make-feature

Creates isolated git worktrees and feature branches for gated TDD development workflows.

9|Updated Jul 8, 2026
One-click install
npx skills add https://github.com/jerrylin96/dotgemini --skill make-feature-jerrylin96
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: make-feature
Source: https://github.com/jerrylin96/dotgemini/tree/main/skills/make-feature
Command: npx skills add https://github.com/jerrylin96/dotgemini --skill make-feature-jerrylin96

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? It prevents unreviewed or untested code changes from polluting your primary working branch by enforcing an isolated git worktree workflow with mandatory spec, plan, test, and review gates before any merge. ## Core Features & Use Cases - Isolated Worktree Development: Automatically creates a gemini/<feature-name>-<hash> branch in a dedicated worktree under ~/.gemini/tmp/worktrees/, keeping the user's primary checkout untouched. - Milestone Gate Pipeline: Enforces sequential gates — interactive alignment, spec review, plan review, RED test review, adversarial code review, and human signoff — with subagent-driven adversarial reviews at each stage. - TDD Enforcement with Remote Sync: Commits and pushes failing RED tests before GREEN implementation, creating cryptographic proof of test-driven rigor for external reviewers and CI bots. - Use Case: When asked to add a new API endpoint, the agent spins up an isolated worktree, drafts a spec and plan for human approval, writes failing tests first, implements the feature, passes adversarial code review, and hands off a clean branch for manual PR creation and merge. ## Quick Start Ask the agent to start a new feature with /make-feature followed by a short description of the change you want to build.

Frequently Asked Questions about make-feature

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I develop a feature in an isolated git worktree?

Invoke /make-feature with a feature description. It creates a gemini/<feature-name>-<hash> branch in a worktree under ~/.gemini/tmp/worktrees/, drafts a spec and plan, and guides the change through review gates before human merge.

How does the skill enforce test-driven development?

It requires writing a failing RED test suite first, which is reviewed by an adversarial test subagent, committed, and pushed to origin before any implementation code is written. GREEN code must then achieve a 100% pass rate verified by empirical test runs.

Can the AI agent merge the feature branch into main?

No. Pull request creation and merging into the base integration branch are always performed manually by the human engineer. The agent may only rebase its own feature branch inside the isolated worktree to resolve drift.

What happens to spec and plan files before merge?

All ephemeral artifacts (spec.md, plan.md, review prompts, scorecards) live only in the feature worktree and are purged via git rm before signoff, so no planning files pollute the primary branch after merge.

When should I use /make-feature heavy mode?

Use heavy mode for complex multi-slice features. It executes task slices sequentially with subagents, enforcing a strict two-stage commit cadence per slice: RED tests committed and pushed first, then GREEN implementation after slice code review.

What are the limitations of this worktree workflow?

It requires a POSIX environment (macOS or Linux) and git worktree support, and it does not apply to non-repository files such as scratch notes or agent configuration artifacts. Offline mode degrades remote review gates to local diff inspection.