The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data in the form of user pastes from other sessions, which is its primary function as a companion interpreter.
Ingestion points: User-provided pastes of session content from other technical discussions (SKILL.md).
Boundary markers: Absent; the skill is instructed to 'read it cold' and treat it as a direct input for analysis.
Capability inventory: The skill reads local files (e.g., repository code and docs/agents/project.md) and uses the research tool for external facts. It is platform-restricted by disable-model-invocation: true and a 'read-only' constraint that prevents commits or publishing.
Sanitization: Absent; the skill relies on interpreting the provided context directly.

interpret-session