The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes content from local markdown files that may contain untrusted or adversarial instructions.
Ingestion points: Files identified by scripts/find_links.py are read and their contents are passed to the preprocess_model and pro_model via the logic in scripts/draft_prompt.py.
Boundary markers: Notes are delimited using markdown headers (e.g., ### Note: {title}) within the prompt, which provides structural separation but does not explicitly instruct the model to ignore embedded commands.
Capability inventory: The skill has the ability to read from the local file system (within the configured zettel_dir) and write generated markdown drafts to an output_dir.
Sanitization: Content extracted from notes is included in the final prompts without sanitization or filtering for known prompt injection patterns or control markers.

zettel-brainstormer