Specrouter

Configure your AI team.
Ship the workflow.

SpecRouter generates the agent prompts, project conventions, and bootstrap skill your AI coding agents need to ship spec-driven code. Four agents plus one entry-point skill — ready to install in OpenCode or Claude Code.

Free during beta. No spam, ever.

specrouter.config.json
Your tool
toolOpenCode
agentsPO · SDD · Coder · QA
stackLaravel 11
testingPest
entry-point@bootstrap
Your agents
bootstrapsetup.md
product-owner.md
sdd.md
coder.md
tests.md
qa.md
6 files ready · 4.9 KB

AI agents are powerful. Configuring them is a mess.

Hand-crafted prompts rot. Stack-specific conventions get lost. Memory doesn't persist. You spend weekends rebuilding what you already built in another project.

You're copy-pasting markdown between projects

Every new repo starts with the same 800 lines of AGENTS.md / CLAUDE.md. Manually. Every time.

Your agents drift between sessions

PO forgets the project context. SDD reinvents scope. QA reviews against yesterday's spec. Chaos compounds.

Multi-agent setups are tribal knowledge

One teammate knows how to wire Claude Code + OpenCode + a memory MCP. Everyone else copies their config and hopes.

From blank repo to working AI team in 5 minutes.

STEP 01

Pick your tool

Choose OpenCode or Claude Code. SpecRouter will generate the right files in the right paths for that tool. Stack-specific conventions come later, from your project.

STEP 02

Generate the agents

SpecRouter generates the four agents (PO, SDD, Coder, QA) and the bootstrap skill. You can describe what each agent should focus on, and the LLM customizes the system prompts to your project — your key is not stored or logged.

STEP 03

Install in seconds

Download a single ZIP, drop it into your project, fill in your API key, run the bootstrap. Your agents are working before your coffee gets cold.

Works with your AI coding tool.

SpecRouter generates the right files in the right paths for your tool. Pick yours in the wizard — more tools coming next.

OpenCode
multi-mode
Supported

Open-source AI coding agent. Supports Claude, GPT, Gemini. Agents are markdown files in .opencode/agents/, skills in .opencode/skills/.

Agents.opencode/agents
Skills.opencode/skills
Invoke@bootstrap
Claude Code
latest
Supported

Anthropic's official CLI. Agents live in .claude/agents/, skills in .claude/skills/. Invoked via slash commands or skill loader.

Agents.claude/agents
Skills.claude/skills
Invoke@bootstrap

Don't see your tool? See the roadmap.

Four agents, one entry-point.
One flow.

PO → SDD → Coder → QA is the opinionated loop. The bootstrap skill initializes your project and hands off to the Product Owner. Every step is a human approval gate.

Skill
skill: bootstrap
Bootstrap

An entry-point skill that lives in your project. Initializes docs/project-context.md by asking you a few questions, then hands off to the Product Owner. Works in OpenCode and Claude Code.

> @bootstrap
{
  step: "init",
  asks: ["project_name", "domain", "stack", "constraints"],
  next: "product-owner"
}
Agent
agents/product-owner.md
Product Owner

Understands what to build. Calls project context for the stack and recent decisions, drafts the user story with testable acceptance criteria, hands off a clean spec.

> @product-owner
> "Given this feature request,
   write a HU with 3-5
   acceptance criteria."
Agent
agents/sdd.md
SDD Planner

Knows your codebase. Breaks the approved spec into ordered, testable tasks. Pulls conventions from the existing code so the task list matches your project.

> @sdd
> "Spec received. Reading
   existing services…
   Generating 6 tasks."
Agent
agents/coder.md
Coder

Implements the task list. Writes the code, runs tests, measures coverage. Reads project context for current conventions before changing anything.

> @coder
> "Task 3/6. Writing
   UserRepository.
   Following your service pattern."
Agent
agents/qa.md
QA Reviewer

Cross-checks the user story against the shipped code. Pulls the original spec and verifies every acceptance criterion. Closes the flow.

> @qa
> "Acceptance criteria 3/5 met.
   Edge case: empty cart state…"
>
Entry-point skill (lives in your project)
Static agent prompts (.md)
Human approves between every step

See what your agents are doing.

Live activity from every agent in every project. Catch drift early, spot bottlenecks, and prove the workflow is actually shipping faster.

specrouter.dev/monitor
LIVE
7d
24
Cycles this week+18% vs last week
7d
2
Drift eventsBoth resolved
7d
1.4M
Tokens (week)Across 4 projects
Cycle activity by agent
last 7 days
PO
24 cycles
SDD
24 cycles
Coder
24 cycles
Tests
24 cycles
QA
24 cycles
Avg cycle time18 min
HU → merged PR2.4 hours
Recent activity
auto-refresh
PR #142 merged · 4 cycles · 18 min avg
2 min ago
Drift detected in ~/projects/billing · resolved
1 hour ago
Cycle completed · HU-218: Add invoice export
3 hours ago
12 cycles completed · 1.1M tokens
yesterday
Drift detection

The MCP compares your local files against the SaaS config every time the IDE starts. If a teammate's local edits diverge, you see it before the next cycle breaks.

