V1
control plane for agentic work

Command
Layer

CmdLayer is infrastructure for governed AI agents. It turns prompts, roles, rules, truth files, reasoning protocols, and verification gates into systems agents can actually run inside.

control-planelive

What this is

CmdLayer is the umbrella: OCR governs execution, CRP governs reasoning, and the Go engines make those rules executable.

core claim
Prompting asks. CmdLayer governs.
plain english

Agents need operating constraints.

Prompts are not infrastructure. CmdLayer gives agents boundaries: what files are true, what skills apply, what templates to use, what phase the work is in, and what proof is required before output is trusted.

OCR

Execution governance

Loads project truth, skills, checklists, templates, and writeback rules so workers cannot drift from the system.

CRP

Reasoning governance

Runs multi-role council reasoning when one-shot answers are too shallow, biased, or premature.

Go engines

Executable runtime

Two compiled Go binaries — ocr-go and crpctl/crpd — turn the protocols into deterministic state transitions, CLI tools, and an MCP daemon.

Stack map

Both engines share one shape: load the rules, run a bounded loop, and refuse to claim success without proof. Orient here before clicking into OCR or CRP.

01 // load

Truth & skills

OCR loads YAML truth packs from .ocr/; CRP loads the request, skill, and council roster. The rules come before the work.

02 // bind

Roles & workers

OCR binds skills to bounded workers; CRP instantiates specialist roles. Neither can rewrite the truth it was given.

03 // execute

Go runtime

Compiled binaries enforce sequence, state, and validation through a finite state machine — not vibes.

04 // prove

Verified output

OCR gates every writeback; CRP records an audit ledger and synthesis. Failures return as failures.

Choose a layer

Each hub explains its framework, walks through the Go engine and how to run it, and links the downloadable binary.