The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches project configuration files, sample source code, and a document manifest from the author's own GitHub repository (etalab-ia/parcours-rag) using curl. This is used to bootstrap the workshop environment without requiring a manual git clone.
[EXTERNAL_DOWNLOADS]: The setup process downloads 17 PDF documents from the official ANSSI (French National Cybersecurity Agency) website or the author's GitHub repository to serve as the local corpus for the RAG tutorial.
[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands including npm install, npx tsx, and npm run dev to install dependencies and run the generated RAG pipeline scripts. These operations are restricted to the local development workspace.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it processes external PDF content through a RAG pipeline. However, the risk is addressed through clear instructional boundaries.
Ingestion points: PDF documents are downloaded and processed in references/bootstrap-workspace.md and references/cp2-ingestion.md.
Boundary markers: The prompt instructions for the generation phase in references/cp5-generation.md explicitly direct the agent to answer using only the provided context.
Capability inventory: The environment has the capability to execute scripts via npx tsx and node as part of the workshop steps.
Sanitization: Standard sanitization is not explicitly implemented in the tutorial-grade scripts, which is typical for a pedagogical 'naïve' RAG implementation.

parcours-rag/module3