The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The ingestion script (scripts/ingest.py) downloads markdown files from an external GitHub repository (github.com/inbn/component-gallery) to populate its local search index.\n- [REMOTE_CODE_EXECUTION]: Automated scans detected a pattern in scripts/ingest.py where content is downloaded from the network (urllib.request.urlopen) and subsequently processed by an external subprocess (rlama rag). While the fetched content consists of markdown documentation rather than executable code, the interaction between remote data and system-level execution tools is noted as a risk vector.\n- [COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run to invoke system commands and CLI tools, including firecrawl, rlama, and python3, for data processing and search operations. It also relies on hardcoded paths to external skill scripts in scripts/query.py.\n- [PROMPT_INJECTION]: The RAG-based design creates a surface for Indirect Prompt Injection (Category 8).\n
Ingestion points: Data is crawled from component.gallery and fetched from github.com/inbn/component-gallery in scripts/ingest.py.\n
Boundary markers: No specific delimiters or protective instructions are implemented to isolate retrieved data chunks from the agent's primary instruction set.\n
Capability inventory: The skill is granted access to Bash, Read, Grep, and Glob tools via SKILL.md.\n
Sanitization: There is no evidence of validation or sanitization of the markdown content fetched from remote sources before it is processed by the search engine or presented to the agent.

component-gallery