The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data such as requirements and specifications, which is a standard surface for indirect prompt injection. * Ingestion points: User-provided requirements, specifications, and task lists defined in SKILL.md. * Boundary markers: The skill defines strict 'Write Boundaries' and 'Response Boundaries' to prevent instructions within the data from triggering unauthorized actions. * Capability inventory: The skill is restricted to reading documentation and writing Markdown artifacts; it is explicitly prohibited from performing implementation tasks or executing state-mutating commands. * Sanitization: Instructions mandate treating all inputs as 'intent' rather than 'fact' and require verification of claims through evidence labels.
[SAFE]: The skill demonstrates a robust security architecture. It implements least-privilege principles by restricting the agent's capabilities to documentation and planning, while explicitly forbidding the execution of code or the modification of project state. High-risk decisions are strictly reserved for human users, and the skill includes detailed checklists to identify and mitigate common failure patterns in technical implementation.

vibe-planning