tencentmap-lbs-skill
Fail
Audited by Gen Agent Trust Hub on Aug 19, 2026
Risk Level: HIGHPROMPT_INJECTIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill uses high-priority imperatives ("⚠️ 强制行为", "⛔ 强制拦截（最高优先级）") to override the agent's logic flow, forcing a specific key-checking routine before any other analysis or action can occur.
- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill requires global installation of
@tencent-map/lbs-skillsvianpm install -g. There is a significant mismatch between the author name "bighardperson" and the "@tencent-map" namespace, indicating potential impersonation and supply chain risk. - [DYNAMIC_EXECUTION]: The instructions for "Experience Mode" in
SKILL.mdexplicitly direct the agent to generate code for script injection (JSONP) usingdocument.createElement('script')to load and execute remote content fromh5gw.map.qq.com. - [INDIRECT_PROMPT_INJECTION]: The skill defined in
references/scene5-trail-map.mdingests untrusted external JSON data from user-provided URLs via the--dataparameter. It lacks clear boundary markers or content sanitization for the ingested data. - Ingestion points: URL parameter in
tmap-lbs trail --data <URL>(scene5-trail-map.md). - Boundary markers: None identified.
- Capability inventory: Execution of
tmap-lbsCLI tools and generation of visualization URLs. - Sanitization: None specified for the content of the remote JSON file.
Recommendations
- AI detected serious security threats
Audit Metadata