generating-patches

Warn

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: MEDIUMPROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION] (MEDIUM): Vulnerable to Indirect Prompt Injection (Category 8) due to processing untrusted codebase data without adequate safeguards.\n
  • Ingestion points: Codebase files in the working directory accessed via git diff and git status.\n
  • Boundary markers: Absent; untrusted content is not delimited from instructions.\n
  • Capability inventory: Execution of git subprocesses and file writing to /mnt/user-data/outputs/.\n
  • Sanitization: Absent; filenames and diff output are used directly in markdown and shell loops. Malicious instructions in the codebase could influence the agent's summary or next-step reasoning.\n- [DATA_EXFILTRATION] (LOW): The skill facilitates data exposure. If directed to sensitive directories (e.g., ~/.ssh, .env), it bundles private information into a patch file for download via the output directory.\n- [COMMAND_EXECUTION] (LOW): The skill uses shell commands with string interpolation for tasks like PR description generation. Fragile shell patterns (e.g., while read f without -r) could be targeted via malicious filenames containing shell metacharacters.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 16, 2026, 08:07 AM