pi-agile-coach

Universal agile coach for pi coding agent and Claude Code — structures any project into sprints, guides teams through the full delivery cycle, and adapts to any technology or project type.

What it does

🧠 Skill: agile-coach

Loaded automatically when you need to organize a project. Works for any project type — software, design, hardware, marketing campaigns, factory processes, operations, or research. Without being asked, the coach will:

• Detect the project type and adapt vocabulary — no technical jargon in non-tech projects
• Detect the tech stack automatically (Odoo, Spring, Angular, React, Django, Node, FastAPI, and more)
• Propose the right depth — Lite (1-2 people, short projects), Standard (2-5 people), or Full (5+ people, complex projects)
• Guide through structured phases — Init → Explore → Proposal → Spec → Design → Tasks → Apply → Verify → Archive
• Generate all project artifacts — backlog, sprint planning, risk register, burndown, retro, engram memory
• Integrate with any skill — works alone or alongside Superpowers, workflow-odoo, code-review-excellence, or any custom skill

📋 Phase System

INIT → EXPLORE → PROPOSAL → SPEC → DESIGN → TASKS → APPLY → VERIFY → ARCHIVE
                                                      ↑________________________↑
                                               (sprint loop + continuous REFINEMENT)

Each phase has:

• Gate conditions — can't skip without explicit acknowledgment
• Artifacts — concrete files generated at each step
• Contract envelope — structured output with status, risks, and next steps

🔧 Tech Stack Adapters

When the project is software, the coach decomposes user stories into layer-specific tasks automatically:

Stack	Tasks generated
Odoo	Model, View (form/list/kanban), Wizard, Report, Cron, Security — delegates to workflow-odoo
Spring Boot	Entity, Repository, Service, Controller, DTO, Unit Test, Integration Test
Angular	Module, Component (smart/presentational), Service, Guard, Pipe, Spec
React/Next.js	Component, Hook, Context, API Route, Store, Test (RTL/Vitest)
Django	Model, Migration, Serializer, ViewSet, URL, Admin, Test
Node/Express	Router, Controller, Service, Middleware, Model, Validator, Test
FastAPI	Router, Schema (Pydantic), Service, Repository, Model, Migration, Test

📦 Artifacts Generated

Every sprint produces structured Markdown and JSON files:

.sdd/config.json           ← project configuration (persistent)
.sdd/phase-status.json     ← current phase state
.sdd/engram.json           ← decision memory across sessions
docs/scrum/project-charter.md
docs/scrum/product-backlog.md
docs/scrum/risk-register.md
docs/scrum/sprint-NNN/planning.md
docs/scrum/sprint-NNN/apply-log.md
docs/scrum/sprint-NNN/verify-report.md
docs/scrum/sprint-NNN/review.md
docs/scrum/sprint-NNN/retro.md
docs/sdd/explore.md
docs/sdd/proposal.md
docs/sdd/spec.md
docs/sdd/design.md
docs/architecture/ADR-NNN-*.md

🤝 Skill Integrations

The coach is a generic orchestrator — it never depends on external skills but enriches when they're available:

Function	Compatible skills	Built-in fallback
Brainstorming	superpowers:brainstorming	A/B/C proposals inline
Implementation plan	superpowers:writing-plans	Detailed planning.md
Execution with sub-agents	superpowers:subagent-driven-development	Sequential guided execution
Odoo development	workflow-odoo	Generic task decomposition
Code review	code-review-excellence	Structured review checklist
Verification	superpowers:verification-before-completion	Evidence checklist (H-VERIFY)

Install

# With npx (recommended — works with pi and Claude Code)
npx skills add EdgarNaranjo/pi-agile-coach -g -y

# With pi from npm
pi install npm:pi-agile-coach

# With pi from GitHub
pi install git:github.com/EdgarNaranjo/pi-agile-coach

# Try without installing
pi -e git:github.com/EdgarNaranjo/pi-agile-coach

What makes this different

Most project management tools for AI agents are templates — they give you a backlog format or a sprint table. This skill gives the agent judgment: when to skip phases, how to adapt to a factory process vs. a marketing campaign, how to detect that a sprint is at risk, and how to integrate seamlessly with other skills without being told.

The Engram Memory is unique — decisions, preferences, and fixes are stored in .sdd/engram.json and recovered automatically in new sessions, so the agent never forgets what was decided three sprints ago.

Gentle AI behaviors

The coach never overwhelms. Built-in rules enforce:

• Max 3 questions per turn
• Announces before generating large outputs
• Adapts vocabulary to the project type (no "commit" or "deploy" in a marketing campaign)
• Reminds the user they're in control after every long response
• Improves existing structure — never replaces without asking

Compatible with

• pi coding agent
• Claude Code (npx skills add)
• Any project type: software, design, hardware, marketing, research, operations

Works well alongside

• pi-odoo-workflow — Odoo 18/19 technical tasks
• superpowers-extended-cc — execution engine with sub-agents, TDD, and verification
• code-review-excellence — structured code reviews

References included

File	Content
scrum-guide-summary.md	Scrum Guide 2020: roles, events, artifacts, timeboxes
scrum-ceremonies.md	DoR, Risk Register, Burndown, Refinement, Retro methods, Stakeholder Map, onboarding
tech-adapters.md	Task decomposition by stack (Odoo, Spring, Angular, React, Django, Node, FastAPI)
phase-contracts.md	Detailed artifact contracts per phase
skill-integration-protocol.md	Generic protocol for integrating any external skill
harnesses-detail.md	Secondary harnesses: sub-agent isolation, scope review, model routing, backup/rollback
architecture-checklist.md	Architecture decision checklist for software projects

License

LGPL-3.0-or-later — see LICENSE

---

Links

• npm: pi-agile-coach
• GitHub: EdgarNaranjo/pi-agile-coach
• pi.dev: pi-agile-coach
• Changelog: CHANGELOG.md