The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from user responses and metadata extracted from uploaded files to populate an application specification.
Ingestion points: User-provided text answers and absolute file paths or example references passed to the extract_metadata tool (SKILL.md).
Boundary markers: The agent is required to emit questions using a specific fenced JSON envelope (```question) and is instructed not to update the specification until clarifying ambiguous free-form answers.
Capability inventory: The skill utilizes update_spec for state modification via JSON Patch, extract_metadata for file analysis, and check_completeness for final validation (SKILL.md).
Sanitization: The instructions mandate that the agent should not blindly accept free-form answers and must ask for clarification. The runtime is also described as validating the update_spec operations.

ultraapp-interview