feature-loop-scheme
npx machina-cli add skill YoniChechik/claude-code-config/feature-loop-scheme --openclawPrerequisites: Must be in a feature clone directory.
Process
Step 1: Gather Context
Use the explorer subagent to gather context about the codebase relevant to the feature:
- Explore existing code patterns and architecture
- Identify related files and components
- Understand dependencies and integration points
- Set thoroughness level to "medium" for balance between speed and depth
This context will inform the planning phase.
Step 2: Plan or Analyze
Determine feature name from branch: FEATURE_NAME=$(git rev-parse --abbrev-ref HEAD)
- If plan_$FEATURE_NAME.md doesn't exist → Create plan using planner subagent with opus model, ask questions if needed
- If plan_$FEATURE_NAME.md exists → Analyze current progress compared to origin/main, examine plan_$FEATURE_NAME.md and documentation, identify next steps
Step 2.5: Commit Plan (if just created)
If plan was just created in Step 2, commit and push it immediately so it survives session crashes:
git add plan_$FEATURE_NAME.md && git commit -m 'Add feature plan' && git push
Step 3: Create Task List
Create a structured task list based on the plan or next steps:
- Break down into actionable tasks
- Use the TaskCreate tool to create the task list
- Each task should be specific and measurable
- Mark the first task as "in_progress" to begin work
Step 4: Implement
- Use coder-agent to write code
- If problems occur, use coder-agent to fix them
- After each significant change, commit and push (main agent does this directly)
Step 5: Quality
Run quality skill to fix code style, types, and remove AI slop.
Step 6: Review
Use Task tool with subagent_type="reviewer-agent" for final code review and validation.
Step 7: Summary
Report findings and confirm ready for PR (or list remaining issues).
Source
git clone https://github.com/YoniChechik/claude-code-config/blob/main/skills/feature-loop-scheme/SKILL.mdView on GitHub Overview
feature-loop-scheme orchestrates the end-to-end feature development process used by feature-new and feature-continue. It guides you through gathering codebase context, planning, creating actionable tasks, implementing with a coder-agent, enforcing quality, and final review for PR readiness. Prerequisites: you must be in a feature clone directory.
How This Skill Works
It begins with the explorer subagent gathering context about code patterns, components, dependencies, and integration points. It then checks for plan_$FEATURE_NAME.md; if missing, it uses the planner subagent with the opus model to create the plan; if it exists, it analyzes progress against origin/main and next steps. Next, TaskCreate builds a structured, measurable task list and marks the first task as in_progress. Implementations and fixes are done by the coder-agent, followed by quality checks with the quality skill, and final validation by the reviewer-agent before PR.
When to Use It
- Starting a brand-new feature on a branch and needs a structured workflow
- Resuming an in-progress feature with an existing plan to align with origin/main
- Coordinating work that touches multiple components or services with dependencies
- Onboarding a new teammate to an ongoing feature to align on tasks
- Requiring formal validation before PR after a large refactor or architectural change
Quick Start
- Step 1: Ensure you are in the feature clone directory and on the feature branch
- Step 2: Run Gather Context, then Plan or Analyze, and create the Task List
- Step 3: Implement with coder-agent, run Quality, complete Review, and prepare PR
Best Practices
- Be in a feature clone directory before starting
- Create and commit plan_$FEATURE_NAME.md early to survive crashes
- Break work into specific, measurable tasks using TaskCreate
- Run the quality skill after significant changes to fix style and types
- Use a reviewer-agent for final validation before PR
Example Use Cases
- Example: feature-new adds user profile pages across UI and API
- Example: feature-continue enhances search with ranking and facets
- Example: feature-loop coordinates billing integration with existing modules
- Example: feature-new introduces multi-tenant support with feature flags
- Example: feature-continue consolidates logging and observability during refactor
