The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external voice notes which can influence agent behavior.
Ingestion points: Fetches note data using the command getnote notes --limit 50 -o json in Step 1.
Boundary markers: None identified; the skill processes the content field directly to extract tasks and diary entries.
Capability inventory: The agent can perform file writes to ~/code/makerjackie/01mvp/, update existing notes via getnote note update, and invoke other potentially sensitive skills like mj-deploy or mj-cf-dns.
Sanitization: No explicit sanitization of note content is described.
Mitigation: A mandatory human-in-the-loop checkpoint is enforced in Step 4, requiring the user to confirm the execution plan before any action is taken.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses local file system paths for its operations.
Evidence: Accesses and modifies ~/.agents/mj-notes-workflow/state.json for persistent state tracking.
Evidence: Writes content to documentation paths within ~/code/makerjackie/01mvp/apps/01mvp-web/content/docs/.
[COMMAND_EXECUTION]: The skill relies on the execution of a local CLI tool to interact with the Getnote service.
Evidence: Uses getnote notes, getnote note update, and getnote save to manage note lifecycle and summaries.

mj-notes-workflow