Tinker
Resources
Agent logs
Agent memories
Agent sessions
Agent terminal logs
Agents
Comments
Epics
Projects
Proposals
Tickets
Avo user
Resources
Agent logs
Agent memories
Agent sessions
Agent terminal logs
Agents
Comments
Epics
Projects
Proposals
Tickets
Avo user
Home
Tickets
Create Git Workflow Skills for Tinker Agents
Edit
Create Git Workflow Skills for Tinker Agents
Cancel
Actions
Transition Ticket
Approve & Merge
Save
Title
*
Priority
low
medium
high
critical
Ticket type
epic
story
task
bug
subtask
Project
*
Choose an option
alpha
tinker
Create new project
Parent Ticket
Choose an option
Create OrchestratorPingJob for periodic coordination
Add Chat to Remote Support Sessions
Migrate from jsonapi-serializer to Alba gem
Allow cancel and approve transitions from any ticket state
Simplify data model and encourage knowledge accumulation
Remove memory_consolidation_job and all references
Add draft status for tickets
Investigation: Port sandboxed Docker + MCP + skills to other repositories
Remove unused columns from tickets table
Integrate chat panel into OperatorCommandCenter
Agent Visibility System: Live terminal streaming, session logs, and debug dashboard
Part 1: Live Terminal Streaming - Capture agent PTY output and stream to WebSocket
Autonomous Agent Coordination via WebSocket
Part 2: Agent Session Logs - Store terminal output persistently for replay
Remove agent assignments - use status-only queue workflow
Add mark_busy and mark_idle MCP tools for agents
Improve dashboard/logs page visual appearance
Add reject transition from in_progress status
Auto-block tickets with unsatisfied dependencies (phased development)
Forbid bare HTTP status checks in tests (hides debugging info)
Orchestrator assigns merged tickets despite workflow instructions
Fix ticket updates from UI
Simplify transition_ticket to allow any status (remove finalize_task)
Add UI to update tickets
Implement Automated Task Prioritization Engine
Create Git Workflow Skills for Tinker Agents
Fix reviewer guidelines: Require test coverage for new functionality
Escalation: Reviewer agent unresponsive - not picking up pending audits
Change default ticket status from backlog to draft
Escalation: Clear 11 tickets stuck in pending_approval status
Fix worker workflow: submit_review transition missed due to context overflow
Make planner a first-class agent member (not wildcard)
Register archive/unarchive MCP tools in mcp-bridge
Extract Critical Workflows to Skills (Test Set)
Create Automated Code Generation and Refactoring Engine
Fix memory_consolidation_job reference to removed acceptance_criteria column
Test Alba migration and verify API compatibility
Add API key authentication (no user model)
Bug: Worker marked busy but no session created - logs have nowhere to go
Cleanup and finalize Alba migration
Update list_tickets MCP tool to add pagination with max limit of 20
Add formal ticket dependencies: prevent orchestrator assigning blocked tasks
Update AgentChannel to only log (no state management)
Fix list_tasks response size - remove verbose fields
Fix kanban view - archived tickets should not appear
Task 2: Role Validation Logic Implementation
Task 4: Guardrails and Error Message Implementation
Step 4: Ticket Detail Page with daisyUI Components
Add planned tickets page (draft/backlog/todo)
Add planner to default agent creation set
Terminal App: WebSocket-to-Claude-Code Bridge
Add web UI for browsing and managing agent memories
Reduce WebSocket log noise - hide normal reconnection messages
Add agent status tracking (status, status_updated_at)
Add availability_updated_at to list_members MCP output
Make planner a standalone agent (not tied to orchestrator)
Update Gemfile: Add alba gem, remove jsonapi-serializer
Drop artifacts and code_diffs database tables
Step 8: Theming and Preferences with daisyUI
Update BaseController render_jsonapi methods for Alba
Modify dashboard controller to support showing all projects
Update Kanban view to display project information when showing all tasks
Add missing MCP tools to TypeScript bridge
Add project selector dropdown to Kanban board UI
Test cross-project Kanban functionality
Memory Consolidation Skill - Pattern Extraction & Deduplication
Add "todo" status with "plan" action to Tinker workflow
Optimize list_tickets output size and add multi-status filtering
Rename api_key_plaintext column to api_key
Step 7: Responsive Design with daisyUI Components
Task 1: System Prompt Implementation for Role Enforcement
Add assign_ticket and list_agents MCP tools
Add missing /api/v1/tickets/:id/claim endpoint
Update agents.rb with deliverable parts guidance
