fusion-issue-solving

fusion-issue-solving is an experimental workflow skill from Equinor's fusion-skills collection that handles GitHub issue resolution end-to-end. It activates on short prompts like "solve #123," "work on #123," or a pasted GitHub issue URL, and drives the full loop from issue intake through planning, implementation, validation, and a PR-ready summary. It solves the problem of turning a terse issue reference into a disciplined, low-token implementation session rather than an ad-hoc scramble.

The workflow first asks whether to use a dedicated git worktree, then confirms issue scope and out-of-scope items, checks assignee intent for the primary issue and any sub-issues, and builds a concrete dependency-ordered todo plan. A significant emphasis is a low-token GitHub strategy: prefer MCP-backed tools over ad hoc gh api or GraphQL calls, cache per-run context (issue metadata, labels, issue types, assignee candidates) and reuse it, avoid retry loops on rate limits, respect retry-after and rate-limit-reset headers, and stay within a documented per-session call budget. It defers commit, validation, changeset, and PR conventions to repository-local policy, and ships a companion checklist asset for progress tracking. It explicitly does not apply to issue drafting only, PR review only, or non-implementation research, and is not for repositories where write operations are disallowed.

The skill targets engineering teams — particularly Equinor Fusion contributors — who want a repeatable, rate-limit-aware way to implement GitHub issues with an agent. It operates on repositories the user already has access to via official GitHub and Fusion MCP tooling, handling no credentials directly; the main caveats are its experimental status and that it performs real repository write operations when permitted.