Valletta OpenClaw Onboarding v1.7 State-Embedded Queue Model

Multi-Level Onboarding Architecture

Level 1 shows OpenClaw multi-agent orchestration where queue is embedded in onboarding state. Level 2 shows internal execution contracts for each agent/tool binding.

Master process owner: Onboarding Agent Queue = state.pending_actions[] (not external) Tools are adapters with explicit confirmations Worker = agent with a tool binding Single-page artifacts with modal preview
Open Onboarding Admin Preview

Program Metrics

1

Master onboarding agent controlling progression and state.

3

Worker agents: Admin, Document, Consultant (agent + tool model).

1

Queue implementation: pending_actions[] inside state.

16

Concrete adapter tools listed and mapped to execution actions.

Level 1: Actor and Agent Interaction Map

Deterministic Execution Layer (OpenClaw adapters plus workers) Security Boundary: secrets, VPN keys, signed contracts, ACL-sensitive actions conversation guard by state event -> rule -> notify pending | confirmed | failed masked logs + ACL scope + retention Recruiter Control Plane Routing Layer Slack Entry Onboarding Agent (Master) State Owner Conversation Guard State-Gated Dialog Candidate Slack Primary Human Owners HR/Legal/Admin Heartbeat Engine SLA Watcher SLA Timer Per-Step Deadlines State Machine Engine Allowed Transitions State Layer Current + Event Log + Pending Actions Notification Rules Event-Driven Triggers Telegram Fallback Preboarding + Rescue Agent Action Loop Pick + Execute + Retry + Update Calendar Adapter AI + Tech Events Admin Worker Execution Document Worker Structured Result Only Consultant Worker FAQ + English Error / Escalation Channel Manual Takeover Pipeline Idempotency Gate No Duplicate Actions Security Gate Masking + ACL + TTL Notification Adapter Slack + Email + TG Fallback Admin Adapter Account Actions Contract Adapter Template + Registry Storage Adapter Docs + Signed Contracts Account Manager ACL Command Path
Primary orchestration path Worker to adapter execution Fallback or escalation channels Risk control layers (queue, idempotency, security)

Level 2: Internal Flow Inside Each Agent

Select an agent to inspect its low-level execution sequence, inputs, outputs, and handoff contracts.

Onboarding Agent (Master)

Owns process progression, state writes, stakeholder communication, and escalation decisions.

Critical Integration and Approval Gates

Gate Condition to Pass Evidence If Not Passed
Routing Gate Start payload arrives in approved Slack context Validated payload and onboarding id Reject and request strict template correction
Account Provision Gate Google/Jira/Slack/VPN actions confirmed adapter evidence ids and refs move to manual_required with admin escalation
Contract Selection Gate Deterministic rule match found rule id, template id, registry id contract_selection_review_required and HR/Legal review
Signed Contract Gate Signed copy verified and stored signed contract URI and notifications follow-up cycle and accounting escalation
Completion Gate Contract, English, handoff, and required tails complete completion event record stay open or transition to manual takeover path

OpenClaw Skills and Integration Overlay

Architecture is mapped to OpenClaw-native capabilities where possible, and to explicit custom adapters where native coverage is not guaranteed.

Architecture Block OpenClaw Capability Delivery Mode Control Notes
Routing + Recruiter Start Thread context, strict payload parser prompts, policy docs native + config Approved channel only; RM-thread ignore rule enforced.
State Machine + State Layer JSON state schema, append-only event log, pending_actions[], state adapter contract custom adapter on OpenClaw workspace No stage transition without state-machine validation.
Agent Action Loop (No External Queue) Onboarding-agent runtime picks and processes pending_actions[] agent runtime logic Agent owns pick/execute/retry/update cycle and evidence validation.
Calendar + Mail + Docs Google connector family (calendar, mail, storage) or delegated API adapters native if available, otherwise custom adapter All outputs must return event id / uri evidence.
Admin Provisioning + Account Manager Admin adapter and command bridge to ACL-restricted account-manager path custom adapter Mandatory feedback loop: pending/confirmed/failed written back to state.
Contract and Storage Flow Template registry adapter, storage adapter, deterministic rule file custom adapter Document worker returns structured result only, no direct state writes.
Heartbeat + SLA OpenClaw heartbeat/cron scheduling plus SLA checker logic native scheduler + custom rules Each state has deadline checks and escalation targets.
Error/Escalation + Security Runbook automation, masked observability logs, manual takeover templates policy + custom pipeline Sensitive data stays in restricted storage and masked audit logs.

