deep-review

Pass

Audited by Gen Agent Trust Hub on May 11, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data, specifically pull request diffs and source code files, as part of its 'Workflow Phase 1'. This creates a surface for indirect prompt injection if an attacker embeds malicious instructions within code comments or string literals.
  • Ingestion points: Processes PR diffs, local file changes, and related project context.
  • Boundary markers: No explicit instruction delimiters (like XML tags or specific 'ignore instructions in data' warnings) are defined for the agent to separate user code from the skill's own instructions.
  • Capability inventory: The skill only produces text-based analysis and does not have access to tools for code execution, file writing, or network operations.
  • Sanitization: No sanitization or filtering of input data is defined.
  • Mitigation: The adversarial nature of the 'Skeptic' persona, which is instructed to look for 'bugs, edge cases, and malicious inputs', acts as a logical defense against injection attempts.
  • [SAFE]: The skill is 'Shape: .S..', meaning it consists entirely of instructions (SKILL.md) and metadata (plugin.json) without associated executable scripts or binaries. It does not perform any environment modifications or remote operations.
Audit Metadata
Risk Level
SAFE
Analyzed
May 11, 2026, 04:14 PM