# @lssm-tech/example.agent-workflow-command-center Website: https://contractspec.io **Agent Workflow Command Center - All-in-one operator cockpit for supervised AI agents in business workflows.** ## What This Demonstrates Agent Workflow Command Center is a ContractSpec example for operating supervised AI agents inside business workflows. It combines: - **Agent Management:** Fleet overview, configuration, tool permissions, lifecycle controls - **Workflow Orchestration:** State-machine workflows with agent assignments - **Human Review:** Review queue with approve/reject/escalate actions - **Governance:** Tool policies, risk thresholds, blocked actions - **Audit Trail:** Complete event timeline with deterministic evidence - **ROI & Handover:** Adoption metrics and handover readiness checklist ## Business Scenario A PME industrielle (85 employees) undergoing finance transformation: - Finance team under pressure with RAF departure - Cash tension and reporting delays - Fragmented Excel/BI tooling - Weak or outdated finance procedures NDconsulting installs supervised AI-assisted workflows and helps the client operate the system after handover. ## Core Demo Message "ContractSpec delivers a controllable operating system where the client team can: - Manage agents and assign them to workflows - Review AI-generated outputs before operational use - Enforce governance policies and block risky actions - Inspect audit logs and track adoption/ROI - Prove what happened through deterministic evidence" ## Included Agents 1. **Cash Aging Agent** - Prioritizes receivables follow-up using deterministic aging rules 2. **Reporting Narrative Agent** - Drafts reporting commentary with variance analysis 3. **Procedure Draft Agent** - Structures finance procedures with roles and controls 4. **Mission Intake Agent** - Triages transformation requests and classifies urgency 5. **AI Adoption Coach** - Tracks use cases and ROI without employee surveillance ## Workflows 1. **Cash Recovery Prioritization** - Intake → Agent Analysis → Human Review → Approved/Changes Requested 2. **Monthly Reporting Narrative** - Data Input → Narrative Draft → Variance Review → Finalization 3. **Finance Procedure Drafting** - Request → Procedure Structure → Control Review → Publication 4. **Mission Intake Triage** - Request → Classification → Priority Assignment → Follow-up 5. **AI Adoption / ROI Logging** - Usage Event → Impact Calculation → Trend Analysis ## Canonical Demo Path 1. Open `/sandbox?template=agent-workflow-command-center` 2. Start on the Landing screen - understand the value proposition 3. Enter the Command Center Overview - see active agents and workflows 4. Inspect the Cash Aging Agent - review allowed tools and governance 5. Open the Cash Recovery workflow - see state machine 6. Assign the Cash Aging Agent to the workflow 7. Trigger a deterministic agent run 8. Inspect run logs and emitted events 9. Open the review queue 10. Approve one recommendation, request changes on another 11. Show workflow state transition 12. Simulate a high-risk tool call - show it's blocked 13. Pause an agent, then resume after policy review 14. Open ROI & Handover - generate checklist 15. Export or display replay/audit proof ## Replay Proof The `./proof` export now exposes a deterministic replay contract for the canonical Managed CompanyOS journey: - `happyPath`: create/configure a synthetic agent, assign a deterministic tool, compose a review-gated workflow, dispatch a simulator run, submit human approval, and inspect audit replay evidence. - `blockedPath`: request a high-risk external tool and attempt a run with that blocked tool; both events must stay authorization-required with no external side effect. - `managedCompanyOsRouteRefs`: documents app-relative `pillarRoute` values and host-prefixed `hostRoute` values for agent, tool, and workflow managed surfaces without importing the managed bundle at runtime. - `atomicCrudCoverage`: proves the example covers tool, agent, and workflow list/create/update/detail expectations plus loading, empty, error, denied, read-only, allowed, dirty-invalid, review-required, and blocked states. - `assertCommandCenterReplayProof()` validates stable audit event IDs, evidence refs, correlation IDs, a human-review gate, a blocked path, and redacted/synthetic-only proof metadata. ## Tool Management Loop The command-center preview includes a Tools surface that demonstrates the full controlled loop: create/edit/archive deterministic tools, inspect governance readiness, assign tools to demo agents or host-owned workflow projections, and simulate calls with audit/replay evidence. The demo uses `@lssm-tech/module.agent-fleet` reusable presentation contracts with local state only; it performs no