Research
Analyze the bounded work order and return evidence.
Connect an approved Agno or OpenClaw runtime, queue a bounded social planning brief, and review its complete artifact before your team uses it.
External runtimes can execute L0-L2 research and drafting work. Each run has an immutable brief, per-order cost ceiling, daily workspace guardrails, scoped result callback, and mandatory human decision.
Analyze the bounded work order and return evidence.
Create ideas, briefs, captions, and recommendations.
Return an artifact for mandatory human review without publishing.
Social actions remain outside the external runtime work-order boundary.
Validate a public HTTPS runtime endpoint and store its credential with authenticated encryption.
Freeze current capability evidence and require a different administrator to approve the runtime.
Send one bounded brief through a durable, idempotent dispatcher without social credentials.
Accept one size-limited artifact through a short-lived dispatch-scoped callback.
Accept or reject the result with an audited reason. Neither decision publishes a social post.
Enforce action and model-cost ceilings with zero as the fail-closed default.
The Agent Safety Center returns only tenant-safe connector, binding, principal, dispatch, and recent-run evidence. It never returns runtime URLs, credentials, fingerprints, provider tokens, raw responses, or hidden model reasoning.
The emergency stop disables automation, cancels work that has not begun, and blocks new claims. Ambiguous in-flight outcomes become human reconciliation tasks and are never retried blindly.
Start with one workspace, then connect channels and invite team members as you grow.