author-contributions

Pass

Audited by Gen Agent Trust Hub on Jul 30, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted metadata from the git repository, such as author names, filenames, and commit messages, as its primary input for analysis.
  • Ingestion points: Extracts commit author strings and file paths via git log and git diff-tree commands in SKILL.md.
  • Boundary markers: No explicit delimiters or instructions are used to separate untrusted git metadata from the agent's logic.
  • Capability inventory: The skill utilizes subprocess execution of git commands and local file system writes (temporary Python scripts).
  • Sanitization: Metadata is interpolated into command strings and Python scripts without visible sanitization or validation.- [COMMAND_EXECUTION]: The skill uses dynamic execution by instructing the agent to write and run temporary Python scripts to handle complex data processing tasks. It also provides shell scripts that interpolate repository-derived variables, which presents a potential surface for command injection if commit metadata contains shell-sensitive characters.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 30, 2026, 01:46 PM
Security Audit — agent-trust-hub — author-contributions