The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill is designed to ingest and follow instructions from external files, which creates a vulnerability surface if those files contain malicious instructions.
Ingestion points: Files located in .claude/homunculus/instincts/personal/ and .claude/homunculus/instincts/inherited/.
Boundary markers: Absent. Content is read directly into the context using cat with only a filename header, which may not be sufficient to prevent the agent from confusing file content with system instructions.
Capability inventory: The skill uses bash to execute cat and basename commands to list and read files.
Sanitization: Absent. The skill does not validate or sanitize the content of the markdown files before displaying them to the agent.

instinct-apply