BLUEPRINT SDLC

One command. Your entire software development lifecycle — automated.

Up to 200x faster. Fully automatic. Stack-agnostic. Zero paid dependencies.

macOS + Linux · Apple Silicon + Intel · zsh + bash · Requires Claude Code

---

The Problem

You know the drill. You open Claude Code, start a feature, and 45 minutes later you're debugging a context overflow because the model forgot what it was building. You've tried CLAUDE.md files, custom prompts, manual checklists — but none of it sticks. The model drifts. Tasks get skipped. Tests don't run. PRs go out half-baked.

Claude Code is the most powerful coding tool ever built. But without structure, it's a Ferrari in a parking lot.

BLUEPRINT gives it the road.

---

What BLUEPRINT Does

BLUEPRINT turns Claude Code from a code assistant into a disciplined engineering partner. It wraps the entire software development lifecycle — from idea capture to merged PR — into a structured pipeline of slash commands, enforced by a compiled Go audit hook that fires on every single tool call.

The result? Features that used to take hours of back-and-forth now ship in minutes. Not because the model is faster — because it never gets lost.

GTD taught you to capture everything. BLUEPRINT ships it.

The Speed Difference

Without BLUEPRINT	With BLUEPRINT
Manually describe the task every session	/plan captures everything once — persists across sessions
Model forgets context mid-feature	Audit hook enforces the plan on every tool call
You babysit every step	/flow-auto runs the entire pipeline autonomously
One agent, sequential work	/plan-approved spawns parallel subagents — true concurrency
"Did I run the tests?"	Audit rule #8 blocks PRs without test runs
PRs with missing context	/pr auto-generates summary from the plan file
Features take 2-4 hours of Claude time	Same features in 5-15 minutes with parallel execution

Up to 200x faster isn't marketing — it's math. When you split 20 tasks across 4 parallel subagents, each running at full speed with no context drift, the throughput is incomparable to one model doing everything sequentially while you re-explain the task every 10 minutes.

---

Install

Option 1: Claude Code Plugin (Recommended)

/plugin marketplace add skaisser/blueprint-plugin
/plugin install blueprint

2-second install. All 27 skills, audit hooks, MCP servers, and status line — active instantly.

Option 2: Homebrew (CLI only)

brew tap skaisser/tap
brew install blueprint

Installs the CLI binary only. Use with the plugin for full functionality.

Option 3: Manual (Legacy)

curl -fsSL https://raw.githubusercontent.com/skaisser/blueprint/main/install.sh | bash

Full install: binary + skills + hooks + settings + MCP servers.

Migration from install.sh

If you previously installed via curl | bash, the plugin install works alongside it. Skills from the plugin take priority. To clean up the old install:

./install.sh --uninstall   # Remove old install
# Then install via plugin

What you need

• Claude Code installed and working
• curl and git (pre-installed on macOS and most Linux)

Uninstall

Clean removal — keeps your other Claude Code settings intact:

./install.sh --uninstall
# or remotely:
bash <(curl -fsSL https://raw.githubusercontent.com/skaisser/blueprint/main/install.sh) --uninstall

Verify

blueprint version

---

What Gets Installed

The installer sets up everything in one shot — no menus, no choices, no dependencies. Both curl | bash and ./install.sh install the same thing.

Component	Target	What it does
Blueprint CLI	~/.blueprint/bin/blueprint	Compiled Go binary — blueprint audit, blueprint status, blueprint update. Pre-built for macOS (arm64/amd64) and Linux (amd64). Zero runtime dependencies.
27 SDLC skills	~/.claude/skills/	The slash commands that drive the entire pipeline — from /backlog to /finish. Each skill is a self-contained prompt with its own references.
Audit hook (15 rules)	settings.json	Fires on every Claude Code tool call. Blocks dangerous commands, enforces test parallelism, requires plan-check before PR.
Status line	~/.blueprint/statusline.sh	Live display — model, context bar, estimated time remaining, git branch, code changes, session duration.
Permissions	settings.json	Pre-approves git add, git push, sequential-thinking MCP, and /commit.
Agent Teams	settings.json	Enables experimental coordinated teams — /plan-approved can spawn multiple workers.
Plugins	settings.json	ralph-loop (recurring tasks), skill-creator (build/test skills), playground (interactive HTML explorers).
MCP servers	mcp.json	Context7 (live docs for any library, free) + Sequential Thinking (structured reasoning). Both via npx, zero config.
Git hooks	~/.blueprint/templates/	commit-msg enforces emoji+type format. pre-push blocks pushes to main. Copied to projects via /start.
GitHub Action	~/.blueprint/templates/	claude-pr-reviewer.yml triggers @claude code review on PRs. Copied to projects via /start.

