ito-commit

Pass

Audited by Gen Agent Trust Hub on May 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local git commands (git status, git diff, git log, git add, git commit) to perform its primary function. Every destructive action is preceded by a required user confirmation step as seen in Step 8.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes the contents of local files via git diffs and logs to generate commit messages.\n
  • Ingestion points: SKILL.md (Step 2 and Step 5) through git output.\n
  • Boundary markers: Absent; the skill filters some metadata but does not use formal delimiters to isolate ingested file content.\n
  • Capability inventory: The skill has the ability to execute git add and git commit based on its processing.\n
  • Sanitization: Absent; the file content is used without sanitization to influence AI behavior.\n- [SAFE]: The skill proactively implements security measures by scanning for sensitive file patterns (e.g., .env, .key, credentials) in Step 4 and requiring confirmation before adding untracked files in Step 2.5. It also avoids executing git push, reducing the risk of accidental data exfiltration.
Audit Metadata
Risk Level
SAFE
Analyzed
May 13, 2026, 06:48 AM
Security Audit — agent-trust-hub — ito-commit