peeps

Fail

Audited by Gen Agent Trust Hub on Apr 14, 2026

Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Suggests creating a persistent cron job (*/30 7-22 * * *) for the Peeps: check heartbeat, allowing the agent to maintain autonomous access and execute tasks outside of active user sessions.
  • [COMMAND_EXECUTION]: Employs shell commands like mkdir and grep to interact with the local filesystem and query the contact database stored in kyp/peeps/.
  • [REMOTE_CODE_EXECUTION]: Includes a mechanism for the agent to update its own logic by fetching and overwriting its instruction file (SKILL.md) from a remote GitHub repository.
  • [DATA_EXFILTRATION]: Performs automated web searches using names and organization details found in local private files to enrich metadata, leaking personal information from the local database to third-party search providers.
  • [EXTERNAL_DOWNLOADS]: Documentation directs the use of curl and npx to fetch skill components and updates from the Know-Your-People/peeps-skill repository.
  • [PROMPT_INJECTION]: Exhibits an indirect prompt injection surface through the ingestion of external data.
  • Ingestion points: Web search results and local Markdown files.
  • Boundary markers: Absent; no delimiters are used to separate untrusted content from instructions.
  • Capability inventory: File system access (grep, mkdir), network access (web search), and persistence via cron.
  • Sanitization: Absent; no validation or escaping is applied to ingested data before processing.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Apr 14, 2026, 09:58 AM