The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests and processes untrusted data from various external project files.
Ingestion points: The skill reads exploration notes, mechanism maps, critiques, Interview Briefs, measurement records, literature notes, and existing project code (SKILL.md, "Before Drafting" section).
Boundary markers: No specific delimiters or "ignore instructions" warnings are defined to isolate external data from the agent's instructions.
Capability inventory: The skill writes local Markdown files and interacts with tools like write-notion, interview, ml-experiment, and evaluation-design.
Sanitization: No explicit sanitization or filtering of the ingested content is documented.
[SAFE]: The skill references design principles from a public repository on GitHub (github.com/mattpocock/skills), which is a well-known service.

design-doc