OpenClaw · Hermes Agent · OpenCode

Coding runtimes become receipt-emitting execution engines.

AgentFoundry does not replace the strongest runtimes. It wraps them with WorkGraph leases, ToolMesh permissions, policy checks, trace capture, proof packets, rollback notes, and human approvals.

runtime-fabric.yaml

openclaw

plan · execute · verify · review · receipt

ready

hermes-agent

plan · execute · summarize · emit_receipt

ready

opencode

execute · verify · summarize · receipt

ready

claude-code

plan · execute · review · summarize

ready

codex

plan · execute · verify · summarize

ready

Runtime ConsoleAPI mirror · /v1/runtime/fabric

Worker routing view for real runtimes, not mock agents.

OpenClawprimary

workspace mounted under branch/path lease

plan/execute/verify/review/receiptready

Hermes Agentorchestrator

skill + cron + handoff receipts normalized

plan/execute/summarize/receiptready

OpenCodeverifier

terminal-native execution isolated behind sandbox policy

execute/verify/receiptleased

Codexparallel worker

bounded implementation candidate with conflict arbitration

plan/execute/verify/receiptgated

Claude Codereview worker

deep review + risk register behind proof policy

plan/review/receiptreview

Gemini CLIalternate

swappable reasoning/execution worker

plan/execute/verifystandby

openclawadapter

OpenClaw

primary autonomous engineering worker

mode: plan · execute · verify · review · receipt
source: lexopedia-stack/vendor/openclaw
governance wrap: leases + policy-scoped tools + proof packet receipts

hermes-agentadapter

Hermes Agent

orchestrator, skills, scheduled work, tool governance

mode: plan · execute · summarize · emit_receipt
source: lexopedia-stack/vendor/hermes-agent
governance wrap: skill registry + cron/event runs + approval-aware tool calls

opencodeadapter

OpenCode

terminal-native coding and verification runner

mode: execute · verify · summarize · receipt
source: lexopedia-stack/vendor/opencode
governance wrap: isolated workspace + checks + reviewer evidence

claude-codeadapter

Claude Code

deep implementation and review worker

mode: plan · execute · review · summarize
source: lexopedia-stack/vendor/claude-code
governance wrap: branch lease + diff receipt + policy review

codexadapter

OpenAI Codex

parallel implementation/verifier worker

mode: plan · execute · verify · summarize
source: lexopedia-stack/vendor/openai-codex
governance wrap: bounded tasks + test receipts + conflict arbitration

openhandsadapter

OpenHands

computer-use and repo automation worker

mode: execute · browser · terminal · artifact
source: lexopedia-stack/vendor/openhands
governance wrap: sandbox egress + trace capture + rollback notes

ToolMesh Console/v1/tool-registry/capabilities

GitHubchecks/read by default; PR/write gated

WorkGraph lease + Proof Packet

approval_required

Shellsandboxed commands only; destructive denied

Execution receipt + side-effect summary

leased

Browserobserved navigation, screenshots, console logs

Execution Chamber artifact

leased

MCP resourcesOAuth/OIDC-sensitive resources exposed by contract

ToolMesh capability + denial receipt

review

SkillsOpenClaw/OpenAI/agent skills wrapped as governed actions

policy scope + proof ref

ready

EvalsDeepBrainz evals/guardrails feed Quality Lab

quality gate + risk note

approval_required

backend service

WorkGraph Engine

objectives, DAG nodes, edges, leases, state transitions, replay

backend service

Governance Core

OpenFGA/OPA policy, branch/path authority, approval gates, denial receipts

backend service

Runtime Fabric

OpenClaw, Hermes Agent, OpenCode, Codex, Claude Code, OpenHands, Gemini adapters

backend service

ToolMesh

MCP tools, skills, browser actions, terminal actions, OAuth capabilities

backend service

Proof Packet

evidence compiler, artifact references, reviewer checklist, rollback lineage

backend service

Quality Lab

tests, evals, guardrails, risk scoring, acceptance proofs

apps/* + vendor/* leverage

Runtime Fabric is assembled from functioning engines and app contracts.

Runtime Fabric backend

lexopedia-stack/vendor/openclaw + hermes-agent + opencode + claude-code + codex + openhands

adapter contract, capabilities, leases, receipts, approval boundaries

GET /v1/runtime/fabric and GET /v1/runtime/source-wrappers

Repo/work proof machinery

lexopedia-stack/apps/gitclaw + gitpilot + gitwork + gitcode

WorkGraph, repo/path ownership, reviewer gates, proof handoff contracts

GET /v1/source-leverage and Proof Packet sample

ToolMesh + Execution Chamber + Quality Lab

mcp SDKs, E2B/browser-use, DeepBrainz evals/guardrails

capability contracts, sandbox/browser receipts, eval-backed risk gates

GET /v1/toolmesh/sources and evidence-plane readbacks

lexopedia-stack/apps/gitclaw/gitpilot/gitwork/gitcoderepo-native WorkGraph appswrap as services and proof workflows

vendor/openclaw/hermes-agent/opencodefunctioning execution runtimesadapter-wrap behind Runtime Forge

vendor/mcp/e2b/browser/evalstool, sandbox, browser, quality machinerygovern through ToolMesh, Execution Chamber, Quality Lab

backend leverage

Backend control plane

Python 3.14-ready FastAPI + Pydantic v2 + Granian, event-sourced around WorkGraph, runtime adapters, and execution receipts.

backend leverage

Governance spine

OPA policy decisions + OpenFGA authority graph for identities, repo/path ownership, delegation, and approvals.

backend leverage

Runtime fabric

OpenClaw, Hermes Agent, OpenCode, Codex, Claude Code, Gemini, and OpenHands operate as swappable engines under AgentFoundry authority.

backend leverage

Execution chamber

E2B, Daytona, Docker, Kubernetes, Playwright, Browser Use, and terminal substrates run with isolation and proof receipts.