The Agent Skills Directory

[SAFE]: No malicious patterns were identified in the skill instructions. The skill emphasizes using local evidence, specifically mentioning standard tools like git and manual file inspection for auditing. It includes explicit 'Anti-Patterns' that forbid dangerous behaviors such as uploading repository data to external services or running unpinned remote code.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted project data (source code, configuration files, and web responses). While this presents an indirect prompt injection surface, the skill includes guidelines for the agent to check for such vulnerabilities within the audited application itself. 1. Ingestion points: Source code files, git history, configuration manifests, and web content from user-specified URLs (SKILL.md). 2. Boundary markers: No explicit instructions are provided to the agent for using delimiters when processing these inputs. 3. Capability inventory: Local file system read access, git command execution, and HTTP network requests for browser checks (SKILL.md). 4. Sanitization: No specific sanitization or filtering of external content is required by the instructions.

production-audit