Tools for structured AI development. 15 agents, 16 commands, 2 skills, 1 MCP server.
Bulk resolve GitHub PR comments using checkpoint-based verification flow
First-time setup for Unit Work - check dependencies, configure Hindsight bank, explore codebase
Browser automation using Vercel's agent-browser CLI. Use for web interaction, form filling, screenshots, scraping, and UI/backend verification. Supports ref-based selection, parallel sessions, and headed debugging. Includes verification workflows with confidence assessment. Triggers on: browse website, fill form, click button, take screenshot, scrape page, web automation, UI verification.
Extract and store learnings to Hindsight memory and local documentation
Autonomously fix failing CI by cycling through analyze → fix → commit → push → verify
Use this agent to analyze the downstream impact of resolving a merge/rebase conflict. This agent identifies affected tests, dependent files, and behavioral implications of different resolution approaches.
Use this agent to analyze the intent behind each branch's changes in a merge conflict. This agent determines whether conflicts are purely textual (same goal, different implementation) or semantic (conflicting goals).
Use this agent for codebase exploration with memory integration. This agent recalls prior exploration findings from Hindsight, validates key file locations still exist, explores the codebase, and retains summarized findings including traversal strategies. Use instead of the standard Explore agent when you want exploration results to compound across sessions. Examples: - <example> Context: The user needs to understand where authentication code lives in a new codebase. user: "I need to find where authentication is implemented" assistant: "I'll use the memory-aware-explore agent to find authentication code and document the findings for future sessions." <commentary> Use memory-aware-explore when exploration results should be retained for future sessions. </commentary> </example> - <example> Context: Planning a feature that requires understanding existing patterns. user: "I'm planning to add a new API endpoint" assistant: "Let me explore the codebase for existing API patterns. I'll use memory-aware-explore so we don't repeat this exploration next time." <commentary> Memory-aware exploration compounds knowledge - findings from this session inform future sessions. </commentary> </example>
Use this agent to review a draft plan for technical feasibility issues. This agent identifies impossible or blocked units, unclear verification strategies, hidden dependencies, unrealistic confidence estimates, and units too large for a single session. Should be invoked during /uw:plan after draft plan creation. Examples: - <example> Context: Reviewing a draft plan with a unit that depends on an external API. user: "Check if this plan is technically feasible" assistant: "I'll analyze each unit for blockers, dependencies, and verification clarity." <commentary> Use feasibility-validator to catch blockers before implementation starts. </commentary> </example>
Use this agent to review a draft plan for information gaps that would require investigation during implementation. This agent identifies unclear requirements, missing API contracts, unspecified edge cases, and ambiguous specifications. Should be invoked during /uw:plan after draft plan creation. Examples: - <example> Context: Reviewing a draft plan for a new feature. user: "Check this plan for gaps that would block implementation" assistant: "I'll analyze the plan for missing information, unclear requirements, and unspecified edge cases." <commentary> Use gap-detector agent to find gaps before presenting plan to user. </commentary> </example>
This skill should be used when reviewing code or running /uw:review. It contains 47 issue patterns with severity tiers, team standards to enforce, and implementation checklists for the Unit Work review system.
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.
External network access
Connects to servers outside your machine
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
A Claude Code plugin with tools for structured AI development. Use them together as a workflow, or pick individual tools as needed.
/plugin marketplace add ryan-relevanceai/unitwork
/plugin install unitwork@ryan-relevanceai-unitwork
AI completes features 70-80% correctly. Reviewing massive diffs to find the remaining 20% is painful. Unit Work addresses this by:
These tools work independently. Know what to build? Skip the plan, use vanilla Claude Code, then run /uw:review and /uw:pr. Want the full structured workflow? Go /uw:plan → /uw:work → /uw:review → /uw:pr.
