Skill flagged — suspicious patterns detected
ClawHub Security flagged this skill as suspicious. Review the scan results before using.
Context Compactor
v1.0.0上下文压缩代理,采用分层压缩策略,基于内存使用触发机制。处理HOT/WARM/COLD三层数据,优化token使用。当用户需要压缩对话上下文、优化内存使用、管理会话历史、减少token消耗时使用此技能。特别适用于长时间对话、复杂任务处理、需要保留重要历史信息的场景。
⭐ 0· 247·0 current·0 all-time
by@geoshan
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
Name/description, configuration, and included code (compactor.py, hierarchical_compactor.py, monitor.py, integration.py, api_server.py, control scripts) align with a Context Compactor that monitors session memory, classifies HOT/WARM/COLD and compresses or archives items. No unrelated cloud credentials, binaries, or config paths are requested.
Instruction Scope
SKILL.md and scripts instruct the agent to monitor ~/.openclaw/workspace memory files, run background monitor/integration services, write logs and a local SQLite DB, and add cron/tasks or heartbeat entries. Those actions are consistent with the stated purpose but grant the skill read access to the user's OpenClaw workspace and persistent write access (logs, DB). The SKILL.md suggests cloning from a repository URL but the package already contains code; the placeholder git clone line is vague and could encourage fetching remote code if followed.
Install Mechanism
There is no remote install spec; the package is instruction-and-code-only and uses a small Python dependency (Flask) declared in requirements.txt. No external downloads, URL shorteners, or archive-extract steps were found in the provided files.
Credentials
The skill asks for no environment variables or external credentials (primary credential none). It does read files under the user's OpenClaw workspace (~/.openclaw/workspace/memory) which is appropriate for a memory-aware compactor, but that access could expose sensitive conversation history. config.json includes auto_start_with_openclaw:true, meaning it is designed to be started automatically if integrated—this increases its access surface.
Persistence & Privilege
The skill runs background processes (monitor, integration), persists a SQLite DB and logs, and includes start/stop scripts. 'always' is false and it does not force-enable itself in the registry, but its files and config indicate it is intended to be integrated (auto-start) with OpenClaw; that persistent presence is reasonable for a monitoring/self-managing service but should be reviewed by the operator.
What to consider before installing
This package is largely coherent with a context-compression agent, but review a few things before installing:
- Inspect integration.py and api_server.py to confirm whether any HTTP endpoints are exposed and whether they require authentication (Flask-based APIs often default to no auth). An unauthenticated API could leak conversation memory or compressed data.
- Search the code for any outbound network calls (requests, urllib, socket, or subprocess calls that curl/wget) to ensure data is not silently sent to external hosts. If you find any external endpoints, verify they are legitimate and appropriate.
- Confirm the service only reads OpenClaw workspace paths you expect (~/.openclaw/workspace). If you need tighter control, run the skill in an isolated environment (container or VM) or adjust file permissions and configuration to limit accessible paths.
- Because SKILL.md suggests cloning a repository but the package already contains code, avoid running any git clone or install commands from untrusted URLs; use the included files instead or obtain the upstream source from a known homepage. Lack of a homepage / unknown source lowers trust and is why this is flagged.
- If you allow the skill to auto-start or add cron/heartbeat entries, ensure you understand and control that integration (review heartbeats, cron entries) and monitor logs (logs/ directory) for unexpected behavior.
If you want, I can scan integration.py and api_server.py for network calls and any hard-coded hosts/keys, or point out exact lines to review for authentication and outbound requests.Like a lobster shell, security has layers — review code before you run it.
latestvk97eep2dma7vqj3zbt2w0pk6ph82me31
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
