This repository contains instruction for coding agents.
Find a file
Repository files (latest commit first)
Filename Latest commit message Latest commit date
2026-06-13 08:59:49 +02:00
skills Added rule about comments. 2026-05-14 11:01:28 +02:00
workflow Rename and refactor of the workflow and skills. 2026-05-14 09:25:40 +02:00
.gitignore Initial commit. 2026-05-09 11:00:44 +02:00
README.md Updated README. 2026-06-13 08:59:49 +02:00

Coding agent workflow

A worflow for implementing features and fixing bugs using coding agents development, based on Boris Tane's workflow.

Overview

This repository documents our approach to working with coding agents, emphasizing a structured three-phase process:

  1. Research - Deep analysis of the codebase
  2. Planning - Iterative refinement of implementation approach
  3. Implementation - Executing the approved plan

Core Principle

Never let the AI write code until you've reviewed and approved a written plan.

Separating thinking from execution prevents wasted effort, keeps the developer in control of architectural decisions, and produces better outcomes.

Quick Start

To start a new task, use the /start-task command which will guide you through:

  • Selecting task type (feat, fix, chore, test)
  • Linking to a JIRA ticket (optional)
  • Creating a properly formatted git branch
  • Setting up research and plan documents

Documentation

Using the skill templates

The skills under skills/ are templates, not ready-to-run skills. They are parameterized (see PARAMETERS.md) so each repository can supply its own values for things like the default branch and the task-artifacts folder.

To use them in a consuming repository, add this repo as a submodule and create a thin wrapper skill under the repository's own .claude/skills/<name>/SKILL.md. The wrapper:

  1. Declares the skill name/description so the agent can discover it.
  2. Provides the values for the template's required parameters (default_branch) and any overrides for optional ones (task_folder_path).
  3. Delegates to the template by path — e.g. it tells the agent to read submodules/agent-workflow/skills/start-task/SKILL.md and execute it with the parameters substituted for ${param} / {{param}}.

Because the wrapper references the template by path rather than copying it, updates to the template flow through automatically. Keep wrappers thin: they inject parameters only and add no workflow logic.

See EXAMPLE-USAGE.md for worked examples. The Himarked repository is a live example of this pattern — its .claude/skills/start-task and .claude/skills/resume-task wrappers bind these templates to master / doc/tasks/.

Repository Structure

├── README.md              # This file
├── workflow/
│   └── research-plan-implement-workflow.md  # Core workflow documentation
└── skills/
    ├── start-task/
    │   └── SKILL.md       # Task initialization wizard
    └── resume-task/
        └── SKILL.md       # Resume interrupted task