The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data in the form of student writing and conversation transcripts for feedback and correction purposes.
Ingestion points: Untrusted content is incorporated into the agent's context through functions described in SKILL.md (Error Correction section) and templates/prompt-main.md (Patterns 4 and 7).
Boundary markers: The skill's instructions and prompt templates do not utilize delimiters (such as XML tags, triple backticks, or specific markers) to separate user-provided text from the agent's internal instructions. This absence makes it easier for adversarial input to be interpreted as instructions by the underlying model.
Capability inventory: The skill's primary function is text-based analysis and instructional output. While the instructions suggest saving results to markdown files, no dangerous capabilities like arbitrary subprocess execution, network exfiltration, or access to sensitive system credentials were detected.
Sanitization: There are no specific instructions within the skill to sanitize, escape, or validate the user-provided text before it is processed.

language-teacher