OpenClaw Simplified Execution Model

Rule Implementation Why It Matters
Queue is not external state.pending_actions[] in onboarding state JSON No hidden queue service; full traceability in one source of truth.
Tools act as adapters Deterministic calls like admin_adapter.create_google_user Agent does not improvise external actions.
Agent owns action loop Pick next pending action, execute, retry, update state, append event Prevents false success and uncontrolled progression.
Worker = agent with a tool Admin/Document/Consultant workers are scoped tool-bound agents Specialization without process ownership fragmentation.

Concrete tool bindings used in this architecture:

Worker or Layer Concrete Tools Expected Confirmation
State Layer state_adapter.write_current_state, state_adapter.append_event state version/hash and event id
Admin Worker admin_adapter.create_google_user, admin_adapter.invite_slack, admin_adapter.invite_jira, admin_adapter.create_vpn_ticket, admin_adapter.resolve_account_manager_status pending/confirmed/failed with evidence
Calendar Layer calendar_adapter.create_event event id and meet link
Document Worker contract_adapter.select_rule, contract_adapter.generate_contract rule id, registry number, contract uri
Storage Layer storage_adapter.create_candidate_folder, storage_adapter.store_document, storage_adapter.store_signed_contract folder and artifact uris
Notification Layer notification_adapter.send_slack, notification_adapter.send_email, notification_adapter.send_tg_fallback delivery status and message id

Real Example: Ivan Ivanov (John Maltovich)

Below is a real execution walk-through in this exact OpenClaw model.

Input Field Value
Candidate full nameIvan Ivanov
Display nameJohn Maltovich
RoleBackend Developer
Start date/time2026-05-12 10:00
Personal emailivan@example.com
Telegram@ivan_ivanov
VPN requiredyes
Interview linkprovided

Execution flow:

Step What Happens State and Tools
1Recruiter sends strict start payload in onboarding Slack entrypoint.Routing sends request to onboarding-agent.
2Agent creates onboarding record.onboarding_id=ONB-2026-0512-IVANOV, status=created, state_adapter.write_current_state.
3Agent validates fields and initializes process shell.status=preboarding_initialized, initial pending_actions[] created.
4Calendar actions are added and executed.calendar_adapter.create_event for AI and Tech events, save event ids and meet link.
5Agent requests access stack.Pending actions for Google, Slack, Jira, VPN via admin adapter tools.
6Admin worker executes account actions.admin_adapter.create_google_user, invite_slack, invite_jira, create_vpn_ticket.
7Account manager feedback loop is reconciled.admin_adapter.resolve_account_manager_status, statuses move through pending|confirmed|failed.
8Preboarding package is sent.notification_adapter.send_email/send_slack, candidate_preboarding_sent=true.
9Heartbeat checks deadlines before start.Reads state and SLA windows, escalates only if blocked.
10Candidate sends first Slack message at start time.Conversation guard validates stage, set candidate_started=true.
11Agent runs contract questionnaire.contract_questionnaire=collecting then completed.
12Candidate uploads documents.storage_adapter.store_document, documents_status=received|partial.
13Document worker runs deterministic contract flow.contract_adapter.select_rule, contract_adapter.generate_contract, structured output only.
14Agent sends generated contract to candidate.status=awaiting_signed_contract, notification tools used.
15Candidate signs and returns contract.storage_adapter.store_signed_contract, set contract_signed=true.
16Recruiter/accounting are notified.notification_adapter.send_slack, notification_adapter.send_email.
17English written and FAQ/checklist are completed.Structured score saved in state, optional April School request.
18Tech onboarding handoff is sent from stored event artifact.tech_onboarding_link_sent=true.
19Post-onboarding tails are closed and process is finalized.post_onboarding_status=completed, final status=completed.

Control points in this example:

Control Point What Must Be True
CP1 StartState created and onboarding id assigned.
CP2 AccessGoogle/Slack/Jira/VPN are confirmed by tool evidence.
CP3 Show-upFirst candidate Slack ping is recorded in state.
CP4 Contract pre-sendQuestionnaire and required docs are complete.
CP5 Signed contractSigned file is stored and notifications are sent.
CP6 CompletionPost-onboarding tails are complete before final close.

Artifact Viewer (Single Page with Modal)

All artifacts are viewable here without leaving the page. Click Preview to open content in modal.