Remove archived column, use archived_at instead
Fix workflow: pass_audit→pending_approval, fail_audit→todo
Orchestrator should assign one ticket at a time to reviewers (not "two tickets need review")
Step 6: Logs Page with daisyUI Components
Step 9: Polish and Theme Configuration with daisyUI
Modify reviewer prompt to avoid gh pr review command
Find subtasks for Researcher epic
Add delete_proposal MCP tool for Researcher agent
Refactor get_terminal_logs MCP: line-based limiting with TerminalLogCleaner
Add confidence field (0-100) to approvals and proposals
Add list_comments MCP tool for tickets
Task 3: Role-Specific Tool Access Control
Rejection workflow: rejected tickets should go directly to "todo" not "in_progress"
Part 3: Debug Dashboard - Web UI for real-time agent monitoring
Fix list_tickets MCP tool: remove draft exclusion, always include draft tickets
Create comprehensive architecture document for autonomous agents
Add ask_for_memory_deletion MCP tool (with human confirmation)
Add file attachments support to chat
Relax reviewer scope enforcement to allow necessary related changes
Create Memory Skill for Agent Knowledge Sharing
Add test coverage for recent PRs (#56, #57, #58)
Design Autonomous Agent Framework Architecture
Step 4: Ticket Detail Page with daisyUI Components
Escalation: update_ticket MCP tool returns ForbiddenAttributesError
Fix get_terminal_logs MCP tool: limit, pagination, timestamps
Step 5: Approvals Page with daisyUI Components
Phase 3: Feature ideation from ticket analysis
Create comprehensive deployment instructions (Coolify + Neon Postgres + proper Dockerfile)
Part 2: Agent Session Logs (Store terminal output persistently)
Phase 1 (MVP): Memory and ticket pattern analysis
Step 3: Kanban Board with daisyUI Styling
Build Self-Healing System Infrastructure
Develop Intelligent Resource Management System
Add list_agent_logs MCP tool for orchestrator and researcher
Add unified get_status MCP tool for project overview
Create Autonomous Decision-Making Framework
Implement Autonomous Testing and Validation System
Create ActionCable channel for guest chat subscriptions
Introduce Researcher agent: 24/7 autonomous code & system analyst
Improve proposal UI: markdown rendering, evidence display, links, and visibility
Add GitHub label "tinker-reviewed" to PRs after reviewer review
Add backlinks in GitHub comments and PR descriptions
Build Self-Improvement and Learning System
Reject debug dashboard in PR #57 - require feature specs
Create proposal system: storage, API, and admin interface
Implement smarter context refresh conditional on worker availability
Set up Researcher agent: infrastructure, MCP tools, and guardrails
Phase 2: Code quality analysis and test coverage detection
Remove File Lists from Tickets - Constrains Agent Thinking
Fix 500 error when adding comments - undefined method 'parent_id' on Comment
Fix tickets#show comment UI: implement comment creation for humans
Convert TicketSerializer to Alba format
Convert ProjectSerializer to Alba format
Slice 1: Operator Chat (Full Vertical Slice)
Reviewer must run tests and detect missing specs before approving
Slice 2: Guest Chat (Extends Slice 1)
Implement Proactive Anomaly Detection System
Convert AgentSerializer to Alba format
Convert remaining serializers (Comment, CodeDiff, AgentMemory, Artifact) to Alba
Slice 3: Chat History Review (Polish)
Step 2: Dashboard with daisyUI Components
Step 1: Install and Configure daisyUI
Develop Continuous Performance Optimization System
Phase 4: 24/7 operation, daily digest, and batch approval
Implement Autonomous Documentation Generation
Build Autonomous Communication and Coordination Hub
Create Autonomous Security and Compliance Guardian
Implement Continuous Innovation and Experimentation Platform
Tool Usage Analytics from Logs
Integrate All Autonomous Systems into Cohesive Ecosystem
Epic: Implement Strict Role Enforcement for Tinker Agents
Fix Planner behavior - stop writing implementation details in ticket descriptions
Fix N+1 queries on kanban board page
Implement Proper Ticket Blocking/Dependency System
Implement ticket archival system with cascade and auto-archive
Task 5: Testing Role Compliance
Integrate chat panel into GuestKiosk
UI Modernization with daisyUI
Add unified set_agent_status MCP tool for orchestrator
Agent Escalation: Create Tickets for MCP/Workflow Issues
Create SessionChatPanel React component
Fix syntax error in tickets show view and add feature spec
Add chat message API endpoints for operators and guests
Create SupportSessionMessage model and migration
Create SendChatMessageService for support session chat
