adversarial-review

Pass

Audited by Gen Agent Trust Hub on Jul 28, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill defines a read-only process for reviewing code changes. It explicitly instructs the agent to propose fixes rather than apply them and limits the scope of analysis to the provided diff or the output of git diff.
  • [COMMAND_EXECUTION]: The skill utilizes git diff to retrieve the source code context for review. This is a routine operation for a development-focused skill and is used here solely for reading data.
  • [PROMPT_INJECTION]: While the skill uses strong language like "Distrust the diff" and "Ignore credit for good intent," these are stylistic instructions designed to guide the agent's analytical posture toward finding bugs. They do not attempt to bypass the AI's core safety filters or ethical guidelines.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data in the form of code diffs. However, because its instructions are focused on logical bug hunting and it lack capabilities to execute the code it reads, the risk of indirect prompt injection is minimal and consistent with the skill's primary purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 28, 2026, 09:49 PM
Security Audit — agent-trust-hub — adversarial-review