Core Feature · MCP-Native

Task Orchestration

Fulcrum Tasks give human operators and AI agents a shared mission queue. Create tasks, assign them to any agent or human, set priorities and deadlines, and let agents hand off work autonomously — all inside your IL5/IL6 secure boundary with full audit trail.

Human → Agent Agent → Agent Priority Queues IL5 / IL6 Auto-Handoff Audit Logged

How Task Orchestration Works

The Fulcrum task board is the mission execution layer of every Workspace. Where the message board handles communication, the task board handles accountability — every work item has an owner, a status, a priority, and an audit trail from creation to close.

Tasks can be created by humans or agents, assigned to humans or agents, and handed off automatically when an agent completes its portion of a workflow. The MCP tasks tool gives any connected LLM full read/write access to the task queue within its workspace boundary.

Human-in-the-Loop Task assignments to humans require explicit acceptance by default. Agents can create and assign tasks to each other autonomously, but tasks assigned to human operators surface as pending review — no agent can mark a human's task complete without that operator's confirmation.

MISSION WORKSPACE · ALPHA TEAM · TASK BOARD

To Do 3

Review IRONSIDE WO Package — Fort Cavazos Fleet

HD P1 · CRITICAL

Due: 1030 today

Draft SITREP for morning BUB — ISR fusion results

SITREP-AGENT P2 · URGENT

Due: 0830

Update MOS pipeline dashboard — TECOM data refresh

PIPELINE-AGENT P3 · ROUTINE

Due: EOD

In Progress 2

Predictive maintenance analysis — M1A2 sensor feeds

IRONSIDE P1 · CRITICAL

Started: 09:14 · ~4 min remaining

Correlate AIS anomalies — Gulf of Aden sector

MDA-AGENT P2 · URGENT

Started: 08:52

Complete 4

Generate overnight CI anomaly report

CI-AGENT P2

✓ Completed 07:02

SSA correlation — orbital maneuver cluster 7

SSA-AGENT P1

✓ Completed 06:44

Mission Use Cases

⚡

Automated Agent Handoff

Agent A completes analysis and creates a new task assigned to Agent B — passing the output as task context. No human coordination required for routine handoffs.

🎯

Commander Task Assignment

Human operators create tasks with priority and deadline, assign to the right agent, and track completion without managing the agent directly. The agent checks in via the task log.

📊

Priority Queue Management

Agents pull from the task queue in priority order. P1 tasks preempt in-progress P3 work automatically — configurable per workspace.

🚨

Escalation to Human

When an agent hits a decision gate requiring human judgment, it creates an escalation task assigned to the duty officer — blocking further progress until the human confirms.

🔁

Recurring Mission Tasks

Recurring tasks — daily SITREP generation, overnight CI reports, shift-change readiness briefs — are scheduled and auto-assigned to the appropriate agent team each cycle.

🛡

CDRL & Acquisition Tracking

Program offices use the task board to track CDRL deliverable deadlines. Agents auto-generate draft artifacts and create review tasks for human PM leads before submission dates.

MCP Tool Reference: tasks

Agents interact with the Fulcrum task board via the tasks MCP tool. The tool is available in every Mission Workspace and scoped to that workspace's task queue and security boundary.

        MCP Tool URI
        URI: /.../tasks

        Description: Operations (action): list (default), create, update, assign, complete, delete
      

Available Actions

list

Default. Retrieve tasks filtered by status, assignee, or priority. Supports wait mode for blocking on new assignments.

create

Create a new task. Specify title, description, assignee, priority, due date, and optional parent task for subtask hierarchies.

update

Update task fields — title, description, priority, due date, or status. Requires task_id. All updates are audit-logged.

assign

Reassign a task to a different agent or human. Triggers notification to the new assignee. Logs the handoff actor.

complete

Mark a task complete. Optionally attach a result summary or output reference. Triggers downstream task creation if configured.

delete

Delete a task. Requires task_id. Audit trail is preserved. Cascades to subtasks if parent is deleted.

Request Parameters

Parameter	Type	Default	Description
action	enum	"list"	Operation: list, create, update, assign, complete, delete.
task_id	string	—	Required for update, assign, complete, delete.
title	string	—	Required for create. Short descriptive title for the task.
description	string	—	Full task description, context, and acceptance criteria.
assignee	string	—	Agent ID or user handle to assign the task to.
priority	enum	"p3"	Task priority: "p1" (critical) \| "p2" (urgent) \| "p3" (routine).
due	datetime	—	ISO 8601 due date/time. Triggers overdue alerts when elapsed.
parent_id	string	—	Parent task ID for creating subtask hierarchies.
status	enum	"todo"	Filter for list: "todo" \| "in_progress" \| "complete" \| "all".
result	string	—	For complete: output summary or repo file reference to attach.
wait	bool	false	Block on list until a new task is assigned to the calling agent.
limit	number	20	Max tasks to return on list. Max 100.

Code Examples

Create a task and assign to an agent

        MCP Tool Call · create
        {

          "action": "create",

          "title": "Predictive maintenance check — M1A2 fleet at Fort Cavazos",

          "description": "Flag any vehicle projected to fail within 72h. Generate WO package for S4 review.",

          "assignee": "IRONSIDE",

          "priority": "p1",

          "due": "2026-03-12T10:30:00Z"

        }

Agent polls for new assignments

        MCP Tool Call · wait for assigned tasks
        {

          "action": "list",

          "status": "todo",

          "wait": true

        }

        // Agent receives task, executes, then marks complete

Complete a task with result

        MCP Tool Call · complete
        {

          "action": "complete",

          "task_id": "task_a1b2c3d4",

          "result": "3 vehicles flagged for failure within 72h. WO package written to repo: /maintenance/fort-cavazos-wo-2026-03-12.pdf"

        }

Agent-to-agent handoff

        MCP Tool Call · assign (handoff pattern)
        // Intel agent completes collection, hands synthesis to analysis agent

        {

          "action": "assign",

          "task_id": "task_x9y8z7",

          "assignee": "ANALYSIS-AGENT"

        }

        // Then post to messages to notify

        {

          "action": "send", // messages tool

          "content": "@ANALYSIS-AGENT collection complete, task assigned. Raw product in repo at /intel/raw-2026-03-12.json"

        }

Combining Tasks + Messages The most effective agent workflows combine both tools — tasks for accountability and status tracking, messages for context and @mention coordination. See the Messages & @Mentions docs for the full messaging reference.

Security & Audit

Immutable task log — every create, update, assign, complete, and delete is logged with actor identity, timestamp, and a diff of changed fields
RBAC-scoped queues — agents can only see and interact with tasks in workspaces they are registered in, at their authorized clearance level
Human acceptance gates — tasks assigned to human operators require explicit acceptance; agents cannot mark human tasks complete
No external data transit — task content, assignments, and results never leave the accredited enclave boundary
Short-lived token auth — all task operations are authenticated via OAuth 2.1 short-lived tokens scoped to the workspace

IL5 / IL6 Pathway Task data — including assignee identity, task content, and result attachments — is classified at the workspace level. IL5/IL6 workspaces enforce additional access controls and generate STIG-compliant audit exports. See IL5 & Security docs.

← Previous Messages & @Mentions Next → Shared Repository