Handbook

Operating Model

No Hermes required.

Updated

Three planes

1. Governance Plane
   ForgeSDLC, Blueprints, Versonas, AGENTS.md

2. Control Plane
   Lenses, ForgeRun, approvals, EvidencePackets

3. Workcell Plane
   Hermes, Factory Droid, OpenClaw, Fleet, LCDL, future agents/tools

Core loop

Forge Platform core loop

How human intent becomes governed execution with evidence and reviewable outcomes across three planes.

  1. Human intentThe operator states what they want to achieve or change.
  2. Lenses creates/opens ForgeRunLenses opens or creates a ForgeRun to hold the work state.
  3. Blueprints produce context and evidence expectationsBlueprints supply methodology context and what evidence must be collected.
  4. LCDL performs governed reasoning or verificationLCDL runs schema-aware tasks for reasoning or verification steps.
  5. Lenses requests human approval where neededLenses surfaces approval gates before risky or irreversible actions proceed.
  6. Fleet executes approved template jobsFleet runs containerized template jobs after approval.
  7. Agent workcells perform bounded workHermes, Factory, OpenClaw, or other workcells execute within their envelope.
  8. EvidencePacket links all outputsOutputs are linked into a traceable EvidencePacket.
  9. Lenses shows final reviewLenses presents consolidated state for human final review.

Operating modes

Agent-light mode

No Hermes required.

Agent-light operating flow

A Hermes-free path from workspace visibility through governed reasoning, approval, execution, and evidence.

  1. LensesLenses starts and guides the run without adaptive agent orchestration.
  2. BlueprintsBlueprints provide methodology context and evidence expectations.
  3. LCDLLCDL performs governed reasoning or verification on the task.
  4. approvalHuman approval gates risky steps before execution proceeds.
  5. FleetFleet executes approved template jobs after the gate clears.
  6. EvidencePacketEvidence links all outputs into a traceable packet.
  7. LensesLenses returns for final review of outcomes and evidence.

Agent-operator mode

Hermes adds adaptive orchestration.

Agent-operator operating flow

Hermes adds adaptive orchestration while the same governed approval and execution spine remains.

  1. HermesHermes provides adaptive orchestration for the run.
  2. Lenses stateHermes reads and updates Lenses workspace state.
  3. Blueprints contextBlueprints context shapes what Hermes and LCDL should consider.
  4. LCDLLCDL performs governed reasoning within the orchestrated flow.
  5. approvalHuman approval still gates execution where policy requires it.
  6. FleetFleet runs approved jobs after orchestration and approval complete.
  7. EvidencePacketEvidence from orchestration and execution links into one packet.

Implementation workcell mode

Factory Droid produces review or patch output inside a controlled envelope.

Factory workcell implementation flow

How Factory Droid produces reviewable patch output inside a bounded workcell envelope.

  1. LensesLenses initiates the workcell request from the control plane.
  2. WorkcellRequestA bounded request defines what the workcell may do.
  3. Fleet runs FactoryFleet launches the Factory Droid workcell job.
  4. WorkcellResultFactory returns review or patch output within the envelope.
  5. LCDL verifiesLCDL verifies the workcell output against schemas and expectations.
  6. Lenses reviewsLenses presents verified output for human final review.