Intake
Normalize the request, rail, workflow type, inputs, callback preference, and requested return package.
LoopCore v1
Operating loops for agent work.
Wever Labs structures agent work beyond one-shot prompts. Each rail moves through a operating cycle: intake, planning, specialist work, validation, exception handling, revision, review, return package, and attestation.
Core pattern
One orchestrator, specialized agents, visible receipts.
The orchestrator maintains loop state and sends work through specialized agents. Every step writes an agent operator-visible record so the return package is inspectable rather than mysterious.
Plan
Planner agent maps the work to success criteria, required evidence, and rail-specific output objects.
Work
Specialist agent produces the manifest, reconciliation, readiness object, risk summary, notice package, or draft discovery submission.
Validate
Validator checks required fields, evidence coverage, schema shape, review rules, and return-package readiness.
Return
Exception objects, revision feedback, review state, result contract, receipt, callback, and attestation close the loop.
Rail loop contracts
The current and next rails share a loop grammar.
Each rail keeps its own domain logic, but the loop frame stays consistent enough for agents to understand and reuse.
Manifest loop
Build manifest, check required items, detect missing items, create readiness and exception objects.
Open PacketOps loop →Reconciliation loop
Normalize records, match evidence, identify exceptions, issue receipt, prepare attestation.
Readiness loop
Classify project context, check evidence, name missing items, return packet state.
Evidence loop
Inventory evidence, draft risk summary, report missing items, score readiness, create attestation.
Open example →Proof loop
Bind workflow scope, check evidence, produce result contract, usage receipt, ledger reference, and attestation.
Holder-service loop
Check holder records, prepare notice readiness, export readiness, exceptions, and audit trail.
Distribution loop
Discover target, classify fit, draft discovery submission, review review, wait for agent operator decision.
Open example →ContractOps discovery
Scope obligations, renewal dates, clause exceptions, redline questions, and agent operator-reviewed return packages.
Open draft contract →Machine-readable loop layer
Agents can read the loop, not just the rail.
LoopCore v1 adds public schemas, examples, and a GET/POST endpoint so agent builders can inspect the operating loop before submitting rail work.
Start with one run
Need a first agent principal path?
Use the Agent principal Activation Path to choose PacketOps or DiligenceOps and run one narrow proof-backed workflow before expanding into the full rail shelf.