Claude Code (stable) · Codex, Gemini CLI, OpenCode (experimental)

Govern the agents you already ship

getmyorgis a verification harness for AI agent engineering — structure, standards, and runtime gates that make coding agents safe to run with less supervision. The durable leverage isn't more agents; it's the harness that lets you raise their authority level safely. Installs into Claude Code in five minutes.

5-minute setup
Runtime authority gates
70+ specialist roles
claude-code — ~/my-saas-app
$/myorg
Shown in Claude Code (stable) · experimental adapters: Codex, Gemini CLI, OpenCode
Stack-aware config
Encrypted delivery

Teams using AI agents face two problems at once.

A coordination problem — one assistant can't hold an entire organization. And a governance problem — agents will happily ship band-aids, suppress errors, and force-push to main if you let them.

One assistant can't hold the structure of a real organization

Context, standards, and judgment of an entire org don't fit in one prompt window. Agents drift, duplicate work, and miss craft conventions.

Agents will ship band-aids if you let them

Agents suppress errors to make tests pass, disable failing tests, add fallbacks that hide real bugs, and bypass pre-commit hooks. Review density is too low to catch all of it.

Human-org defaults break in agent orgs

The cost ratios that justify shortcuts in human orgs invert when agents iterate at speed. Band-aids compound silently. Memory of 'we patched this' evaporates between sessions.

No runtime check on agent claims

When an agent says 'I fixed it', there's no verification that the claim matches reality. Audit trails tell you what happened — they don't gate what's allowed.

Up and running in under 2 minutes

No complex setup. No manual configuration. Just install into Claude Code, authenticate, and start working with specialists.

Supported hosts: Claude Code stable · Codex, Gemini CLI, OpenCode experimental
01

Install for your host

Add getmyorg to Claude Code (stable). Experimental adapters exist for Codex, Gemini CLI, and OpenCode.

Claude Code:
$ claude plugin add getmyorg
Codex CLI (experimental):
$ codex mcp add getmyorg
02

Authenticate once

Connect your account via OAuth. Your org config loads automatically on each session.

$ getmyorg auth
03

Run /myorg

Scans your repo, detects your stack, and loads the right specialists and standards.

$ /myorg
04

Work with specialists

Route tasks to the right agent. Get structured handoffs, guardrails, and consistent output.

$ @tech-lead review auth flow

Structure. Verification. Authority gates.

Three pillars that solve both the coordination problem and the governance problem. Pre-built and opinionated — installs in minutes.

70+ specialists

Structure

Pre-built agents, teams, functions, and craft standards. Opinionated and cross-host — one definition projects into Claude Code (stable) and the experimental Codex, Gemini CLI, and OpenCode adapters.

Evidence-cited

Verification

Every edit re-checked by the band-aid engine. Every claim cites a run id from the verification plane. Recipes you can run on demand — evidence persisted under .myorg/evidence/.

L0–L3

Authority gates

Graduated L0–L3 autonomy per action type. Today: git push --force to protected branches is gated end-to-end on fresh verification evidence. L3 blocks; L2 confirms; every decision audits.

Auto-detect

Stack-aware setup

Detects your repo's tech stack and maps the right agents, teams, and standards automatically. Updated on every commit, branch switch, and merge.

Full lifecycle

Workflow playbooks

Narrative workflow guides applied with judgment by leads and the routing agent. Feature build, incident response, triage — full lifecycle, not just prompts.

Persistent

Project map

A three-plane graph (System, Execution, Intent) that maps your repo's tech stack and modules to blueprint agents, teams, and playbooks. Shared context for every specialist.

Two demoable proof points.

The harness claim is concrete. These are the two surfaces that ship today — what they gate, how they audit, and what they refuse to claim. Full capability matrix →

PostToolUse output-verification probe

Every edit re-checked. Every claim cites evidence.

validate-postedit — after Edit(auth.ts)
PostToolUse: re-scanning applied diff...
band-aid scan: no bare excepts, no lint disables
no @ts-ignore, no skipped tests
contract judge queued (MYORG_VERIFY_LLM_JUDGE=1)
evidence: .myorg/evidence/ev-a3f8c12d/
run-id: ev-a3f8c12d (cite before claiming complete)

After every Edit/Write, the band-aid engine re-runs synchronously on the applied diff. An optional async contract LLM judge checks the change against intent. The agent's claim (“I fixed it”) is checked against evidence — not just declared.

Advisory-only at PostToolUse — edit already applied. Pre-tool band-aid detector can block pre-apply.

Force-push blast-radius gate

End-to-end gated on fresh verification evidence.

validate-bash — git push --force origin main
PreToolUse: git push --force detected
consulting: getmyorg policy resolve git.push.force_to_protected
preset: Sovereign → checking evidence freshness...
no fresh verification evidence found
resolved: L3 — escalate-and-halt
audit.policy.resolved emitted — observable via: getmyorg audit tail
run `getmyorg verify` first to produce evidence

git push --force to a protected branch is gated end-to-end on fresh verification evidence. No fresh evidence → L3 deny. Fresh evidence → L2 confirmation prompt. Every resolution emits an auditable event.

