@mark-gozner/aigile-method
v1.5.1
Published
AIgile Development Method by Mark Gozner - AI-powered development with custom agents (Claude Code, GitHub Copilot, Google Antigravity & Kiro), skills, workflows, handoffs, and delegation
Maintainers
Readme
AIgile Development Method
A pragmatic, agent-driven agile method designed for IDE-native AI assistants. It ships with ready-to-run agents, executable tasks, MCP server wiring, and docs/templates to bootstrap product delivery.
Supported platforms: Claude Code, GitHub Copilot, Google Antigravity, and Kiro — one installer, the same agent team everywhere.
Roles covered: Analyst, Architect, Dev, React Engineer, DevOps, Security, QA, Automation Tester, PM, PO, SM, UX-Expert, UI-Expert, Code Tour, plus two orchestrators (automatic and human-in-the-loop).
Installation
Quick Start (npm)
Prereqs: Node.js 18+ (20+ recommended)
Install AIgile into your project:
npx @mark-gozner/aigile-method installThe installer detects which IDEs your project already uses (.kiro/, .github/, .agent/, .claude/ or CLAUDE.md, or being run from a Kiro or Claude Code session) and pre-selects those targets. In a Kiro or Claude Code workspace you get that platform's agent team, context, and skills without passing any flags. Projects with none of those markers default to GitHub Copilot, as before.
The interactive installer asks which platforms to target: Claude Code, GitHub Copilot, Google Antigravity, Kiro (any combination). You can also skip the prompt with flags:
npx @mark-gozner/aigile-method install --for-claude --yes # Claude Code only
npx @mark-gozner/aigile-method install --for-copilot --yes # GitHub Copilot only
npx @mark-gozner/aigile-method install --for-antigravity --yes # Antigravity only
npx @mark-gozner/aigile-method install --for-kiro --yes # Kiro only
npx @mark-gozner/aigile-method install --for-copilot --for-antigravity --for-kiro --for-claude --yesClaude Code plugin (no npm install)
Claude Code can install the whole agent team straight from this repository:
/plugin marketplace add Gmark2000/AIgile-Method
/plugin install aigile@aigile-methodThat gives you the 17 subagents plus every workflow namespaced under /aigile: — /aigile:method for routing help, /aigile:dev-develop-story, /aigile:qa-quality-gate, /aigile:workflow-greenfield-fullstack, and so on. The plugin carries its own copy of the task, template, and checklist library, so there is nothing else to install. Use the npm installer instead when you want the framework files checked into your repo as .aigile-core/, or when you want .mcp.json wiring.
Add MCP config (optional):
npx @mark-gozner/aigile-method install --with-mcpBundle (optional text bundles to dist/):
npx @mark-gozner/aigile-method bundleWhat this gives you in the project:
.aigile-core/– roles, tasks, templates, checklists, data, workflows.github/agents/– GitHub Copilot custom agents (--for-copilot).agent/rules/,.agent/workflows/,GEMINI.md– Google Antigravity (--for-antigravity).kiro/agents/,.kiro/steering/,.kiro/skills/– Kiro custom agents, steering, and Agent Skills (--for-kiro).claude/agents/,.claude/skills/,CLAUDE.md,.claude/settings.json– Claude Code subagents, skill workflows, and project context (--for-claude).vscode/mcp.json,.kiro/settings/mcp.json, or.mcp.json– only if you pass--with-mcp.aigile-manifest.json– manifest summary
To update later (e.g., after a version bump):
npx @mark-gozner/aigile-method install --yesClone-and-Run (No npm Install Needed)
# 1) Clone the repository once
git clone https://github.com/Gmark2000/AIgile-Method.git ~/aigile-method
# 2) In any target project
cd /path/to/your/project
node ~/aigile-method/tools/cli.js install
# Optional: include MCP config
node ~/aigile-method/tools/cli.js install --with-mcp
# Bundle (team + agents)
node ~/aigile-method/tools/cli.js bundleOptional convenience:
- Bash:
export AIGILE_HOME="$HOME/aigile-method"thennode "$AIGILE_HOME/tools/cli.js" install - PowerShell:
$env:AIGILE_HOME = "$HOME/aigile-method"thennode "$env:AIGILE_HOME/tools/cli.js" install
Custom Agents
Every supported platform gets the full AIgile agent team out of the box, generated from the same source definitions in core/agents/.
