merge

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses string interpolation to build shell commands for git operations and Python script execution. Variables such as {session-id}, {winner}, and {base_branch} are inserted directly into command strings without evident sanitization. This allows for potential command injection if these identifiers contain shell metacharacters such as semicolons or pipes.\n
  • Evidence: Multiple shell blocks in SKILL.md sections for merging, archiving, and state updates.\n- [INDIRECT_PROMPT_INJECTION]: The skill accepts external input (session IDs, agent names) which flows into sensitive execution sinks (shell commands) without sanitization or boundary markers.\n
  • Ingestion points: SKILL.md processes arguments and variables derived from session context.\n
  • Boundary markers: None identified in the interpolation logic.\n
  • Capability inventory: Shell access through git CLI and python subprocesses.\n
  • Sanitization: None identified in the instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 06:04 PM
Security Audit — agent-trust-hub — merge