The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from external research files.
Ingestion points: The skill reads various markdown files including change.md, research/*.md, and frame.md from the local file system.
Boundary markers: The skill explicitly instructs the agent to invoke dx-references with untrusted-content when encountering research marked as kind: external before acting on the reported content.
Capability inventory: The skill has the capability to read extensive documentation structures and write new plan.md files and update change.md metadata.
Sanitization: The instructions mandate treating summaries from external sources as data rather than instructions, providing a layer of protection against accidental command obedience from embedded content.

dx-plan