The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading documentation files (.md, .html) and passing the content to LLM sub-agents. Ingestion points: The bundled scripts/link_graph.py script reads file content, and Phase 2 sub-agent prompts in SKILL.md include content from landing pages and representative documents. Boundary markers: The prompt templates for the sub-agents (e.g., 'Findability narrative review') interpolate document content directly without clear delimiters or instructions to ignore embedded commands. Capability inventory: The orchestrator agent has access to Bash, Read, Grep, and Glob tools. The bundled scripts/link_graph.py script executes shell commands via subprocess. Sanitization: No sanitization or escaping of the ingested document content is performed before interpolation into sub-agent prompts.
[COMMAND_EXECUTION]: The bundled Python script scripts/link_graph.py uses subprocess.run to execute git rev-parse --show-toplevel. While this specific usage is benign and used to locate the project root, it represents a system-level interaction capability using the Python runtime.

doc-architecture-review