AgentCommunity★ 500

Agentic Sprint

Updated Jul 7, 2026

A Claude Code plugin by damienlaine (part of the Agentic Forge ecosystem) that runs autonomous spec-driven sprints. A project-architect agent coordinates implementation agents (python-dev, nextjs-dev, cicd-agent, allpurpose-agent, website-designer) and testing agents (qa-test-agent, ui-test-agent). The sprint loops up to 5 iterations, shrinking completed work from specs each pass until all gates close.

View on GitHub

What it does

/sprint
Runs the full convergent sprint loop: architect analyzes specs, implementation agents build, testing agents validate, loop repeats until all gates close.
/sprint:new
Creates .claude/sprint/N/specs.md for the next sprint with a structured template.
/sprint:setup
Interactive onboarding that creates project-goals.md (business vision) and project-map.md (technical architecture) through guided questions.
/sprint:test
Opens a live browser session for manual UI testing outside of a sprint; saves a report that feeds into the next sprint run.
/sprint:generate-map
Generates or updates project-map.md by analyzing the current codebase structure.
/sprint:clean
Removes old sprint directories to keep the workspace tidy.

Files (1)

OVERVIEW.mdprimary · markdown · 2.0 KB

## What it does

A Claude Code plugin implementing convergent spec-driven sprints. The /sprint command drives a self-iterating state machine: read specs.md → architect creates api-contract.md → implementation agents build in parallel → testing agents validate → architect removes completed work from specs and iterates. The picture starts noisy; each pass reduces noise until only working code remains. Most sprints converge in under 5 iterations; if they don't, the system pauses and asks what to do.

Two persistent "second brain" files anchor agent memory across sprints:
- .claude/project-goals.md — product vision and constraints (you maintain)
- .claude/project-map.md — architecture, API surface, and file layout (architect maintains)

Agents read these instead of scanning the whole codebase, keeping context focused.

Implementation agents: python-dev (FastAPI/PostgreSQL), nextjs-dev (Next.js 16/React 19), cicd-agent (GitHub Actions/Docker), allpurpose-agent (any stack), website-designer (static HTML/CSS).

Testing agents: qa-test-agent (pytest/jest/vitest), ui-test-agent (E2E via Chrome browser MCP), nextjs-diagnostics-agent (runtime monitoring for Next.js).

Manual testing mode: set "UI Testing Mode: manual" in specs.md to open a Chrome tab, interact yourself, and feed your session report back to the architect before the next iteration.

## Install

From Agentic Forge marketplace (recommended):

/plugin marketplace add damienlaine/agentic-forge
/plugin install sprint

For local development:

git clone https://github.com/damienlaine/agentic-sprint.git
claude --plugin-dir ./agentic-sprint

## When to use

Use /sprint:setup first to create project-goals.md and project-map.md, then /sprint:new to create specs.md. Keep specs small — multiple small sprints beat one large one. Commit before each sprint run so you have a rollback point. Good for: new features with clear acceptance criteria, iterative UI work with real browser validation, and full-stack features where backend and frontend need coordinated implementation.