provider, network, backend, ERP, email, payment, or secret-storage side effects. ## Safety Boundaries - All agent outputs require human review before operational use - No real emails, payment instructions, or accounting entries sent - No connection to live finance systems - Synthetic fixtures only - no real client or financial data - Deterministic and offline-safe - Drafts/recommendations only, not operational commands - Tracks workflow ROI, not intrusive employee surveillance ## Core Entities - **AgentProfile** - Agent configuration, status, autonomy level, allowed tools - **WorkflowTemplate** - Workflow definitions with steps and transitions - **WorkflowInstance** - Running workflow with current state and assignments - **AgentAssignment** - Links agents to workflow steps - **AgentRun** - Execution record with inputs, outputs, events - **HumanReview** - Review queue with decisions and feedback - **ToolPolicy** - Governance rules for tool usage - **AuditEvent** - Immutable event log - **AdoptionMetric** - ROI and usage tracking ## Public Entry Points - Export `.` resolves through `./src/index.ts` - Export `./agent` resolves through `./src/agent/index.ts` - Export `./workflow` resolves through `./src/workflow/index.ts` - Export `./review` resolves through `./src/review/index.ts` - Export `./governance` resolves through `./src/governance/index.ts` - Export `./fixtures` resolves through `./src/fixtures/index.ts` - Export `./proof` resolves through `./src/proof/index.ts` - Export `./ui` resolves through `./src/ui/index.ts` - Export `./example` resolves through `./src/example.ts` - Export `./docs` resolves through `./src/docs/index.ts` ## Code Quality Review Review date: 2026-05-20. - The example package exposes deterministic domain, fixture, proof, UI, example, and docs surfaces; the root export also re-exports `./docs` for consumers that import the aggregate package. - Preview screens render from static PME finance fixtures and do not call live LLM, ERP, CRM, BI, email, payment, or accounting systems. Provider names in the fixture are labels only. - Governance and human-review screens are part of the default demo path, so supervised operation is visible before ROI or handover claims. - Replay proof stays under `src/proof/` and should be extended with any new workflow path before release. ## Production readiness evidence Use [`docs/agent-workflow-command-center-production-readiness.md`](../../../docs/agent-workflow-command-center-production-readiness.md) as the release-facing checklist for the boil-lake command-center slice. It defines the promotion gates, verification command matrix, redaction/import-boundary proof, replay determinism evidence, and authority gates that must be attached before claiming production readiness. ## Validation Checklist Before publishing changes to this template, run from `packages/examples/agent-workflow-command-center`: 1. `bun run lint:check` - package lint and formatting gate 2. `bun run typecheck` - public TypeScript entrypoint validation, including `./docs` 3. `bun run test` - preview shell and adaptive agent-fleet DataView tests 4. Open `/sandbox?template=agent-workflow-command-center` and follow the canonical demo path ## Local Commands - `bun run dev` - Development build - `bun run build` - Production build - `bun run test` - Run tests - `bun run typecheck` - TypeScript check - `bun run lint` - Lint and fix - `bun run preflight` - Full validation ## Architecture - `src/agent/` - Agent profile domain - `src/workflow/` - Workflow template and instance domain - `src/review/` - Human review domain - `src/governance/` - Tool policies and governance - `src/fixtures/` - Synthetic demo data - `src/proof/` - Replay evidence - `src/ui/` - Preview components and screens ## Notes Works alongside `@lssm-tech/lib.contracts-spec`, `@lssm-tech/lib.design-system`, `@lssm-tech/lib.schema`. ### Tool Effects The Tools screen demonstrates typed tool effects: `contract_effect` entries select a ContractSpec operation, and `integration_effect` entries select an already-integrated provider catalog item. The example validates and simulates effects deterministically by default. Real execution is represented as an explicit adapter boundary so the template stays secret-free and provider/network-free unless a host wires an adapter deliberately. ### Tool Effect Workbench The Tools screen now demonstrates a decision-quality operation workbench for tool effects. Operators can inspect richer ContractSpec and integration-call fixtures, compare candidates, configure input mappings, validate readiness, and simulate deterministically before saving. Real execution remains an explicit host-adapter boundary so the example stays provider-free, backend-free, and secret-free by default.