Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

User Context Scanner

v0.7.0

用户画像扫描器v4:自动挖掘用户偏好、行为模式、兴趣图谱,支持实时上下文更新。触发条件:用户表达偏好、检测到行为模式、上下文更新请求。

0· 55·0 current·0 all-time
byLouis Z.@zhaoguoqiang-hub
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Suspicious
medium confidence
!
Purpose & Capability
The skill is a user-profile scanner and the code implements local scanning, evidence storage, confidence calculation and quiz generation; that matches the stated purpose. However documentation repeatedly claims "data only saved locally, not uploaded" while multiple configuration flags (publishPatternSignals, publishContradictionSignals, publishProfileUpdateSignals, publishQuizSignals, publishQuizSignals in config-loader) indicate the skill is designed to publish signals to an external 'proactive-engine' or other listeners. This discrepancy between 'local-only' claims and explicit publish flags is unexplained and concerning.
!
Instruction Scope
SKILL.md and examples only show running node scripts/scanner.js and processing signals, but the code scans ~/.openclaw/workspace/memory for all .md files and extracts personal data with regexes, then writes evidence and profile JSON under ~/.openclaw/workspace/.soul. That scanning can ingest arbitrary local content (potentially sensitive). The SKILL.md/collect-user-context.md emphasize privacy and local-only storage, yet the scanner code will process any files placed in the memory directory and the configuration allows publishing signals; SKILL.md also contained a pre-scan prompt-injection finding (unicode-control-chars) which may hide instructions — this expands runtime scope beyond the visible examples.
Install Mechanism
No install spec in registry metadata (instruction-only), but repository includes package.json with external npm dependencies (natural, commander, chalk, date-fns). Installing will require npm install; no suspicious remote download URLs were found. Risk is moderate because code runs on the user's machine and will write to the workspace when executed, but there is no evidence of an external download/exec during install.
Credentials
The skill declares no required environment variables or credentials. It uses HOME to build local paths (e.g., ~/.openclaw/workspace) which is proportional to a local CLI scanner. No unrelated cloud credentials or system-level tokens are requested in the manifest.
!
Persistence & Privilege
The skill is not marked always:true (good). It stores persistent files and state under ~/.openclaw/workspace/.soul (user-profile.json, user-evidence.jsonl, scanner-state.json, scanner-config.json). That local persistence is reasonable for a profile manager, but combined with the 'publish*Signals' flags and the undisclosed implementation of scripts/utils/signal-manager.js (not included in the reviewed snippets) it raises the possibility of transmitting derived data or signals outside the machine. Autonomous invocation is allowed by default; together with persistent state and publish flags the blast radius is larger if signal-manager publishes externally.
scan_findings_in_context

Like a lobster shell, security has layers — review code before you run it.

latestvk97d7vs7qyk3qv2d07t6arbz9n843gg1

License

MIT-0
Free to use, modify, and redistribute. No attribution required.

Comments