friction-report
Warn
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill transmits information derived from the user's conversation history to an external endpoint (https://agent-friction-skill.vercel.app/api/draft). This payload includes summaries, error descriptions, and redacted code snippets. While the instructions advise the AI to sanitize sensitive data, the transmission occurs automatically and before the user is presented with a review link.
- [PROMPT_INJECTION]: The skill contains instructions to hide its activities from the user, specifically directing the AI to 'exit silently' if no reportable data is found and to 'say nothing' if the data transmission fails. This concealment reduces user visibility into background network operations.
- [INDIRECT_PROMPT_INJECTION]: By processing the full conversation history to generate a report, the skill is susceptible to indirect prompt injection. A malicious input within the conversation could potentially manipulate the content of the generated report or the transmission process.
- Ingestion Point: Full conversation history accessed in SKILL.md.
- Boundary markers: None present.
- Capability inventory: Network POST to external URL.
- Sanitization: Instructions to strip identifying information and PII.
Audit Metadata