From unitwork
This skill should be used when implementing features with human-in-the-loop verification. It provides the core Unit Work methodology including checkpoint-based development, confidence assessment, memory integration with Hindsight, and verification strategies. Use this skill for planning features, executing with checkpoints, reviewing code, and compounding learnings.
How this skill is triggered — by the user, by Claude, or both
Slash command
/unitwork:unitworkThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
> "The largest task an AI can self-validate to 100% accuracy, that is also able to get validated by the minimum amount of human review"
"The largest task an AI can self-validate to 100% accuracy, that is also able to get validated by the minimum amount of human review"
Unit Work replaces arbitrary development phases with a verification-driven approach:
AI is strong at verifying:
AI is weak at verifying:
The plugin adapts confidence and checkpoint behavior based on these strengths/weaknesses.
/uw:plan -> /uw:work -> /uw:review -> /uw:compound
| | | |
Spec.md Checkpoints Code Review Learnings
+ Verify.md + Fix Loop to memory
|
Create PR
See decision-trees.md for detailed decision flows:
Unit Work creates this structure in your project:
.unitwork/
├── specs/ # {DD-MM-YYYY}-{feature}.md
├── verify/ # {DD-MM-YYYY}-{n}-{name}.md
├── review/ # Code review findings
├── learnings/ # Compound phase output
└── test-plans/ # {DD-MM-YYYY}-{feature}.md
See interview-workflow.md for the complete interview protocol including confidence-based depth assessment and stop conditions.
CRITICAL: Memory recall is the foundation of compounding. Skip it and you lose all accumulated learnings.
See checkpointing.md for the complete checkpointing reference including:
See hindsight-reference.md for complete patterns including:
# Bank name (config override → git remote → worktree → pwd)
BANK=$(jq -re '.bankName // empty' .unitwork/.bootstrap.json 2>/dev/null || git config --get remote.origin.url 2>/dev/null | sed 's/.*\///' | sed 's/\.git$//' || basename "$(git worktree list 2>/dev/null | head -1 | awk '{print $1}')" || basename "$(pwd)")
# Recall with ANSI stripping
hindsight memory recall "$BANK" "query" --budget mid --include-chunks 2>&1 | sed 's/\x1b\[[0-9;]*m//g'
# Retain (always async)
hindsight memory retain "$BANK" "narrative" --context "context" --doc-id "id" --async
Context7 provides framework documentation lookup via MCP. Use it when implementing unfamiliar APIs.
Step 1: Resolve library ID
mcp__unitwork_context7__resolve-library-id
query: "what you're trying to implement"
libraryName: "framework-name"
Step 2: Query documentation
mcp__unitwork_context7__query-docs
libraryId: "/org/project" (from step 1)
query: "specific API or pattern"
| Subagent/Skill | Purpose | When to Use |
|---|---|---|
| test-runner | Execute tests | Changed test files or tested code |
| api-prober | Probe API endpoints | Changed API endpoints |
| /uw:browser-test (command) | UI verification | Changed UI components |
| Agent | Focus |
|---|---|
| type-safety | Casting, guards, nullability |
| patterns-utilities | Existing solutions, duplication |
| performance-database | N+1, indexes, parallelization |
| architecture | Structure, coupling, boundaries |
| security | Injection, auth, data exposure |
| simplicity | Over-engineering, YAGNI |
| memory-validation | Learnings from Hindsight memory |
Start at 100%, subtract:
>= 95%: Checkpoint and continue < 95%: Checkpoint and pause for human review
/uw:plan - Interview and create spec/uw:work - Execute with checkpoints/uw:review - Parallel code review/uw:compound - Extract learnings/uw:bootstrap - First-time setup/uw:pr - Create/update GitHub PRs/uw:action-comments - Resolve PR comments/uw:fix-ci - Autonomously fix failing CI/uw:fix-conflicts - Intelligent rebase conflict resolution/uw:test-plan - Generate manual testing steps from git diffsnpx claudepluginhub ryan-relevanceai/unitwork --plugin unitworkGuides the full SDLC workflow: planning, implementation, testing, and deployment. Automates checklist-driven development for features, bug fixes, refactoring, and releases.
Orchestrates all code-modifying development tasks like bug fixes, enhancements, and new features using adaptive phases for analysis, specs, TDD, implementation, and verification.
Enforces a gated Spec → Plan → Build → Test → Review → Ship lifecycle for multi-file features and projects, preventing AI coding agents from skipping specification and verification steps.