One destructive action class gated end-to-end today. General deploy and migration gating is explicitly not built — it's tracked in the open in #379.
not a marketplacenot LangChain — agents already existnot observability — it tells you what agents did; we gate what they can dohonest about what's not yet built

See how it works inside your agent

The core surfaces you'll touch daily. Shown in Claude Code (stable); the same flow runs in the experimental Codex, Gemini CLI, and OpenCode adapters.

claude-code — ~/my-saas-app
$/myorg
Scanning repository structure...
Framework: Next.js 16 (App Router)
Language: TypeScript 5.4
Styling: Tailwind CSS 4.0
Database: Supabase (PostgreSQL)
Auth: Supabase Auth
Payments: Stripe
Loading org preset: Sovereign
12 specialists activated
Project HQ initialized
Guardrails enabled (pre-commit, security)
Ready. Use @specialist-name to route tasks.

Initialize in seconds

One command scans your repo, detects your stack, and loads the right configuration.

Cross-host ensemble

Use the plans you already pay for

If you pay for Claude, Codex, and Gemini separately, /myorg:ensemble runs all three against the same question in parallel — then synthesises one answer. You keep the verdict; it surfaces perspectives, not consensus.

How it works
  1. 01Fans your prompt to every installed peer in parallel — each gets the same brief, no session history.
  2. 02Waits for all peers (per-peer timeout). Timed-out peers appear in the merge as “(no response)” — no hang.
  3. 03The calling agent synthesises a final answer. You decide what’s right — ensemble surfaces perspectives, not verdicts.
$/myorg:ensemble “Review the auth flow in server/internal/auth/”

Good for

  • Factual questions where agreement signals confidence
  • Architecture and design reviews
  • Code reviews on non-trivial diffs
  • Writing specs — three drafts, one synthesis
  • Security threat models

Honest tradeoffs

Adds latency

You wait for the slowest peer. Default timeout: 120s per peer.

Three calls, three bills

Each peer is billed under its own subscription. Ensemble doesn't save money — it uses the capacity you already pay for.

No shared state

Peers run in parallel sandboxes. One peer's file writes aren't visible to the others. Build self-contained briefs.

Tier note

/myorg:ensemble requires the Lean tier or above. The per-host delegation commands (/myorg:codex, /myorg:gemini) are available on all tiers.

The passive org layer — specialists, project map, standards — works on every tier regardless of which hosts you have installed.

Harness configurations, not just org sizes.

Each preset is a default mapping from action class to autonomy level. Pick the harness posture that fits today — per-action overrides compose on top.

Start here

Sovereign

Assisted — Free

High supervision. Most gated actions default to L2/L3. Solo operator, sequential work, ad-hoc agent help.

All 70+ types available
  • On-demand routing
  • L2/L3 default gates
  • Personal HQ

Lean

Structured

Defined boundaries. Routine actions L0/L1; sensitive actions L2. Small team with standard team patterns.

All 70+ types available
  • Team routing
  • L0/L1 routine gates
  • Workflow playbooks

Focused

Directed

Workflow-centric. Agents handle substantial execution within defined workflows; humans steer at key gates.

All 70+ types available
  • Domain specialists
  • Active blast-radius gates
  • Review chains

Scaled

Autonomous

Low supervision. Most actions L0/L1 within enforced policy boundaries. Full org with deputies and concurrent teams.

All 70+ types available
  • Deputy system
  • Parallel execution
  • Full autonomy policy

All presets share the same verification harness and 70+ specialist types. The preset sets your default autonomy posture — adjust per-action anytime via org or repo instructions.

Security-first architecture

Your IP stays protected

getmyorg is built for teams with proprietary codebases and sensitive workflows. Every layer is designed to keep your intellectual property secure.

Encrypted Asset Bundles

All org configuration, specialist definitions, and standards are encrypted end-to-end before delivery.

No Plaintext on Disk

Your proprietary files, custom agents, and org secrets never exist in plaintext on your local machine.

MCP / OAuth Flow

Secure authentication via OAuth. Assets delivered through the Model Context Protocol. No API keys in config files.

Encrypted Local Cache

Session data and context are cached locally in encrypted form. Cleared automatically or on demand.

Security StatusAll systems secure
Asset Encryption
AES-256-GCM
Active
OAuth Session
Token valid
Active
Local Cache
Encrypted
Active
No plaintext secrets • Encrypted at rest, decrypted locally • Hosted on Cloudflare

Get started in five minutes.
Govern agents in six.

Install the harness, run /myorg:init, and your first org preset is live — with verification, blast-radius gates, and specialist routing configured for your stack.

Free to start. No credit card required. Honest about what's not yet built.

FAQ

Questions, answered plainly

No hand-waving. If something isn't built yet, the answer says so.

Read the source on GitHub