The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a bundled Python script (extract_entities.py) via the command line to scan the user's Obsidian vault and extract metadata for the visualization.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from the user's local files to classify entities and 'calibrate' the agent's tone for the generated temple text.
Ingestion points: The extract_entities.py script reads markdown files from a user-provided directory, and the agent is explicitly instructed in SKILL.md Step 2 to read the content of central nodes.
Boundary markers: The skill does not implement delimiters or explicit 'ignore' instructions when reading the vault content.
Capability inventory: The skill spawns a subprocess to execute the metadata extraction script and writes output files (JSON and HTML) to the local file system.
Sanitization: There is no evidence of sanitization, filtering, or validation of the ingested note content before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The generated visualization (HTML) fetches Three.js and MediaPipe libraries from cdn.jsdelivr.net and storage.googleapis.com at runtime. These are well-known and established services.
[PERSISTENCE]: The generated HTML file utilizes browser localStorage to persist interaction state, such as discovered nodes and unlocked 'secrets', between user sessions.

temple-generator