dev-workflow

Executes keel development tasks via dual-agent TDD with adversarial review and atomic commits.

Updated Jan 12, 2026
One-click install
npx skills add https://github.com/ab300819/keel-workflow --skill dev-workflow-ab300819
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: dev-workflow
Source: https://github.com/ab300819/keel-workflow/tree/main/skills/dev-workflow
Command: npx skills add https://github.com/ab300819/keel-workflow --skill dev-workflow-ab300819

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? It turns documented development tasks into verified code by enforcing a structured test-driven workflow, preventing untested implementations, undocumented changes, and skipped reviews from reaching your repository. ## Core Features & Use Cases - Dual-Agent TDD Isolation: A Test Agent writes skeletons and tests while a separate Impl Agent writes implementations, with an information barrier so neither sees the other's source material. - Risk-Tiered Review Profiles: fast, guarded, and audit profiles control when adversarial code reviews (internal role-play phases plus external codex-based review) run inline or defer to a review drain. - Batch Orchestration & Resume: Execute single tasks, ranges, or feature sets with dependency topological sorting, checkpoint files, and a six-step breakpoint resume state machine. - Use Case: Run /dev-workflow T-01~T-05 to implement five planned tasks from your dev-tasks document, each producing atomic code and documentation commits with verified acceptance criteria. ## Quick Start Ask the assistant to execute task T-03 from your dev-tasks document using the dev-workflow skill with test-driven development.

Frequently Asked Questions about dev-workflow

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

FAQPage Schema
How do I run test-driven development tasks with an AI coding agent?

Invoke dev-workflow with a task ID like T-03 or a range like T-01~T-05. A Test Agent writes skeletons and failing tests first, then a separate Impl Agent implements code until tests pass, with the orchestrator verifying each gate.

How does dual-agent isolation work in AI code generation?

The Test Agent never reads existing implementation code, and the Impl Agent never reads test case documents or requirement files. The orchestrator diffs test files after implementation, treating any test modification as a blocking violation.

Can I run batch development tasks unattended?

Yes, the headless mode automates all decision points using a predefined policy table with fail-fast semantics. It requires a clean workspace, enforces safety invariants like never committing failing tests, and outputs a resume command on failure.

What happens when an interrupted development task resumes?

A six-step detection pipeline checks document status, git history, workspace changes, and evidence completeness before resuming. Completed tasks with verifiable evidence are skipped; incomplete ones resume from the exact failed step.

When should I use dev-flow instead of dev-workflow?

Use dev-flow when you have no keel documentation and want to develop directly from a plan or task description. Use dev-workflow when tasks exist in a dev-tasks document with requirement traceability and acceptance criteria.