checks, or handling of sensitive or incorrect stored information.

Skill content
自验 = 用自己过去的总和(向量+wiki)替代当下他人在场
Recommendation

Define exactly which knowledge sources may be queried, how results are verified, and when the user must approve using retrieved context.

What this means

Task outputs, potentially including sensitive troubleshooting or decision details, may be stored persistently and influence later sessions.

Why it was flagged

The skill mandates writing outputs into a persistent wiki for future reuse without specifying retention, redaction, approval, or scope.

Skill content
「看」的输出必须闭环存入 wiki,为下次「想」提供素材
Recommendation

Make wiki writes explicit and user-approved, and document storage location, retention, redaction rules, and deletion controls.