The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of homework photos and student-provided text.
Ingestion points: Data enters the agent context through photos or pasted text as described in SKILL.md.
Capability inventory: The skill can write results to Busabase via scripts/record_homework.mjs and manage review decisions through scripts/execute_decisions.mjs.
Boundary markers: The skill contains explicit safety guidelines in SKILL.md such as child-friendly tone, avoiding blunt answer dumping, and verifying uncertain OCR.
Sanitization: The homework-model.js file implements sanitizeObject to prevent sensitive keys like API keys or tokens from being exposed in configuration summaries and normalizeAnswer for deterministic marking.
[COMMAND_EXECUTION]: The skill provides several Node.js scripts used to interact with the Busabase backend for management tasks such as recording homework, executing human review decisions, and deploying the associated web application. These scripts do not execute arbitrary or untrusted code and are intended for administrative use by the agent or user.
[EXTERNAL_DOWNLOADS]: The skill relies on standard dependencies, including busabase-sdk and the hono web framework, fetched from the public NPM registry. These are well-known and reputable services used for the skill's legitimate functionality.
[DATA_EXFILTRATION]: The skill demonstrates best practices for data privacy by ensuring that raw student photos are never uploaded to the database. It only stores a textual description of the image, minimizing the risk of sensitive data exposure.

kelly-homework-coach