Agent Introspection

Security checks across static analysis, malware telemetry, and agentic risk

Overview

This is an instruction-only self-reflection skill with no code, credentials, or install steps, though it includes guidance about using past reflection files and taking corrective action that users should understand.

This skill appears safe to install as an instruction-only reflection aid. Before using it, decide where any introspection notes should live and ensure your agent still asks before taking irreversible or external actions.

Static analysis

No static analysis findings were reported for this release.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Risk analysis

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.

What this means

If used in an agent that has powerful tools, this guidance could cause the agent to take corrective steps during reflection rather than only reporting its reasoning.

Why it was flagged

The skill contemplates a non-interactive scheduled mode and tells the agent to take a corrective action rather than only write a reflection. No tool or destructive action is specified, and this is aligned with the introspection purpose, but the action scope and approval expectations are not explicit.

Skill content
当自省由定时任务触发时(非用户实时交互)... 如果发现和之前重复,不要写新的反思,而是直接执行那个正确的动作。
Recommendation

Use this skill with normal tool-approval controls, and clarify that any irreversible, external, or high-impact action still needs explicit user confirmation.

What this means

Past reflection records may affect how the agent behaves in later tasks.

Why it was flagged

The skill instructs the agent to use prior introspection files and cross-day reflection history as context for future behavior. This is consistent with its purpose, but persistent notes can influence later actions if stale or inaccurate.

Skill content
读取过去的自省文件,检查历史发现。... 跨天的自省中,同一个问题是否反复出现?
Recommendation

Keep introspection records in a known location, avoid storing sensitive unrelated content in them, and treat prior notes as context to verify rather than unquestioned truth.