ai-review-changes

Pass

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content that could contain malicious instructions to subvert the review process.
  • Ingestion points: Git diff output, commit logs, external issue tracker data, and various local file paths (docs/, specs/, .scratch/, and Obsidian vaults).
  • Boundary markers: Absent. The diff and spec contents are interpolated directly into the sub-agent prompts without delimiters or instructions to ignore embedded instructions.
  • Capability inventory: Shell execution (git commands) and local file read operations.
  • Sanitization: None. The skill processes the raw content of diffs and specs without sanitization.
  • [COMMAND_EXECUTION]: The skill instructs the agent to run git commands (e.g., git diff, git rev-parse) using user-supplied references. This pattern is vulnerable to command injection if the underlying shell execution environment does not properly sanitize these arguments.
  • [DATA_EXPOSURE]: The skill reads files from the local filesystem using absolute paths, including paths involving the $OBSIDIAN_AI_VAULT environment variable. This allows the agent to access files outside the immediate repository context.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 24, 2026, 12:40 PM
Security Audit — agent-trust-hub — ai-review-changes