Run control
Connecting to cockpit overview…
Product cockpit
Run governed autonomous engineering workforces.
This is the product surface: create objectives, inspect the live WorkGraph, control runtime adapters, approve handoffs, review Proof Packets, and keep every autonomous action tied to authority, receipts, rollback, and human decisions.
Live product modeAPI-connectedWorkGraph · Runtime Fabric · ToolMesh · Proof Packet
product workspace connected to live API
This is now an operations cockpit, not static docs.
The first screen reads live state from api.agentfoundry.in: health, run feed, Runtime Fabric, WorkGraph ledger, Proof Packet, governance decisions, readiness, and event replay. A failed read visibly breaks the cockpit.
connecting0/8 backend contracts onlinehttps://api.agentfoundry.in
health
/healthconnecting…cockpit
/v1/cockpit/overviewconnecting…runtime
/v1/runtime/fabricconnecting…workgraph
/v1/workgraph/ledger/sampleconnecting…proof
/v1/proof-packets/sampleconnecting…governance
/v1/governance/decisions/sampleconnecting…events
/v1/events/feedconnecting…readiness
/v1/platform/readiness/summaryconnecting…objective ledger
—events—leases—receipts
Authority state: loading
Run: waiting
Governed workers
proof packet
Decision: loading
Workflow: unreported
Approve / retry / narrow / stop
Start a governed run
Normalize an engineering objective into acceptance criteria, constraints, owner, repo scope, and approval class.
Inspect execution state
View leases, dependencies, runtime receipts, side effects, proofs, risks, rollback refs, and approval gates.
Open WorkGraphApprove or block
Turn Proof Packet evidence into an explicit deploy, retry, narrow, stop, or PR-handoff decision.
Review proofSix-role proof loop
intakeleased / receipted
plannerleased / receipted
implementerleased / receipted
verifierleased / receipted
reviewerleased / receipted
release/handoffleased / receipted
Wrapped engines
OpenClawworkspace mounted under branch/path lease
Hermes Agentskill + cron + handoff receipts normalized
OpenCodeterminal-native execution isolated behind sandbox policy
Codexbounded implementation candidate with conflict arbitration
Claude Codedeep review + risk register behind proof policy
Permissioned tools
GitHubchecks/read by default; PR/write gated
Shellsandboxed commands only; destructive denied
Browserobserved navigation, screenshots, console logs
MCP resourcesOAuth/OIDC-sensitive resources exposed by contract
SkillsOpenClaw/OpenAI/agent skills wrapped as governed actions
EvalsDeepBrainz evals/guardrails feed Quality Lab