soloscrum — solo dev scrum loop with Claude agents

soloscrum is a Claude Code plugin for running a scrum-style development loop as a solo developer. Each stage is driven by a dedicated AI agent operating under a strict contract.

The problem

A solo developer plays every role: shape the idea, file the Issue, break it down, implement it, review it, decide when it ships. Context switching between roles is costly, and the quality gates — review, DoD, “is this Issue well-formed?” — tend to be the first to be skipped.

soloscrum lets a Claude agent hold each role on your behalf. A per-role contract enforces the loop without requiring a second human reviewer.

What you get

soloscrum has four parts:

• skills — machine-readable specs the framework reads as its contract. -define-* for shared definitions, -tracker-* for tracker-profile-specific operations, plus operation-level skills. Stored under skills/.
• agents — five role definitions: soloscrum-po (Product Owner), soloscrum-design (Design), soloscrum-dev (Dev), soloscrum-ui (UI), soloscrum-review (Review). Each agent only mutates the concepts it owns.
• commands — the entry points: /refine, /breakdown, /develop, /review.
• tracker profile — github-only (default) or linear+github. Selects where Subtasks, SP, and state live. The rest of the framework is profile-agnostic. See tracker profile.

The flow

/refine     → idea becomes a GitHub Issue with Background / Goal / AC / Out of Scope
/breakdown  → Issue splits into Sub-issues when it exceeds the size threshold
/develop    → branch, implement, open a draft PR with `Closes #N`
/review     → DoD + AC + CodeRabbit + multi-agent review; verdict; ready handoff

You run gh pr merge at the end — the only irreversible step kept on the human side. Everything before it runs autonomously inside the commands: creating the PR, posting the verdict, promoting draft to ready, transitioning state.

Fit / not-fit

Use soloscrum when:

• You are a solo developer (or operate like one — one human reviewer max per change).
• You center your workflow on Claude Code and let it act autonomously inside an explicit contract.
• Your work flows through GitHub Issues + PRs (optionally Linear on top for the breakdown layer).
• You want a quality gate that fires on every PR.

soloscrum is not a fit when:

• Multiple human reviewers are required before merge.
• You need portability across AI agents. soloscrum is built specifically for Claude Code.
• You want a framework that organises people rather than codifies a contract for an agent.

Next steps

• Getting started — install the plugin, pick a tracker profile, file your first Issue.
• Concept — tracker profiles, agent responsibilities, PR lifecycle, code review process.
• Policies — the rules /refine and /review decide against (Issue format, size, SP, priority, DoD).
• Commands — /refine, /breakdown, /develop, /review in lifecycle order.

Where the AI contract lives

The authoritative source of soloscrum’s behaviour lives in the repository — skills/, agents/, commands/, and CLAUDE.md. This site is the human-readable companion. Spec files are structured for AI parsing; these pages are structured for human readers.