The Agent Skills Directory

[COMMAND_EXECUTION]: The skill runs a project-local script using node .gitnexus/run.cjs. It also suggests commands for global installation and pnpm dlx which includes build permissions for native modules.
[EXTERNAL_DOWNLOADS]: The instructions guide the agent to fetch the gitnexus package from the NPM registry via npx or pnpm to ensure the runner is available.
[CREDENTIALS_UNSAFE]: The skill mentions storing LLM API keys in ~/.gitnexus/config.json and uses the OPENAI_API_KEY environment variable for faster processing.
[PROMPT_INJECTION]: The analysis process creates a surface for indirect prompt injection by parsing repository source files into a knowledge graph.
Ingestion points: Source files in the project root are read during analyze (SKILL.md).
Boundary markers: None are provided to distinguish between code and instructions.
Capability inventory: The tool writes to .gitnexus/ and creates context files like CLAUDE.md, which are subsequently read by the agent.
Sanitization: No sanitization process for ingested source code content is mentioned.

gitnexus-cli