The Agent Skills Directory

[SAFE]: The skill is designed exclusively for generating text-based documentation and architectural diagrams. It lacks any functionality for command execution, persistence, privilege escalation, or remote code downloads.
[DATA_EXPOSURE]: The skill scans common build configuration files (such as package.json, Cargo.toml, and pyproject.toml) to detect the project's programming language. This operation is limited to project metadata and does not involve accessing sensitive files like private keys, environment variables, or cloud credentials.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content by reading the user's source code to generate citations and descriptions.
Ingestion points: The agent reads local repository files and build configurations as part of its analysis workflow.
Boundary markers: The skill does not explicitly define delimiters for external content, relying on standard text generation guardrails.
Capability inventory: The skill is restricted to text and Mermaid diagram generation. It has no capabilities for file writing, network communication, or shell command execution.
Sanitization: No specific sanitization is implemented for the ingested code content, but the lack of executable capabilities significantly mitigates the risk of a successful injection attack.

wiki-onboarding