What problem does it solve?
It helps human-AI pairs reliably execute Linear issues through plan-driven TDD without losing alignment between research, implementation, and approvals.
Core Features & Use Cases
- Human-approved phase workflow: Claude researches, plans, and implements while the human explicitly approves at Phase 2-3 and Phase 4 boundaries.
- Repo-aware phase detection: It determines the right next step based on whether the branch exists and whether the issue-specific plan file exists and has been audited.
- Dependency-aware branching: It selects a base branch by inspecting Linear blocking issues and uses discrete merge commits for unmerged blockers to enable clean rebasing later.
- Plan Mode enforcement: It requires using Claude’s native Plan Mode for creating the persistent
_PLAN_{issueKey}.md handoff artifact.
- TDD execution guardrails: It directs commit delegation and requires explicit human review before marking the PR ready.
Quick Start
Use the getting-shittdone skill to start or resume execution of a specific Linear issue key with branch setup, plan creation in Plan Mode, and human checkpoints at each phase boundary.