The Agent Skills Directory

[SAFE]: The skill instructions define a structured Socratic dialogue for gathering user intent. It incorporates safety best practices, such as requiring explicit user confirmation before finalizing a brief and prohibiting implementation actions within the same invocation without separate authorization.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential attack surface for indirect prompt injection as it is instructed to inspect repository code and documentation to ground its interview process. 1. Ingestion points: Repository files, including code and documentation, and user-provided notes (e.g., notes/product-brief.md). 2. Boundary markers: The skill uses a 'Closure Gate' and 'Refinement checkpoints' where it must restate facts and decisions for user verification. It also uses explicit source labeling (e.g., [from-user], [from-code]). 3. Capability inventory: The skill can write an 'Interview Brief' to a file (defaulting to interview-brief.md) and is designed to feed downstream vendor-authorized tools like gigio-project-setup and gigio-write-plan. 4. Sanitization: The methodology relies on iterative human feedback and a 'one-question-at-a-time' constraint which reduces the efficacy of potential injection exploitation.

deep-interview