The Agent Skills Directory

[SAFE]: The skill is designed with a clear separation of concerns, ensuring that the goal selection and planning phase remains read-only. It explicitly avoids repository mutations during discovery and uses a structured routing envelope to pass context between sessions, which reduces the risk of unintended behavior.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes external repository data (roadmaps and plans) that could potentially contain malicious instructions, creating an indirect prompt injection surface.
Ingestion points: The skill reads AGENTS.md, ROADMAP.md, and specific plan documents (e.g., plans/*.md) to establish project state and infer goal boundaries.
Boundary markers: The agent generates a closed "routing envelope" using structured fields like Outcome, Included results, and Authority to define a strict execution contract for the next session.
Capability inventory: The skill produces prompts that eventually drive tools such as $progress, $interview, and $create-pr in downstream agent sessions.
Sanitization: The process incorporates a mandatory "Readiness Gate" to identify unresolved decisions and a multi-step workflow that requires the user to explicitly confirm or adjust the recommended scope before any execution prompt is emitted.

next-goal