Public beta

Give coding agents explicit logic boundaries.

Install a reviewable instruction contract for requirement shaping, PCD authoring, supported operations, and evidence handoff.

Agent instruction contract

Shape inputs, outputs, and failure behavior before writing code.

Use supported monomers instead of inventing hidden operations.

Report commands, changed files, assumptions, and missing evidence.

Managed-file writes require explicit reviewable consent.

Instructions an agent can follow and a reviewer can inspect.

Skills shape how the agent works inside a repository. They do not replace the CLI, MCP, or the human review decision.

Declare inputs, outputs, assumptions, and fail-closed behavior before implementation.

Select supported monomers and keep branches and returns explicit.

Return commands, changed files, evidence state, assumptions, and remaining blockers.

agent skill install

$Open the BRIK64 agent installation guide

$Review the managed instruction section

$Approve the repository change

The instruction contract is visible before the agent uses it.

The two surfaces are complementary and intentionally separate.

Repository-facing instructions for how an agent should shape, author, change, and report bounded logic work.

A structured resource and tool surface for discovering current BRIK64 contracts and documentation.

Managed-file changes remain explicit and reviewable; the skill does not silently authorize repository writes.

Start with the public skill, then connect MCP when the agent needs structured BRIK64 context.