sync-feedback

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection surface: The skill processes session transcripts from ~/.claude/learnings/sessions/ using an LLM to identify feedback. While an attacker could theoretically insert malicious instructions into a session transcript to influence the extraction, the workflow requires an explicit manual review and approval step (/sync-feedback --review) before any data is exported or acted upon. \n
  • Ingestion points: ~/.claude/learnings/sessions/ (SKILL.md)\n
  • Boundary markers: Absent in the processing instructions.\n
  • Capability inventory: Local file manipulation, Python script execution, and interactive usage of the GitHub CLI.\n
  • Sanitization: Not explicitly implemented in the skill instructions; the system relies on human validation of the extracted markdown content.\n- [COMMAND_EXECUTION]: The skill executes a local script (process-sessions.py) via Python 3 and uses standard system utilities like find and grep for file management. These operations are scoped to the skill's own directory and the user's local feedback repository.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 06:50 AM