The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input, such as learning objectives and curriculum standards, and possesses the capability to modify and create files using tools like write_file and create_file.
Ingestion points: The skill determines the subject domain, curriculum standards, and learner ability from external user input as described in the 'Suggested Process' in SKILL.md.
Boundary markers: The instructions lack specific requirements for the agent to wrap or delimit external content to prevent the execution of embedded instructions.
Capability inventory: The skill has access to read_file, write_file, edit_file, create_file, and search_code as defined in the YAML frontmatter of SKILL.md.
Sanitization: While the skill mentions an 'ASI05 RCE Guard' for sanitizing test environments, it does not provide specific instructions for the agent to sanitize the input data itself before processing it into exercise files.
[COMMAND_EXECUTION]: The skill documentation and upgrade plan reference the execution of code and scripts.
SKILL.md includes a guardrail (ASI05) for executing student-submitted code or tests, which implies the agent may interact with execution environments.
UPGRADE_PLAN_2026.md mentions running local validation scripts such as python3 core/scripts/validate-all.py and validate-rules.py.
These references describe functional requirements or development tasks rather than malicious intent, but they highlight the skill's reliance on code execution capabilities.

create-exercises