The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the hermes CLI tool via subprocess.run in adapters.py to export memory records. This interaction is implemented securely using a list-based argument structure to prevent command injection and does not utilize a shell environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from memory sources (JSON files or tool outputs) to generate wiki pages. It mitigates injection risks through comprehensive sanitization and rigorous path validation.
Ingestion points: Data is read from .knowledge/memory.json or through the output of the hermes mnemosyne export command.
Boundary markers: Uses  and  tags to delimit generated content within index and overview files, and neutralizes Markdown markers like [[ and ## in processed text.
Capability inventory: Performs file writes using an atomic write pattern (_atomic_write) with built-in path validation.
Sanitization: Implements _sanitize_inline and _sanitize_name functions to escape special characters and normalize text before interpolation into Markdown files, preventing forgery of headers, links, or markers.

curator