Closed Circuit Consultants · Las Vegas, NV · remote US
Agentic engineering for your dev team, with guardrails you can audit.
Your engineers already use Claude Code, Codex, or Cursor. We install the rules around them: what agents may do alone, what needs a human, and the tests that decide whether agent output ships. Fixed scope. Your repos, your CI.
Free 30-minute call. Written findings in 48 hours that you keep. Fixed-scope install.
How it starts
Call → 48h findings → fixed-scope install
- 1. Free 30-minute call. Team size, repos, which agents people already run, and what scares you about them.
- 2. 48-hour findings you keep. Where agents can reach secrets or production today, what your CI would let through, and the install we would do first.
- 3. Fixed-scope install. Policy, gates, harness, MCP, and training in your repos. Fixed fee. A retainer from $2,500/month only if you want us to stay.
What we install
Agent autonomy policy
Committed to the repo. Agents may edit source and tests, run build, test, and type-check, branch, commit, and open pull requests. Merge to main, force push, CI and secrets changes, and database operations need a human.
Permission gates on tool calls
Hooks and allow and deny lists in the agent configuration, so the policy is enforced and not just written down. Risky commands stop and ask.
CLAUDE.md, AGENTS.md, and MCP
Repo instructions agents read every session, and MCP servers that connect agents to your real systems instead of pasted context.
Verification harness
Three layers. Agents write and run tests against stated success criteria until green. A human reviews every diff before main. Behavioral checks exercise the running app, not the source.
Prompt-injection guard
Repository content, issues, and fetched pages are treated as data, not instructions. Agent-authored pull requests are tagged for attribution and audit.
Supply-chain gate
A pre-commit scan for leaked secrets, malicious dependency behavior, and obfuscated code before anything reaches a remote. See supply-chain compromise check.
Model routing and cost
Which jobs go to Claude, Codex, or a self-hosted open-weight model. Per-token, per-seat, and self-hosted costs compared at your team size, because most agent pilots stall on unit cost, not capability.
Team training
Pairing sessions on your codebase, then a written playbook. Adoption should not depend on one engineer's tool fluency.
Proof (client names withheld)
We run this ourselves
1,700+ commits
A consumer social platform (Vue 3 and TypeScript on Laravel 12 and PostgreSQL) built this way since December 2025, by a five-person team plus distinct agent identities in the history.
190+ spec files
Vitest and Playwright specs in the primary codebase, under the same three-layer harness we install for you.
Self-hosted models
A self-hosted GPU inference cluster runs open-weight models for bulk and low-stakes work, next to Claude Code, Codex, Cursor, and OpenCode for the rest.
8+ active codebases
The same spec-first delegation, agent implementation, and automated verification across client and product work every day.
Questions
Is this vibe coding for teams?
No. Vibe coding trusts the author. This setup trusts the harness: tests written against stated success criteria, a human review on every diff, and checks against the running app. Code nobody typed ships only when those pass.
Do agents get to merge to main?
Not under the policy we install. Agents may edit source and tests, run builds and tests, branch, commit, and open pull requests. Merging to main, force pushes, CI and secrets changes, and database operations need a human.
Which tools do you set up?
Claude Code, Codex, Cursor, and OpenCode, plus MCP servers that connect agents to your systems of record. Where it pays, a self-hosted open-weight model handles bulk and low-stakes work. We do not resell licenses.
What does it cost?
A fixed fee for a fixed scope, set from the 48-hour findings. If you want us to stay after the install, a retainer starts at $2,500/month.
Can this pass an audit?
The autonomy policy, the gates, and the attribution of agent-authored pull requests are written down in your repo so you can show them to an auditor or a customer security review. We have worked under NIST 800-53 and HIPAA controls; we do not certify compliance.
Let the agents type. Keep the judgment.
Book the free call. Bring your repo list and the agents your team already runs.
Closed Circuit Consultants · SOC LLC d/b/a · Las Vegas, NV · remote across the US