
from specs.md170
Discover pending work items in the workspace and propose run grouping options (single, batch, wide) with dependency-aware ordering and user-preference learning.
Run-Plan scans the filesystem and state to discover intents and work items, reconciles differences with state.yaml, and generates run-scope options. It produces single/batch/wide grouping recommendations, respects dependencies (sequential execution), and records user preferences for future recommendations.
Use before executing agent runs when you need to gather pending work items and decide how to group them for execution. Triggers include: after decomposition completes, when a user wants to start execution, or when pending work items exist.
Agents that can read workspace state and file systems (CLI-capable agents, workflow orchestrators) are compatible. This is primarily a planning/coordination skill rather than a content-generation skill.
Run-plan is a workflow orchestration skill for the fabriqaai FIRE framework that discovers pending work items and proposes run grouping options (single, batch, wide) with dependency-aware ordering and preference learning. No scripts were bundled — it's purely instructional via SKILL.md. The skill is well-documented with clear flow steps, clarification sections, and template outputs, but is tightly coupled to the FIRE framework's state.yaml and .specs-fire directory conventions, limiting its applicability.
Monolithic SKILL.md with no scripts/ or references/ directories. Skill is well-written internally but highly specific to the fabriqaai FIRE framework — limited general usefulness.