Claude Code (.claude/agents/, .claude/skills/, CLAUDE.md)
- One subagent per role, delegated to with the
Agenttool (subagent_type: "qa") or addressed directly as@agent-qa - Specialists inherit the full session toolset, including whatever MCP servers you have configured — only the two orchestrators are restricted, to
Read, Glob, Grep, Agent, TodoWrite, AskUserQuestion, WebFetch, WebSearch, so "delegation-only" is enforced by the toolset and not just by the prompt - Role-relevant AIgile skills are preloaded per agent via the
skills:frontmatter field - Copilot/VS Code tool names in the source definitions are translated to their Claude Code equivalents (
runSubagent→Agent,vscode_askQuestions→AskUserQuestion,manage_todo_list→TodoWrite) CLAUDE.mdgets an AIgile section between<!-- AIGILE:START -->/<!-- AIGILE:END -->markers, so re-running the installer never disturbs anything you wrote around it.claude/settings.jsonis merged, not overwritten — the installer only adds read permissions for.aigile-core/.mcp.jsonwhen combined with--with-mcp
GitHub Copilot (.github/agents/*.agent.md)
- Strong system directives for persona maintenance
- YAML frontmatter with tools, handoffs, and model configuration
- Subagent delegation support via
runSubagent
Google Antigravity (.agent/rules/, .agent/workflows/, GEMINI.md)
- One rule file per agent plus one workflow file per agent command
GEMINI.mdas the project-level agent index
Kiro (.kiro/agents/, .kiro/steering/, .kiro/skills/)
- One custom agent per role, so the whole team shows up in Kiro's agent selector and is available for subagent delegation
- Tool access declared with Kiro tool categories (
read,write,shell,web,context,subagent,spec) plus@builtinandincludeMcpJson: true, so subagent delegation works across Kiro builds - Orchestrators are delegation-only by prompt, not by tool restriction — they need
@builtinfor the subagent tool to be reliably present .kiro/steering/aigile-method.md— always-included framework context.kiro/steering/aigile-<agent>-workflows.md— manual steering files that show up as slash commands (e.g./aigile-dev-workflows).kiro/skills/— AIgile Agent Skills (see below).kiro/settings/mcp.jsonwhen combined with--with-mcp
Each generated agent keeps a collapsible appendix with the original agent source for transparency. If an agent stops behaving according to its role, re-run the installer to regenerate it.
Agent Skills (Kiro and Claude Code)
--for-kiro and --for-claude install four AIgile skills (into .kiro/skills/ and .claude/skills/), following the open Agent Skills standard. They load on demand when a request matches their description, and each generated agent lists the skills relevant to its role.
| Skill | What it covers |
|-------|----------------|
| aigile-orchestration | Routing matrix for the whole team, delegation protocol, sequencing rules, when to escalate |
| aigile-story-delivery | Story readiness, task breakdown, implementation loop, definition of done, QA handoff |
| aigile-quality-gate | Evidence gathering, risk scoring, and PASS / CONCERNS / FAIL / WAIVED decisions |
| aigile-spec-bridge | Mapping spec phases (requirements, design, tasks) onto AIgile artifacts and roles |
Two of these describe platform mechanics rather than pure process, so Claude Code gets hand-written versions from core/skills-claude/ instead of a machine translation of the Kiro wording.
Workflow skills (Claude Code)
--for-claude additionally turns every agent command and every delivery workflow into its own invocable skill, so the whole method is reachable from the / menu:
| Skill | What it does |
|-------|--------------|
| /aigile | Routing index — which agent, which workflow, and the sequencing rules |
| /aigile-<agent>-<command> | One per agent command (33 of them): /aigile-dev-develop-story, /aigile-po-create-story, /aigile-qa-quality-gate, ... |
| /aigile-workflow-<name> | One per delivery workflow (6): the greenfield and brownfield sequences, rendered as a step table with the agent, artifact, and preconditions for each step |
Each workflow skill sets disable-model-invocation: true — they run when you ask for them, not on Claude's initiative — and accepts free-form context as $ARGUMENTS.
