The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents the processing of untrusted external data (Excel spreadsheets and RFP PDF files).
Ingestion points: Documented in Step1Ingestion.tsx (Excel upload) and rfpExtractor.ts (PDF extraction).
Boundary markers: The documentation does not specify boundary markers or explicit sanitization instructions for handling these untrusted inputs in the prompt flow.
Capability inventory: The system has capabilities to write to a PostgreSQL database via Prisma, generate PDFs via a headless Chrome service (Browserless), and execute form actions via actionExecutor.ts.
Sanitization: Not explicitly documented in the provided bible content.
[UNVERIFIABLE_DEPENDENCIES_REMOTE_CODE_EXECUTION]: The skill references external services and official repositories for deployment and AI functionality.
References the author's own GitHub repository khaledbashir/rag2 for source code.
Interacts with AnythingLLM and Browserless services for RAG and PDF rendering.
Utilizes Puter.js for client-side vision processing via Kimi K2.5.
These references target the author's own infrastructure or well-known services, which is consistent with standard development practices.
[DATA_EXPOSURE_EXFILTRATION]: The skill contains internal infrastructure metadata for the project.
Lists the production VPS IP address: 138.201.126.110.
Lists specific production hostnames on the easypanel.host domain.
These represent the vendor's own infrastructure and are documented for project context.

anc-bible