fusion-issue-task-planning

Fusion-issue-task-planning is an experimental Equinor skill that breaks a GitHub User Story issue into an ordered, traceable set of task-issue drafts before implementation begins. It addresses the gap between a high-level story and actionable engineering work, producing dependency-ordered tasks each carrying an objective, scope boundary, deliverable, verification method, and mapped acceptance-criteria references. It is explicitly not for direct code implementation, PR review, or GitHub mutations without confirmation.

The workflow is safety-first. It defaults to draft-only mode and requires explicit same-turn confirmation before any GitHub mutation. It classifies its own operating mode based on which companion skills are available (orchestrated, direct-subordinate, or inline) and degrades gracefully when preferred skills are missing. It researches the story (title, body, acceptance criteria, scenarios, ancestor chain, existing sub-issues), clarifies only when needed with at most three questions per batch, and runs a devil's-advocate review that inspects the task set for architecture-ambiguity signals such as unresolved design decisions, implicit backend/frontend contracts, vague sequencing, and contested ownership. When two or more signals appear it routes to an interrogator review or emits an ambiguity checklist the user must resolve before drafts are generated. The skill requires the GitHub MCP server and suggests a Fusion MCP; issue mutation is delegated to a companion authoring skill with MCP-first behavior and GraphQL fallbacks.

It targets engineering teams (its owner is @equinor/fusion-core) who plan work in GitHub Issues and want disciplined, reviewable task breakdowns with explicit publish gates. Being experimental, its structure and outputs may change between versions.