Community skills
AIgile also ships a curated registry of third-party skills. Nothing third-party is bundled in the npm package — skills:add downloads each skill from its source repository at your request and writes an ATTRIBUTION.md recording source, author, license, and retrieval date.
npx @mark-gozner/aigile-method skills:list # browse the registry
npx @mark-gozner/aigile-method skills:add # recommended set
npx @mark-gozner/aigile-method skills:add caveman grill-me # specific skills
npx @mark-gozner/aigile-method skills:add --all --global # everything, into ~/.kiro/skills
npx @mark-gozner/aigile-method install --for-kiro --with-community-skills --yesThe registry currently covers caveman (token-thrifty replies), grill-me, junior-to-senior, loop-factory, context-canary, interface-kit (all MIT, by Julius Brussee), and webapp-testing, mcp-builder, skill-creator, frontend-design from Anthropic's public skills repo. Set GITHUB_TOKEN if you hit GitHub API rate limits. These are third-party instructions that will steer your agent — review them before use.
Using AIgile in your IDE
Claude Code
- Run
/aigilefor routing help, or ask directly ("use the po subagent to draft stories for FEAT-12") - Pin a specific specialist with
@agent-dev,@agent-qa, … when you do not want Claude choosing for you - Type
/aigile-to browse the workflows, e.g./aigile-dev-develop-story STORY-142 - Hand a whole piece of work to a coordinator by starting the session as one:
Or pin it for the project withclaude --agent aigile-orchestrator # coordinates autonomously claude --agent aigile-manual-orchestrator # asks before every delegation"agent": "aigile-orchestrator"in.claude/settings.json. Spawning an orchestrator as a subagent also works — subagents nest up to three layers below the main conversation — but you only see its final summary - Run
/agentsto inspect or tweak the installed team
GitHub Copilot
- Pick an agent for a role (e.g., PO, PM, Architect) from the Agents dropdown in Copilot Chat
- Ask for help with a specific command (e.g., "*help" or "*create-epic")
Kiro
- Open the workspace and trust it when prompted (required for
.kiro/agents/to load) - Choose a role from the agent selector, or ask Kiro to run one as a subagent ("Use the
posubagent to draft stories") 2b) Runaigile-orchestrator/aigile-manual-orchestratoras the main agent when you want it to coordinate the team. They declare thesubagenttool, but a subagent cannot spawn further subagents, so an orchestrator launched as a subagent can only hand briefs back - Type
/in chat to pull in a workflow or skill (e.g./aigile-qa-workflows,/aigile-quality-gate) - Manage everything from the Agent Steering & Skills panel in the Kiro side bar
Antigravity
- Open the project — rules and workflows are picked up from
.agent/
In all cases, when you execute a task the agent loads the exact file from .aigile-core/tasks/... and follows it step by step.
Key principles:
- Tasks are executable instructions; follow them exactly.
- Elicitation steps are mandatory; agents will ask for missing inputs.
- Save-to-docs prompts help persist artifacts under
docs/for traceability.
CLI Commands
After installation you can use:
| Command | Purpose |
|---------|---------|
| npx @mark-gozner/aigile-method install --yes | Re-install / regenerate agents |
| npx @mark-gozner/aigile-method install --for-claude --yes | Install the agent team for Claude Code |
| npx @mark-gozner/aigile-method install --for-kiro --yes | Install the agent team for Kiro |
| npx @mark-gozner/aigile-method install --for-antigravity --yes | Install the agent team for Antigravity |
| npx @mark-gozner/aigile-method install --with-mcp --yes | Include MCP config (.vscode/mcp.json, .kiro/settings/mcp.json, or .mcp.json for Claude Code) |
| npx @mark-gozner/aigile-method bundle | Build text bundles and regenerate the Claude Code plugin in plugin/ |
| npx @mark-gozner/aigile-method status | Show installed version, agents, agent count, skills, MCP info |
| npx @mark-gozner/aigile-method list:agents | List installed agent IDs |
| npx @mark-gozner/aigile-method skills:list | List AIgile skills and the curated community registry |
| npx @mark-gozner/aigile-method skills:add [ids] | Fetch community skills into .kiro/skills/ |
Consumer package.json helper scripts are auto-added (unless AIGILE_ADD_SCRIPTS=0) for convenience:
"scripts": {
"aigile:install": "aigile-method install --yes",
"aigile:bundle": "aigile-method bundle",
"aigile:status": "aigile-method status",
"aigile:list": "aigile-method list:agents"
}Remove or customize as needed.
The installer writes an .aigile-manifest.json summarizing what was installed (version, agents, agent count, MCP components). Re-running install updates it.
Publishing (npm)
This package is configured to publish to the public npm registry.
1. Authentication
Use standard npm authentication:
npm loginOr set NPM_TOKEN environment variable for CI/CD.
