The Agent Skills Directory

[SAFE]: The skill establishes a clear security boundary by isolating local file operations (ingestion and export) to trusted scripts meant for the operator, while the browser UI strictly interacts with the Busabase API through a sanctioned local gateway.
[INDIRECT_PROMPT_INJECTION]: The skill processes external curriculum materials and teacher drafts to generate or parse lesson plans.
Ingestion points: Untrusted data enters the agent context via scripts/ingest_plan.mjs when parsing payload files.
Boundary markers: The system uses a review queue workflow requiring dean approval for all decisions.
Capability inventory: The skill has the capability to write to the local filesystem (scripts/export_plans.mjs) and mutate records in the Busabase backend (scripts/execute_decisions.mjs).
Sanitization: UI outputs are sanitized using standard HTML escaping in lesson-views.js.
Conclusion: The risk is mitigated by the mandatory human review process before any feedback is sent or plans are published.
[COMMAND_EXECUTION]: The skill includes functional Node.js scripts (scripts/ingest_plan.mjs, scripts/run_checks.mjs, scripts/export_plans.mjs) for local data management. These scripts use standard environment variables for credentials and do not contain hidden execution patterns or remote download-and-execute logic.

kelly-lesson