Add chat history review in session detail view
Update worker-workflow skill: add explicit git branch checking before starting tickets
Orchestrator assigns merged tickets - missing PR status check tool
Create researcher-workflow skill
Remove MemoryDeletionRequest - migrate memory deletion to proposals-only workflow
Add offset parameter to list_memories MCP tool
Fix approvals page reject - add rejection reason modal
Create proposal-execution skill: Enable researchers to execute approved proposals
Fix bare HTTP status check warnings in specs
Role-based skill scoping: agents only access their own skills
Create new parent ticket
Description
## Problem Worker agents have poor git workflow habits: 1. **PR branch confusion** - When asked to fix an issue in PR #33, they create a new random branch instead of using the PR's existing branch 2. **No branch stacking** - All phases branch from main instead of stacking (phase-2 from phase-1, etc.) 3. **Incorrect push targets** - Changes are pushed to wrong branches 4. **Bad commit messages** - Vague, inconsistent format ## Solution Create **one comprehensive Claude Code skill** that teaches agents proper git workflows. The skill is automatically invoked based on context and provides step-by-step instructions for all git operations. ## Skill to Create ### Skill: `git-workflow` **Location:** `.claude/skills/git-workflow/SKILL.md` **Purpose:** Comprehensive git workflow guide for all operations **When to use:** Whenever git operations are needed (fixing PRs, creating commits, creating branches, pushing changes) ```yaml --- name: git-workflow description: Use for any git operations: fixing PRs, creating commits, creating branches, pushing changes. Handles proper branch management, commit conventions, and stacked PRs. allowed-tools: Bash, gh --- # Git Workflow - Complete Guide ## Golden Rules 1. **Never push to main** 2. **Fix PRs on their existing branch** - don't create new branches 3. **Stack phased work** - Phase N branches from Phase N-1 4. **Use conventional commits** - type(scope): subject --- ## Scenario 1: Fixing an Existing PR When user says: "Fix PR #33" or "Update pull request 45" or "Modify #123" ### Steps 1. **Get the PR branch:** ```bash gh pr view --json headRefName --jq '.headRefName' ``` Save this branch name - use it for ALL operations. 2. **Checkout the PR branch:** ```bash git fetch origin git checkout ``` 3. **Make changes** using Edit/Write tools 4. **Commit with proper format:** ```bash git add git commit -m "fix(scope): description of fix" git push origin ``` 5. **No new PR needed** - same PR URL ### Example ``` User: "Fix the failing tests in PR #33" 1. gh pr view 33 --json headRefName → "feature/ticket-81-ui-foundation" 2. git checkout feature/ticket-81-ui-foundation 3. [fix test files] 4. git commit -m "fix(tests): resolve assertion failures in dashboard spec" 5. git push origin feature/ticket-81-ui-foundation ``` --- ## Scenario 2: Creating a New Feature (Single PR) When user says: "Implement X" or "Add feature Y" (no mention of phases/stacking) ### Steps 1. **Start from main:** ```bash git checkout main git pull origin main ``` 2. **Create feature branch:** ```bash git checkout -b feature/descriptive-name ``` 3. **Make changes** using Edit/Write tools 4. **Commit and push:** ```bash git add -A git commit -m "feat(scope): description" git push -u origin feature/descriptive-name ``` 5. **Create PR:** ```bash gh pr create --base main --title "feat(scope): description" ``` --- ## Scenario 3: Creating Stacked PRs (Phased Epic) When user says: "Create Phase 1, Phase 2, Phase 3" OR working on epic subtasks sequentially ### Critical Concept Each phase builds on the PREVIOUS phase, NOT main. ``` main ← phase-1 ← phase-2 ← phase-3 ``` ### Phase 1 (Foundation) ```bash git checkout main git pull origin main git checkout -b feature/phase-1-description # Make changes git commit -m "feat(phase1): description" git push -u origin feature/phase-1-description gh pr create --base main --title "Phase 1: Description" ``` ### Phase 2 (Builds on Phase 1) ```bash # IMPORTANT: Branch from phase-1, NOT main git checkout feature/phase-1-description git pull origin feature/phase-1-description git checkout -b feature/phase-2-description # Make changes git commit -m "feat(phase2): description" git push -u origin feature/phase-2-description gh pr create --base feature/phase-1-description --title "Phase 2: Description" ``` ### Phase 3+ (Continue stacking) ```bash git checkout feature/phase-2-description git checkout -b feature/phase-3-description # ... continue pattern ``` ### After Phase 1 Merges to Main ```bash