The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-submitted exercises and self-reports which may contain malicious instructions designed to influence the agent's behavior during assessment.
Ingestion points: The agent ingests data from learner answers and reports as described in SKILL.md and scripts/learning_store.py.
Boundary markers: While the skill instructs the agent to preserve actual answers, it lacks explicit instructions to treat user-provided text as untrusted data or to ignore embedded commands.
Capability inventory: The agent possesses the capability to read/write local files via the learning_store.py script and execute generated build scripts for document rendering.
Sanitization: The learning_store.py script validates JSON structure and identifiers using regex but does not sanitize the content of the text fields provided by the user.
[DYNAMIC_EXECUTION]: The instructions for creating printable materials in references/workbooks.md suggest that the agent generate an editable content source and a builder script (e.g., build.py) to be executed locally to produce PDF artifacts. This involves the runtime generation and execution of code within the local environment.

japanese-learning