Guide

How to Work with DevCrew

The core of DevCrew is INSTRUCTIONS.md — it turns any AI into a dev team that follows a structured workflow. After installation, there are two ways to invoke Skills:

1. Natural Language (Recommended)

Describe what you need in AI chat. The AI automatically follows the PDEVI workflow:

You: I need to add auth middleware to the API
AI:  [PdM] Creating change add-api-auth, mode: Standard …

Common natural language triggers:

What you say	Skill triggered
"Let's make a plan" / "I want to add a feature"	plan
"What's the status?" / "Show progress"	status
"Run a checkpoint" / "Audit this"	checkpoint
"Archive it" / "This one's done"	release

2. Skill Commands

Use skill commands directly in your AI chat:

/crew.plan        # Create a change plan
/crew.status      # Check progress
/crew.checkpoint  # Phase audit
/crew.release     # Archive changes

Mix and match: Use natural language for daily collaboration, skill commands for quick operations.

---

Skills Reference

Use these skill commands in your AI chat after installation:

Skill	Command	Purpose
init	/crew.init	Initialize workspace + agent memory files
plan	/crew.plan	Create a change and start working
status	/crew.status	Check current progress
checkpoint	/crew.checkpoint	Phase audit + consistency check + memory sync
release	/crew.release	Archive changes + consolidate memory
agents	/crew.agents	List available specialists

Natural language works too — "run a checkpoint" triggers the checkpoint skill

---

Three Work Modes

Mode	Flow	Best For
Standard	Plan → Design → Execute → Verify → Iterate	New features, refactoring
Express	Plan → Execute → Verify	Bug fixes
Prototype	Plan → Design → Execute	Quick prototyping

---

What You Need to Do

During the workflow, the AI only needs your input at two points:

When	What you do	Why
Plan confirmation	Review AI's requirements summary, say "confirmed" or suggest edits	Ensure correct understanding
Verify confirmation	Review test results, say "confirmed" or point out issues	Ensure quality

Prototype mode has no Verify step, so you only confirm once. The rest of the time, AI progresses automatically.

---

Built-in Team

PjM assembles the team on demand based on your needs. Common roles:

Agent	Responsibility
PjM Project Manager	Task decomposition, agent coordination, progress tracking
PdM Product Manager	Requirements analysis, PRD import, acceptance criteria
Architect	Tech decisions, task decomposition, dependency analysis
Implementer	Code generation, refactoring, dependency management
Tester	Test execution, acceptance checks, coverage
Reviewer	Code review, security scanning, best practices

Team size is dynamic — PjM creates additional agents (DBA, tech writer, ops) as needed. No manual assignment required.

Agent Memory

Each agent maintains a long-term memory file in dev-crew/memory/, accumulating project knowledge, patterns, and preferences across changes. Memory is consolidated on release and auto-loaded on new sessions — the team gets better the more you use it.

---

File Structure

The AI creates the following files in your project (all under dev-crew/, excluded by .gitignore by default):

project-root/
├── dev-crew.yaml                    ← Project config (committed)
└── dev-crew/                        ← AI workspace (not committed)
    ├── resume.md                     ← Project state snapshot
    ├── blockers.md                   ← Issues and decisions
    ├── changes/
    │   └── {change-name}/
    │       ├── proposal.md           ← Requirements (Plan output)
    │       └── design.md             ← Technical design (Design output)
    └── archive/                      ← Archived changes

You don't usually need to edit these files directly, but you can view them anytime.

---

Configuration

Project root dev-crew.yaml:

project:
  name: my-project

verify:
  test_command: "npm test"        # If set, uses test command for verification; otherwise AI reviews

Most defaults are fine. The only recommended config is verify.test_command.

Domain Specialists (Optional)

If your project involves specific domains, activate built-in specialists — the AI will automatically dispatch experts at relevant PDEVI phases:

# dev-crew.yaml
specialists:
  - game-designer
  - security-engineer

• Not configured = team works independently (zero impact)
• Configured = experts auto-participate at relevant phases (no manual dispatch)

Full directory and recommendations at Specialists.