git checkout main && git pull git checkout feature/phase-2-description git rebase main git push -f origin feature/phase-2-description ``` Now Phase 2 is based on updated main. --- ## Commit Message Format Always use: ``` (): ``` ### Types - `feat`: New feature - `fix`: Bug fix - `refactor`: Code refactoring - `test`: Adding or updating tests - `docs`: Documentation changes - `chore`: Maintenance tasks ### Rules - Subject: imperative mood, max 50 chars, no period - Examples: - ✅ "fix(auth): prevent SQL injection in login" - ✅ "feat(dashboard): add real-time agent status" - ✅ "test(models): add ticket dependency coverage" - ❌ "Fixed the bug" - ❌ "Update file.rb" --- ## Quick Reference | Task | Base Branch | PR Base | |------|-------------|---------| | New feature | main | main | | Fix PR #X | PR's branch | (no new PR) | | Phase 1 | main | main | | Phase 2 | phase-1 branch | phase-1 branch | | Phase 3 | phase-2 branch | phase-2 branch | --- ## Critical DOs and DON'Ts | ✅ DO | ❌ DON'T | |-------|---------| | Checkout PR branch when fixing | Create new branch for PR fixes | | Stack phases (phase-2 from phase-1) | Branch all phases from main | | Use conventional commit format | Write vague commit messages | | Push to the branch you're working on | Push to random branches | | Set correct PR base for stacked PRs | Base all PRs on main | ``` --- ## Implementation Steps 1. **Create skill directory:** ```bash mkdir -p .claude/skills/git-workflow ``` 2. **Create SKILL.md** with content above 3. **Update agent configuration** to include the skill: ```yaml --- name: tinker-worker description: Implement features, fix bugs, write code skills: git-workflow tools: Read, Write, Edit, Bash, Grep, Glob --- You are the Tinker Worker agent... When performing any git operations, always follow the git-workflow skill. ``` ## Acceptance Criteria 1. `.claude/skills/git-workflow/SKILL.md` created 2. Agent configuration updated to include skill 3. When asked to "fix PR #33", agent checks out the correct branch and pushes to it 4. When asked to commit, agent uses conventional commit format 5. When creating phased work, agent creates stacked branches 6. No more random branches 7. No more pushes to wrong branches ## Files to Create - `.claude/skills/git-workflow/SKILL.md` ## Files to Modify - Agent configuration files (add `skills: git-workflow`) ## Notes - One skill covers all git scenarios - Skills are auto-invoked based on keywords in the description - Test by giving agents specific git tasks and verifying correct behavior
## Problem Worker agents have poor git workflow habits: 1. **PR branch confusion** - When asked to fix an issue in PR #33, they create a new random branch instead of using the PR's existing branch 2. **No branch stacking** - All phases branch from main instead of stacking (phase-2 from phase-1, etc.) 3. **Incorrect push targets** - Changes are pushed to wrong branches 4. **Bad commit messages** - Vague, inconsistent format ## Solution Create **one comprehensive Claude Code skill** that teaches agents proper git workflows. The skill is automatically invoked based on context and provides step-by-step instructions for all git operations. ## Skill to Create ### Skill: `git-workflow` **Location:** `.claude/skills/git-workflow/SKILL.md` **Purpose:** Comprehensive git workflow guide for all operations **When to use:** Whenever git operations are needed (fixing PRs, creating commits, creating branches, pushing changes) ```yaml --- name: git-workflow description: Use for any git operations: fixing PRs, creating commits, creating branches, pushing changes. Handles proper branch management, commit conventions, and stacked PRs. allowed-tools: Bash, gh --- # Git Workflow - Complete Guide ## Golden Rules 1. **Never push to main** 2. **Fix PRs on their existing branch** - don't create new branches 3. **Stack phased work** - Phase N branches from Phase N-1 4. **Use conventional commits** - type(scope): subject --- ## Scenario 1: Fixing an Existing PR When user says: "Fix PR #33" or "Update pull request 45" or "Modify #123" ### Steps 1. **Get the PR branch:** ```bash gh pr view <PR_NUMBER> --json headRefName --jq '.headRefName' ``` Save this branch name - use it for ALL operations. 