中文 自我改进代理

PassAudited by VirusTotal on May 8, 2026.

Overview

Type: OpenClaw Skill Name: chinese-self-improving-agent Version: 1.0.0 The skill bundle provides a framework for an AI agent to perform self-reflection and log learnings, errors, and feature requests to local files. It includes utility scripts (scripts/activator.sh, scripts/error-detector.sh, scripts/extract-skill.sh) and OpenClaw hooks (hooks/openclaw/handler.js) designed to inject reminders into the agent's context and automate the creation of new skill scaffolds. The code includes safety warnings against logging secrets and implements basic path validation in the extraction script to prevent directory traversal. No evidence of data exfiltration, malicious execution, or harmful prompt injection was found.

Findings (0)

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

Incorrect, overbroad, or sensitive notes could be reused later and influence the agent’s future behavior.

Why it was flagged

The skill explicitly asks the agent to add learnings to persistent memory, which can affect future sessions.

Skill content
4. 自我组织 → 更新记忆 → 永久改进 ... 将新知识添加到 memory.md
Recommendation

Keep memory entries short and redacted, review persistent learnings periodically, and require user approval before promoting notes into long-term guidance files.

NoteHigh Confidence
ASI10: Rogue Agents
What this means

If enabled, future sessions may be nudged to log learnings and update persistent files even when the user did not mention the skill in that session.

Why it was flagged

The optional hook can inject a recurring reminder into the agent context at bootstrap.

Skill content
Injects self-improvement reminder during agent bootstrap ... events: ["agent:bootstrap"]
Recommendation

Enable the hook only in workspaces where persistent learning is desired, and disable it if the reminders become intrusive.

What this means

Cross-session sharing can expose private context if used carelessly.

Why it was flagged

The integration guide documents cross-session transcript reading and messaging, while also limiting it to explicit user need and sanitized summaries.

Skill content
sessions_history ... Read transcript from another session ... Only read another session's transcript when the user explicitly wants shared context
Recommendation

Use inter-session tools only with explicit user consent and share concise, redacted summaries rather than raw transcripts.

What this means

Running the helper changes the local workspace by adding a new skill directory.

Why it was flagged

The helper script can create new local skill scaffold files, but it validates the skill name and prevents absolute paths or '..' path segments.

Skill content
mkdir -p "$SKILL_PATH" ... cat > "$SKILL_PATH/SKILL.md"
Recommendation

Use the documented --dry-run option first and review generated skill files before relying on them.

What this means

The mismatch makes provenance and package lineage less clear, even though the included code appears coherent.

Why it was flagged

The internal metadata does not match the supplied registry metadata, which lists slug 'chinese-self-improving-agent' and version 1.0.0.

Skill content
"slug": "self-improving-agent", "version": "3.0.16"
Recommendation

Verify the publisher and expected package identity before installing, especially if relying on the optional hook scripts.