Cycle metrics

How many full PO → SDD → Coder → QA cycles you completed this week, this month, this quarter. By project, by tool, by team.

Decision log

Every architectural decision the Bootstrap MCP captures (or that your team adds manually) is timestamped and searchable. No more 'why did we do this?' in Slack threads.

Every HU tells its story. Files, diffs, decisions.

The Bootstrap skill captures exactly which files were touched, what was added and deleted, who decided what, and how long each phase took. You never lose the context of how a feature was built.

HU-218Add invoice export
18 min ago·PR #142 merged·18 min · 4 agents
View full diff →
Spec

As a user I want to export my invoices to CSV, filtered by date range. Acceptance criteria: date filter, CSV format, downloadable file.

Files touched
4
Mapp/Models/Invoice.php
Aapp/Exports/InvoiceExport.php
Mtests/Feature/InvoiceTest.php
Aroutes/web.php
Diff summary
+47-12lines
Timeline
PO
2m
SDD
3m
Coder
11m
QA
1m
QA
1m
Decision

"Used Action class for export logic instead of inline controller — easier to test and consistent with existing InvoicePdfExport pattern."

HU-219Fix cart total rounding
2 hours ago·PR #141 merged·10 min · 4 agents
View full diff →
Spec

Cart total shows $19.97 instead of $19.98 due to floating point. Add tests for edge cases (0.01, 0.005 rounding).

Files touched
2
Mapp/Services/CartCalculator.php
Mtests/Unit/CartCalculatorTest.php
Diff summary
+18-8lines
Timeline
PO
1m
SDD
1m
Coder
6m
QA
1m
QA
1m
Decision

"Switched from float to integer cents internally. Display formatted via Laravel's Number helper."

Every HU keeps its original spec, touched files, complete diffs, per-phase timeline, and team decisions. Retention depends on the tier.

Free to start. Pays for itself when it scales.

The wizard stays free forever. Sync and team features live behind a paywall so we can keep building.

Free
$0forever

Generate flows and install them. Manual sync.

Join waitlist
  • Unlimited flow generations
  • OpenCode + Claude Code
  • All 5 agent roles + Bootstrap MCP
  • ZIP download
  • 7-day cycle history (metadata only)
  • Single user, single project
Most popular
Pro
$12per dev / month

MCP sync + full HU history across all your projects.

Get notified
  • Everything in Free
  • MCP sync across all projects
  • Dashboard editor
  • Flow version history
  • 90-day cycle history with full diffs
  • Files touched, decisions & timeline per HU
  • Drift detection & alerts
  • Up to 10 active projects
Team
$50per team / month

Shared flows, full audit, unlimited HU history.

Talk to us
  • Everything in Pro
  • Shared flow library
  • Roles & permissions
  • Unlimited HU history with diffs
  • Decision log & full audit trail
  • GitHub PR sync
  • SSO (Google, GitHub)
  • Priority support

Common questions.

What's the bootstrap skill? Is it just a .md file?+

Yes — it's a .md file (SKILL.md) that runs when you invoke @bootstrap in your AI tool. It reads your project (composer.json, package.json, go.mod, etc.), asks you 3-5 questions to fill in docs/project-context.md, and hands off to the Product Owner. No server, no daemon.

Why tool-centric instead of stack-centric?+

The agent file format changes with the AI tool you use. OpenCode expects .opencode/agents/, Claude Code expects .claude/agents/. We generate the right format for the tool you pick. Your stack (Laravel, Next.js, Go) is detected by the bootstrap reading your project — you don't pick it upfront.

Do I need to configure my stack separately?+

No. The bootstrap skill reads your project on first run and extracts the tech stack, packages, and conventions automatically. Solo devs with existing projects can be up and running in one command.

Is this just for solo devs?+

It's actually built with solo devs in mind. Once the bootstrap knows your project, the rest of the team (PO, SDD, Coder, QA) runs in the background while you focus on shipping. Teams with shared projects can share the generated ZIP as a baseline.

What about Cursor, Codex, Kiro?+

On the roadmap for Phase 2. Each uses a different agent format (MDC, single AGENTS.md, steering files). Today we support OpenCode and Claude Code because both use plain markdown files.

Is the source open?+

The generated configs (agents, skills, opencode.json) are yours to use, modify, and ship. SpecRouter itself is a closed product — the wizard is the value we provide.

Your AI team is five minutes away.

Join the waitlist. We're onboarding new devs every week during the private beta.