The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes conversation transcripts from Claude Code and Codex to build its map structure, creating a potential surface for indirect prompt injection if malicious instructions are embedded in the conversation history.
Ingestion points: Transcripts are read and synchronized from local agent configuration directories via scripts/lib/transcripts/sync.ts and scripts/lib/transcripts/parse.ts.
Boundary markers: The agent is guided by specific protocols in SKILL.md to interpret the map outline and perform structure updates.
Capability inventory: The skill can modify local map files and host a local web server restricted to the loopback interface.
Sanitization: The web interface uses DOMPurify in web/js/markdown.js to sanitize rendered content and prevent XSS from processed conversation data.
[COMMAND_EXECUTION]: The CLI tool dm.ts spawns a local web server process using scripts/server.ts. This execution is a core part of the skill's visualization functionality and the server is limited to the local loopback interface (127.0.0.1).
[EXTERNAL_DOWNLOADS]: The web interface fetches front-end libraries and fonts from well-known services, including Preact and Marked via esm.sh, and Noto Serif SC from Google Fonts. These references are used to provide the user interface and are recognized as established service providers.

diffusion-map