The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted user-supplied materials during teaching tasks. It includes a mandatory evidence chain for mitigation: it identifies ingestion points in user materials, requires boundary markers (quoting/paraphrasing), limits the capability inventory to the 'Read' tool only, and mandates sanitization by stripping embedded instructions.
[COMMAND_EXECUTION]: The skill configuration and documentation explicitly prohibit the use of Bash, Python, or other executable scripts. It is designed purely for data emission and lacks the permissions required for command execution.
[DATA_EXFILTRATION]: No network access, remote downloads, or sensitive file path references were detected. The skill operates in-memory and defers rendering to other components, minimizing data exposure risks.

teacher-skill