The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data (code, docs, errors, examples, commands, data, and state) to build decision trees and guide user interviews. If these external files contain malicious instructions, they could manipulate the agent's logic or the recommendations it provides to the user.
Ingestion points: SKILL.md and INTERVIEW.md explicitly instruct the agent to inspect discoverable context including files, documentation, and command outputs.
Boundary markers: The instructions do not provide explicit delimiters or "ignore instructions" wrappers for the context being processed.
Capability inventory: The skill has the ability to spawn sub-agents (via deep-thinker), conduct interviews, and route the workflow to other skills (to-prd and five-whys).
Sanitization: There is no mention of sanitization or validation of the content discovered in the environment before it is processed by the agent's reasoning loop.

discuss-task