ps-how
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the local codebase to generate explanations and critiques, creating an attack surface for indirect prompt injection.
- Ingestion points: The skill ingests data from the local filesystem through explorer subagents using
Glob,Grep, andReadtools (SKILL.md, explorer-prompt.md). - Boundary markers: Instructions in the
references/directory use placeholders like{EXPLANATION},{FILE_PATHS}, and{EXPLORER_FINDINGS_ALL}to interpolate external content into subagent prompts. - Capability inventory: All subagents are explicitly restricted to "read-only access" (SKILL.md). The skill does not exhibit capabilities for network exfiltration, file modification, or arbitrary code execution.
- Sanitization: The provided prompt templates do not include specific instructions for sanitizing, escaping, or filtering ingested codebase content.
Audit Metadata