NYC

wisdom-accountability-coach

Fail

Audited by Gen Agent Trust Hub on Feb 15, 2026

Risk Level: HIGHPROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • Indirect Prompt Injection (HIGH): The skill is highly vulnerable to instructions embedded in data retrieved from the web.\n
  • Ingestion points: Uses mcp__firecrawl__firecrawl_search, mcp__brave-search__brave_web_search, and WebFetch in SKILL.md to ingest external content.\n
  • Boundary markers: Absent. There are no instructions to delimit or ignore instructions found in external data.\n
  • Capability inventory: Extensive file access tools Read, Write, Edit, and TodoWrite are permitted in SKILL.md.\n
  • Sanitization: Absent. No filtering or validation of web content is performed before processing.\n
  • Risk: An attacker-controlled website could include hidden prompts that command the agent to read sensitive local files and exfiltrate their contents via search queries or fetch parameters.\n- Data Exposure (MEDIUM): The skill is designed to maintain a detailed corpus of 'life, work, writings, and conversations'. While intended for coaching, this sensitive repository becomes a target for exfiltration if the agent is compromised by external inputs.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 15, 2026, 11:23 PM