ai-engineering-review
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute local scripts and command-line utilities for project analysis. This includes
node .claude/scripts/ai-signal-scan.cjsfor identifying AI-related code changes andpython .claude/scripts/code_graph tracefor mapping dependencies. It also utilizes standardgitcommands (e.g.,git status,git diff) to determine the scope of changes. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted code and documentation for review, which creates a potential surface for indirect prompt injection.
- Ingestion points: The skill ingests uncommitted changes, pull request content, and design plans during Phase 1 of the review workflow.
- Boundary markers: The protocol contains explicit rules (Phase 6) to treat all fetched content as untrusted data and strictly forbids obeying instructions found within the analyzed content.
- Capability inventory: The agent has the capability to execute shell commands (
node,python,git), perform web fetches for documentation verification, and spawn sub-agents for specialized review tasks. - Sanitization: The skill's internal protocol (specifically AE-2.2) mandates that model output must be encoded, validated, or sandboxed before being passed to any external sink or system command.
Audit Metadata