All settings are smart-merged — the installer never overwrites your existing settings.json or mcp.json. It only adds missing sections. Existing files are backed up with timestamps before any modification.

---

The Pipeline

Every letter in BLUEPRINT maps to a pipeline phase. This isn't a naming gimmick — it's how the system actually works:

#		Command	Phase	What happens
1	B	/backlog	Backlog	Capture idea → blueprint/backlog/0001-feature.md
2	L	/plan	Layout	Create branch + blueprint file with phases, tasks, acceptance criteria
3	U	/plan-review	Unpack	Assign [H]/[S]/[O] complexity, pick execution strategy, mark model tiers
4	E	/plan-approved	Endorse	Spawn parallel subagents, execute phases, commit per phase
5	P	/plan-check	Preflight	Audit code vs blueprint — detect deleted tasks, orphaned refs, missing tests
6	R	/pr	Raise	Open PR with full context from the blueprint file
7	I	/review	Inspect	Trigger @claude code review via GitHub Action
8	N	/address-pr	Negotiate	Fetch review feedback, categorize, fix, push
9	T	/finish	Tag	Merge PR → move blueprint to upstream/

Skip the Steps — Let It Run

Don't want to drive? Let BLUEPRINT handle the entire pipeline:

/flow                 # Guided pipeline — pauses at review for your approval
/flow-auto            # Zero-touch — model decides everything, PR ready for merge
/batch-flow 2-6       # Execute blueprints 0002–0006 sequentially in one session

/flow-auto is the endgame: you describe what you want, walk away, and come back to a PR. The model plans, reviews its own plan, executes with parallel subagents, audits the result, opens the PR, and stops. You just merge.

---

The BLUE Workspace

Your file path is your status. No database, no dashboard — just folders:

	Folder	Trigger	Meaning
B	blueprint/backlog/	/backlog	Ideas not yet planned
L	blueprint/live/	/plan	Currently in development
U	blueprint/upstream/	/finish	Shipped and merged
E	blueprint/expired/	/backlog --archive	Cancelled or deferred

Files move between folders automatically on each phase transition. Open blueprint/ as an Obsidian vault and your kanban is ready — Dataview compatible out of the box.

---

Why BLUEPRINT?

vs. Raw Claude Code

Raw Claude Code	With BLUEPRINT
Great at coding, bad at process	Process is enforced — 15 audit rules on every tool call
Forgets what it's building after 50% context	Blueprint file is the persistent source of truth
You re-explain the task every session	/resume picks up exactly where it stopped
One agent, one task, sequential	Parallel subagents — 2-6 workers on independent phases
No commit discipline	Emoji+type format enforced, AI signatures blocked
"Did I test this?"	Audit rule blocks PRs without test runs

vs. Other Claude Code Workflows

Other workflows	BLUEPRINT
Loose CLAUDE.md conventions	Compiled Go audit hook — not a suggestion, an enforcement
Manual slash commands you have to remember	9-step pipeline where each step triggers the next
Single-agent execution	Multi-agent parallel execution with coordinated teams
Context overflow crashes your session	/resume recovers from any crash, context limit, or session break
No project structure	BLUE workspace — your folder structure is your kanban

vs. Not Using Claude Code at All

If you're still writing every line by hand, BLUEPRINT won't help you. But if you've adopted Claude Code and want to go from "useful assistant" to "autonomous engineering team" — this is the system.

