The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from legacy requirements files and user input, which could potentially contain malicious instructions aimed at influencing the agent's output or actions.
Ingestion points: The workflow in SKILL.md instructs the agent to read docs/requirements/raw.md, REQUIREMENTS.md, and user statements as sources for migration.
Boundary markers: The skill relies on a rigid output structure and template defined in references/idea-structure.md and uses specific markers like > 无 for missing data, which helps constrain the output format.
Capability inventory: The instructions describe capabilities for reading local files, writing a target document to docs/ideas/idea.md, and deleting source legacy files upon successful validation.
Sanitization: While the instructions emphasize recording only facts and resolving conflicts via user interaction, there is no explicit technical sanitization or escaping of the content ingested from the external source files before it is processed.

ideate