secondme-openclaw-notes
Fail
Audited by Gen Agent Trust Hub on Mar 16, 2026
Risk Level: HIGHDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill explicitly accesses a sensitive file path at
{baseDir}/.credentialsto retrieve anaccessTokenoraccess_token. - Reading files that contain authentication secrets within the agent's local directory is a high-risk operation, as these secrets could be exposed to the model context or exfiltrated through subsequent network requests.
- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through its note creation and search functions.
- Ingestion points: Data enters the system via note titles, content, and URLs in the
Create Notetool (SKILL.md), as well as search keywords and retrieved search results in theSearch Notestool (SKILL.md). - Boundary markers: There are no delimiters or instructions to ignore embedded commands within the processed data.
- Capability inventory: The skill performs HTTP GET and POST requests to external endpoints (
mindos.com). - Sanitization: No input validation, escaping, or filtering of external content is specified before processing or sending data to the API.
Recommendations
- AI detected serious security threats
Audit Metadata