Autonomy control plane

Doctor, benchmarks, agent self-improvement, runtimes, routines, budgets, approvals, artifacts, memory, and operator access. The control surface for unattended runtime work.

The Autonomy tab is the operating control plane for Team-X execution. Mission Control shows what is happening now; Autonomy explains why execution is allowed, how it is governed, and which recurring or external systems are shaping the workload.

Use this page when preparing unattended work, reviewing runtime health, approving risky actions, or turning repeated failures into durable tickets.

Subviews

SubviewUse it for
DoctorRun deterministic readiness checks for database integrity, backups, runtime posture, secrets, providers, MCP health, and budget blockers.
BenchmarksReplay deterministic autonomy scenarios and inspect pass rates, duplicate-work prevention, recovery timing, spend, and artifact evidence.
ImproveRun the agent self-improvement loop and review open improvement tickets plus recent loop history.
RuntimesBind employees to explicit execution profiles and inspect live runtime posture.
RoutinesDefine recurring operating loops that materialize as visible work instead of hidden background automation.
BudgetsReview spend governance, warnings, hard caps, and approval thresholds across company, employee, runtime, and routine scopes.
ApprovalsProcess authority, planner, budget, and routine decisions from one operator inbox.
ArtifactsReview concrete runtime outputs, agent-created files, and evidence captured from autonomous execution.
MemoryInspect thread digests, resumable checkpoints, and packed-context posture for long-running work.
AccessReview local, invited, and cloud-ready operator membership posture for the workspace.

Agent Self-Improvement Loop

The Improve subview turns operational patterns into ordinary tickets so the team fixes process problems through the same durable queue as product work. It is the user-facing agent self-improvement loop: the loop does not silently patch behavior, hide failures, or mutate role prompts without review. It opens visible correction tickets.

When you click Run Improvement Loop, Team-X inspects recent events and tickets for these signals:

  • repeated work.failed events
  • runtime.execution.failed or runtime.session.stale events
  • tickets currently in blocked
  • tickets left in-progress for 48 hours or more

For each signal, Team-X opens a self-improvement ticket unless an open ticket for that same signal already exists. Improvement tickets are labeled with agent-improvement, self-improvement, agent-improvement:auto-created, and the signal-specific label such as agent-improvement:blocked-tickets.

The Improve panel shows:

  • open self-improvement ticket count
  • recent loop-run history
  • the latest run result, including inspected event count, inspected ticket count, recommendations, and created ticket IDs
  • direct links into the Tickets view for any open improvement ticket

How to Use It

Run Improve after any heavy work session, failed runtime run, provider disruption, repeated blocked-ticket pattern, or stale in-progress queue. Review the created tickets like normal work:

  1. Open the self-improvement ticket.
  2. Assign an owner or add participants.
  3. Attach evidence if needed.
  4. Ask the assigned employee to fix the process, prompt, runtime, checklist, or operating gap.
  5. Close the ticket only after the correction is implemented or intentionally rejected.

The loop dedupes by signal so rerunning it is safe. If the same blocked-ticket or stale-runtime pattern is already represented by an open self-improvement ticket, Team-X reports the recommendation without opening a duplicate.

Operating Pattern

Before launching long or external work:

  1. Run Doctor to confirm the workspace is ready.
  2. Check Runtimes so every employee has explicit execution posture.
  3. Review Budgets and Approvals so spend and authority blockers are visible before work starts.
  4. Use Benchmarks when you need repeatable evidence that runtime mechanics still behave correctly.
  5. Run Improve after failures, stalls, or a heavy work session so repeated problems become actionable correction tickets.
  6. Inspect Artifacts and Memory when the question is what a runtime produced or what context a long thread retained.

Agent-Created File Evidence

When an employee creates a deliverable with execution tools, Team-X writes the file inside that employee’s workspace. If vault storage is available, the same output is copied into the File Vault, tagged agent-created, and recorded as an Artifact with employee provenance.

Use Files to browse, verify integrity, search, and attach the output to tickets. Use Autonomy > Artifacts when you need the execution record: which employee created it, which vault record it points to, and how it fits into recent autonomous work.

  • Autonomy Doctor readiness checks
  • Autonomy benchmark harness evidence
  • Runtime operations snapshots
  • Agent wakeup requests
  • Budget governance and approval gates