2. **Checkout the PR branch:** ```bash git fetch origin git checkout <branch_name> ``` 3. **Make changes** using Edit/Write tools 4. **Commit with proper format:** ```bash git add <files> git commit -m "fix(scope): description of fix" git push origin <branch_name> ``` 5. **No new PR needed** - same PR URL ### Example ``` User: "Fix the failing tests in PR #33" 1. gh pr view 33 --json headRefName → "feature/ticket-81-ui-foundation" 2. git checkout feature/ticket-81-ui-foundation 3. [fix test files] 4. git commit -m "fix(tests): resolve assertion failures in dashboard spec" 5. git push origin feature/ticket-81-ui-foundation ``` --- ## Scenario 2: Creating a New Feature (Single PR) When user says: "Implement X" or "Add feature Y" (no mention of phases/stacking) ### Steps 1. **Start from main:** ```bash git checkout main git pull origin main ``` 2. **Create feature branch:** ```bash git checkout -b feature/descriptive-name ``` 3. **Make changes** using Edit/Write tools 4. **Commit and push:** ```bash git add -A git commit -m "feat(scope): description" git push -u origin feature/descriptive-name ``` 5. **Create PR:** ```bash gh pr create --base main --title "feat(scope): description" ``` --- ## Scenario 3: Creating Stacked PRs (Phased Epic) When user says: "Create Phase 1, Phase 2, Phase 3" OR working on epic subtasks sequentially ### Critical Concept Each phase builds on the PREVIOUS phase, NOT main. ``` main ← phase-1 ← phase-2 ← phase-3 ``` ### Phase 1 (Foundation) ```bash git checkout main git pull origin main git checkout -b feature/phase-1-description # Make changes git commit -m "feat(phase1): description" git push -u origin feature/phase-1-description gh pr create --base main --title "Phase 1: Description" ``` ### Phase 2 (Builds on Phase 1) ```bash # IMPORTANT: Branch from phase-1, NOT main git checkout feature/phase-1-description git pull origin feature/phase-1-description git checkout -b feature/phase-2-description # Make changes git commit -m "feat(phase2): description" git push -u origin feature/phase-2-description gh pr create --base feature/phase-1-description --title "Phase 2: Description" ``` ### Phase 3+ (Continue stacking) ```bash git checkout feature/phase-2-description git checkout -b feature/phase-3-description # ... continue pattern ``` ### After Phase 1 Merges to Main ```bash git checkout main && git pull git checkout feature/phase-2-description git rebase main git push -f origin feature/phase-2-description ``` Now Phase 2 is based on updated main. --- ## Commit Message Format Always use: ``` <type>(<scope>): <subject> ``` ### Types - `feat`: New feature - `fix`: Bug fix - `refactor`: Code refactoring - `test`: Adding or updating tests - `docs`: Documentation changes - `chore`: Maintenance tasks ### Rules - Subject: imperative mood, max 50 chars, no period - Examples: - ✅ "fix(auth): prevent SQL injection in login" - ✅ "feat(dashboard): add real-time agent status" - ✅ "test(models): add ticket dependency coverage" - ❌ "Fixed the bug" - ❌ "Update file.rb" --- ## Quick Reference | Task | Base Branch | PR Base | |------|-------------|---------| | New feature | main | main | | Fix PR #X | PR's branch | (no new PR) | | Phase 1 | main | main | | Phase 2 | phase-1 branch | phase-1 branch | | Phase 3 | phase-2 branch | phase-2 branch | --- ## Critical DOs and DON'Ts | ✅ DO | ❌ DON'T | |-------|---------| | Checkout PR branch when fixing | Create new branch for PR fixes | | Stack phases (phase-2 from phase-1) | Branch all phases from main | | Use conventional commit format | Write vague commit messages | | Push to the branch you're working on | Push to random branches | | Set correct PR base for stacked PRs | Base all PRs on main | ``` --- ## Implementation Steps 1. **Create skill directory:** ```bash mkdir -p .claude/skills/git-workflow ``` 2. **Create SKILL.md** with content above 3. **Update agent configuration** to include the skill: ```yaml --- name: tinker-worker description: Implement features, fix bugs, write code skills: git-workflow tools: Read, Write, Edit, Bash, Grep, Glob --- You are the Tinker Worker agent... When performing any git operations, always follow the git-workflow skill. ``` ## Acceptance Criteria 1. `.claude/skills/git-workflow/SKILL.md` created 2. Agent configuration updated to include skill 3. When asked to "fix PR #33", agent checks out the correct branch and pushes to it 4. When asked to commit, agent uses conventional commit format 5. When creating phased work, agent creates stacked branches 6. No more random branches 7. No more pushes to wrong branches ## Files to Create - `.claude/skills/git-workflow/SKILL.md` ## Files to Modify - Agent configuration files (add `skills: git-workflow`) ## Notes - One skill covers all git scenarios - Skills are auto-invoked based on keywords in the description - Test by giving agents specific git tasks and verifying correct behavior
Working memory
{}
Pull request url
Avo
· © 2026 AvoHQ ·
v3.27.0
Close modal
Are you sure?
Yes, I'm sure
No, cancel