---

Execution Strategies

/plan-review analyzes your plan and automatically picks the fastest execution mode:

Complexity	Meaning	Model
[H]	Fast — config, styling, simple tests	Sonnet (fast output)
[S]	Balanced — business logic, services	Opus (maximum quality)
[O]	Deep reasoning — architecture decisions	Opus (full context)

Strategy	When	How
Parallel Subagents (default)	2+ independent phases	Multiple Agent calls in one message — true parallelism
Coordinated Team	Workers need mid-task handoffs	Team messaging between agents
Single Subagent	1 phase or strictly sequential	One Agent call, no spawn overhead
Leader Direct	≤3 [H] tasks total	Lead model handles directly — no agent spawn

---

All 27 Skills

Category	Skills	What they do
Pipeline	/backlog /plan /plan-review /plan-approved /plan-check /pr /review /address-pr /finish	The full BLUEPRINT pipeline — idea to merged PR
Automation	/flow /flow-auto /flow-auto-wt /batch-flow	Run the pipeline automatically — guided or fully autonomous
Fast Tracks	/quick /hotfix /resume	Small fixes without planning, emergency deploys, crash recovery
Git & PR	/bp-commit /bp-ship /bp-push /bp-branch	Git operations with safety checks and emoji format
Testing	/bp-test /bp-tdd-review	Create Pest tests, restructure plans for TDD
Project Setup	/start /bp-context /bp-status /complete	Initialize projects, generate CLAUDE.md, show status
Skill Factory	/skill-creator	Build, test, benchmark, and optimize new skills

---

Audit Hook — 15 Rules

blueprint audit fires on every Claude Code tool call via PreToolUse. It's a compiled Go binary — fast, zero dependencies, impossible to skip.

#	Rule	What it catches
1	Skill read gate	Writing code without reading the relevant SKILL.md first
2	Reference tracking	Skipping key reference files before implementation
3	Team compliance	Using teams without reading team-execution.md
4	Standalone task count	3+ tasks without spawning a team
5	Handoff tracking	Missing checkpoints at /flow pauses
6	Checkpoint audit trail	Skipping /plan-check before /pr
7	Workflow creation gate	Creating CI workflows without a staging branch
8	Test suite enforcement	Running tests without --parallel or --filter
9	Plan task deletion	Agents deleting tasks instead of implementing them
10	Dangerous command block	migrate:fresh, AI signatures, direct push to main
11	Review enforcement	Short @claude review comments — require full prompts
12	Plan-check skip	Opening a PR without running /plan-check
13	Acceptance criteria	PR with unchecked acceptance criteria
14	Flow-auto enforcement	Skipping mandatory pipeline steps
15	Backlog CLI enforcement	Manual file parsing instead of blueprint backlog

---

Commit Format

Every commit follows emoji+type format, enforced by the commit-msg hook. AI signatures (Co-Authored-By, Generated by Claude) are automatically blocked.

<emoji> <type>: <description>   (present tense, lowercase)

Emoji	Type	Use case
✨	feat	New feature
🐛	fix	Bug fix
📚	docs	Documentation
♻️	refactor	Restructuring, no behavior change
🧪	test	Tests only
📋	plan	Blueprint file updates
🔀	merge	Branch merge
🩹	hotfix	Urgent production fix
🚀	deploy	Deployment / CI

---

Platforms

Binary	Platform	Arch
blueprint-darwin-arm64	macOS	Apple Silicon (M1–M4)
blueprint-darwin-amd64	macOS	Intel
blueprint-linux-amd64	Linux	x86_64

---

Roadmap

Version	Scope
v1.0 ✅	27 SDLC skills · blueprint CLI · 15-rule audit hook · Obsidian BLUE workspace
v1.1	Laravel TALL preset · Node/TS preset
v1.2	Standalone skills as optional add-ons
v1.3	Autoresearch eval dashboard + optimizer
v2.0	Improved multi-agent execution · inter-agent blueprint handoffs

---

License

Apache 2.0 — use freely, build openly. See LICENSE for details.

---

Made with ❤️ by Shirleyson Kaisser

⭐ Star this repo if BLUEPRINT makes you ship faster