The Agent Skills Directory

[SAFE]: The skill implements a structured workflow for requirements gathering using markdown-based reference files. It manages conversation state through local file writes to a project-specific log file (progress.txt) and does not request elevated privileges, network access, or the execution of third-party scripts.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by collecting user-provided answers and aggregating them into a handoff file for downstream skills.
Ingestion points: User answers to requirements questions are captured in SKILL.md and various reference files (e.g., core-concept.md, features.md).
Boundary markers: The skill uses structured headers like PHASE, STATUS, and INTERROGATION_TRANSCRIPT to organize data in progress.txt, which helps delineate user content from instructions.
Capability inventory: The skill's capabilities are limited to reading/writing local text files and triggering other agent skills (vibe-coding-document and vibe-coding-ui-ux).
Sanitization: No explicit sanitization or instruction-stripping of user input is documented, relying on the downstream skill's logic to handle the aggregated data safely.

vibe-coding-ideate