agent-blueprint

A Claude Code plugin providing multi-agent orchestration for software projects — planning, code review, security audits, and structured execution workflows.

Installation

# Install to your project
claude plugin install agent-blueprint

# Or load during development
claude --plugin-dir /path/to/agent-blueprint

Agents

Agent	Role	Model	Description
planner	Strategic Planner	opus	Analyzes requirements, researches the codebase, produces detailed implementation plans in .blueprint/plans/. Never writes code.
researcher	External Researcher	sonnet	Searches documentation, APIs, and web sources. Read-only. Cites all sources.
reviewer	Code Reviewer	opus	Reviews completed work and plans. Approval-biased — rejects only for true blockers. Read-only.
auditor	Security Auditor	opus	Audits for security vulnerabilities and spec compliance (OAuth, JWT, CORS, etc.). Skeptical-biased. Read-only.

Skills

Skill	Invocation	Description
plan	/agent-blueprint:plan	Orchestrates the full Plan-Review-Execute workflow: delegates to Planner, runs Reviewer/Auditor review, then hands off to start-work.
start-work	/agent-blueprint:start-work	Executes a plan from .blueprint/plans/ by working through tasks sequentially, tracking progress via checkboxes.
review	/agent-blueprint:review	Runs Reviewer and Auditor agents against recent changes for code quality and security review.

Workflow

The intended workflow follows a Plan-Review-Execute cycle:

1. Plan: Run /agent-blueprint:plan to kick off the full orchestrated workflow, or invoke the planner agent directly for just the planning step.
2. Review: The plan skill automatically runs Reviewer/Auditor review. You can also run /agent-blueprint:review independently at any time.
3. Execute: Run /agent-blueprint:start-work to work through the plan task-by-task.
4. Post-Review: After execution, run /agent-blueprint:review again to audit the completed work.

Structure

.claude-plugin/plugin.json   # Plugin manifest
agents/                      # Sub-agent definitions (.md files)
  planner.md                 # Strategic planner
  researcher.md              # External researcher
  reviewer.md                # Code reviewer
  auditor.md                 # Security auditor
skills/                      # Skill definitions (directories with SKILL.md)
  plan/SKILL.md              # Full Plan-Review-Execute orchestration
  start-work/SKILL.md        # Plan execution workflow
  review/SKILL.md            # Review orchestration
scripts/                     # TypeScript hook/utility scripts

Adding Custom Agents

Create a new .md file in agents/ and register it in .claude-plugin/plugin.json:

---
name: my-agent
description: When Claude should use this agent.
model: sonnet
maxTurns: 20
---

System prompt for the agent goes here.

Then add the path to the agents array in plugin.json:

"agents": [
  "./agents/my-agent.md"
]

Note: Each agent file must be listed explicitly — directory paths are not supported.

Adding Custom Skills

Create a new directory in skills/ with a SKILL.md and register it in .claude-plugin/plugin.json:

---
name: my-skill
description: When and why Claude should use this skill.
---

Instructions for the skill go here.

Then add the skill directory to the skills array in plugin.json:

"skills": [
  "./skills/my-skill/"
]

Note: Each skill directory must be listed explicitly — parent directory paths are not supported.

Skills are invoked with /agent-blueprint:<skill-name>.