commit-confirm

Pass

Audited by Gen Agent Trust Hub on Jul 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions to install a required dependency from the vendor's GitHub repository (LegendApp/legend-skills) using the npx package runner.
  • [COMMAND_EXECUTION]: The agent is instructed to invoke the $commit tool to perform its primary function of repository management.
  • [PROMPT_INJECTION]: The skill contains logic to override the default behavior of the $commit tool, such as forcing plan mode and whole-tree inspection. These are functional instructions intended to customize the workflow and do not target safety guardrail bypass.
  • [PROMPT_INJECTION]: The skill interacts with untrusted data from the local working tree, creating a surface for indirect instructions. 1. Ingestion points: Inspects staged, unstaged, and untracked changes in the working tree as described in SKILL.md. 2. Boundary markers: Mandates that the agent 'wait for explicit approval such as go' before proceeding. 3. Capability inventory: Capable of staging and committing changes through the $commit tool. 4. Sanitization: Relies on manual human-in-the-loop review of the commit plan as the primary safety mechanism.
  • [DATA_EXFILTRATION]: No exfiltration or unauthorized data access patterns were detected. The skill's operations are confined to standard version control workflows.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 17, 2026, 01:00 PM
Security Audit — agent-trust-hub — commit-confirm