| Command | Description |
|---|---|
/uw:plan | Interview-driven planning. Explores codebase, clarifies requirements, creates spec with implementation units. Takes ~15min — skip if you know what to build. |
/uw:work | Implements a spec with checkpoints at verifiable boundaries. Each checkpoint is a commit with a verification document and confidence score. |
| Command | Description |
|---|---|
/uw:review | Spawns 7 parallel review agents (type-safety, patterns, performance, architecture, security, simplicity, memory-validation). Verifies findings before presenting. Fixes P1 issues automatically. |
/uw:pr | Creates or updates GitHub PRs with AI-generated descriptions. |
/uw:compound | Extracts learnings from the implementation journey and stores them for future sessions. Auto-triggered after /uw:review. |
| Command | Description |
|---|---|
/uw:fix-ci | Autonomously fixes failing CI. Cycles through analyze → fix → commit → push → verify until CI passes (max 5 cycles). |
/uw:fix-conflicts | Intelligent rebase conflict resolution. Spawns intent + impact analysis agents, auto-resolves high-confidence conflicts, interviews you for ambiguous ones. |
/uw:action-comments | Bulk resolves GitHub PR review comments with checkpoint-based verification. |
| Command | Description |
|---|---|
/uw:investigate | Read-only codebase investigation. Runs tests and temp scripts to verify hypotheses without modifying production code. |
/uw:test-plan | Generates manual testing steps from git diffs. Analyzes frontend + backend changes and produces one continuous end-to-end test plan. |
/uw:browser-test | Browser automation via agent-browser for UI verification, form filling, screenshots, and scraping. |
/uw:momentic | Run, create, debug, or upload Momentic E2E tests with video recording. |
| Command | Description |
|---|---|
/uw:park | Parks current work session to a GitHub PR comment for multi-device handoff. |
/uw:resume | Resumes a parked work session by reading PR parking comments and local artifacts. |
/uw:harvest | Scrapes merged PR review comments from GitHub repos, synthesizes insights, and stores them for future sessions. |
/uw:bootstrap | First-time setup: configure Hindsight memory bank and explore codebase. Only needed if using Hindsight. |
15 specialized agents spawned by commands — you don't invoke these directly.
| Category | Agents | Used By |
|---|---|---|
| Review (7) | type-safety, patterns-utilities, performance-database, architecture, security, simplicity, memory-validation | /uw:review |
| Plan Review (3) | gap-detector, feasibility-validator, utility-pattern-auditor | /uw:plan |
| Verification (2) | test-runner, api-prober | /uw:work, /uw:fix-ci |
| Conflict Resolution (2) | conflict-intent-analyst, conflict-impact-explorer | /uw:fix-conflicts |
| Exploration (1) | memory-aware-explore | /uw:plan, /uw:bootstrap |
Context7 — Retrieves up-to-date library documentation and code examples. Used during planning and implementation to research framework patterns.
Hindsight gives the plugin persistent memory across sessions. Commands recall past learnings before starting and retain new discoveries when done. Without it, the plugin works fine — you just won't get cross-session memory.
Add to your shell profile (~/.zshrc or ~/.bashrc):
export HINDSIGHT_API_LLM_PROVIDER=openai
export HINDSIGHT_API_LLM_BASE_URL=https://openrouter.ai/api/v1
export HINDSIGHT_API_LLM_API_KEY=<your-openrouter-key>
export HINDSIGHT_API_LLM_MODEL=google/gemini-3-flash-preview
Start the container (runs in the background):
npx claudepluginhub ryan-relevanceai/unitwork --plugin unitworkMulti-agent orchestration for code that matters.
Verification-first engineering toolkit for Claude Code. 15 skills across a 5-phase spine (Investigate → Design → Implement → Verify → Ship), 8 specialist agents, an interactive setup wizard. Every skill has rationalizations + evidence requirements. Built for senior ICs and tech leads.
Corca Workflow Framework — consolidated hooks and skill orchestration for structured development sessions
Persona-driven AI development team: orchestrator, team agents, review agents, skills, slash commands, and advisory hooks for Claude Code
Code transformation: Dev SDLC orchestrator (code-shipping pipeline), plan, assert, audit, review, test, refactor, debug, for-sure. Hosts engineering agents.
Implementation planning, execution, and PR creation workflows with multi-agent collaboration