2. Build & Validate
npm install
npm run validate # builds bundles and validates
npm run pack:dry-run # view what would be published3. Publish
Patch / minor / major helpers are available:
npm run release:patch # bump patch + publish
# or
npm run release:minor
npm run release:majorAlternatively run manually:
npm version patch # or minor / major
npm publishThe prepublishOnly script automatically runs the bundle build.
MCP Servers (optional)
AIgile can use local or remote MCP servers set up via mcp/servers.json. See the dedicated guide: docs/mcp-servers.md.
This repository includes examples for:
- Atlassian (Jira + Confluence): read-only defaults with tokens via prompts
- GitHub.com
- Playwright (browser automation and testing)
- Memory (long-lived notes)
- Sequential Thinking (multi-step orchestration)
- Context7 (external knowledge grounding)
Your IDE may translate these into tool lists automatically. Provide tokens/URLs when prompted.
Where to get API keys and tokens
- Context7: Create or retrieve an API key from your Context7 account dashboard.
- Atlassian Cloud: Create an API token at https://id.atlassian.com/manage-profile/security/api-tokens
- GitHub: Create a PAT at https://github.com/settings/tokens or fine-grained tokens at https://github.com/settings/personal-access-tokens
- SonarQube: Generate tokens in your profile on the SonarQube server.
Core Configuration (core/core-config.yaml)
Important keys used by tasks:
- devStoryLocation: docs/stories
- qa.qaLocation: docs/qa
- prd: prdFile, prdSharded, prdShardedLocation
- architecture: architectureFile, architectureSharded, architectureShardedLocation
Tasks and scaffolding respect these paths for saving outputs and docs.
Output Docs Conventions
- Epics →
docs/epics/epic-{key-or-slug}.md - Stories →
docs/stories/story-{key-or-slug}.md - Tasks →
docs/tasks/task-{key-or-slug}.md
Each file includes cross-links:
- Stories link back to their epic and to related stories.
- Tasks link to their parent story and epic (when available).
Task Catalog by Role
Below are representative task files you can run via an agent:
Analyst
core/tasks/analyst/facilitate-brainstorming-session.mdcore/tasks/analyst/market-research-from-context7.mdcore/templates/project-brief-tmpl.yaml(template)core/templates/market-research-tmpl.yaml(template)
Architect
core/tasks/architect/code-arch-review-with-github.mdcore/tasks/architect/document-project.mdcore/tasks/architect/create-architecture-doc.md
Dev
core/tasks/dev/check-story-implemented.mdcore/tasks/dev/explain-story-from-jira.mdcore/tasks/dev/implement-freeform-work-item.mdcore/tasks/dev/implement-story-from-jira.mdcore/tasks/dev/implement-unit-tests.mdcore/tasks/dev/execute-checklist.mdcore/tasks/dev/validate-next-story.md
PM
core/tasks/pm/sync-jira-backlog.mdcore/tasks/pm/create-prd-doc.mdcore/tasks/pm/create-next-story.md
PO
core/tasks/po/create-jira-epic-from-confluence.mdcore/tasks/po/create-stories-from-epic.mdcore/tasks/po/create-tasks-for-story.mdcore/tasks/po/create-jira-story-from-confluence.mdcore/tasks/po/create-jira-story-from-text.mdcore/tasks/po/groom-jira-story.mdcore/tasks/po/create-prd-doc.md
QA
core/tasks/qa/gate.mdcore/tasks/qa/sonarqube-hotspot-review.mdcore/tasks/qa/test-design.mdcore/tasks/qa/verify-jira-story-e2e.mdcore/tasks/qa/review-story.md
SM
core/tasks/sm/standup-digest.md
UI / UX
core/tasks/ui/figma-audit-design-system.mdcore/tasks/ux/front-end-spec-from-design.md
Try it Now
# 1) Install into your repo
npx @mark-gozner/aigile-method install --with-mcp
# 2) Open a PO agent in your IDE and run a flow
# Example: create an Epic from Confluence, then generate stories and tasks
# *create-epic
# *create-stories
# *create-tasks
# 3) Check generated docs
# docs/epics/*.md
# docs/stories/*.md
# docs/tasks/*.md
# 4) Bundle for external use (optional)
npx @mark-gozner/aigile-method bundleContributing, Security, and License
- Contributing: see
CONTRIBUTING.md - Code of Conduct: see
CODE_OF_CONDUCT.md - Security policy and private reporting: see
SECURITY.md - Roadmap and priorities: see
ROADMAP.md - Changelog: see
CHANGELOG.md - License: see
LICENSE.md
