Is there a pause action?

No. The master has three actions: `continue`, `stop`, and `revert`. There is no soft pause, because agent completion times are non-deterministic and a time-based lock would deadlock or starve. Changing what an agent knows means terminating the task and starting a fresh one.

How far back can a post-mortem reconstruct?

The append-only event log is the source of truth, so the master's state at any prior point is reconstructable by replaying events. The history view reads the same log.

Use case

Plan-stage incident response

Mesh of coordination nodes converging on a central decision point

By Tetherlab team

Updated 2026-05-29

Overview

When something goes wrong across a fleet of agents, the question is not just what changed. It is what each agent intended, in what order, and how to undo it without unwinding everything around it.

Freezing the swarm mid-flight

The master can stop an active task. The task terminates and local files are left as they are. Because the master is the single authority for one git remote, a stop reaches every coordinated agent, not just the one in front of you.

There is no wait or lock to manage. The master changes what agents are allowed to do by terminating tasks, so a freeze is decisive rather than a queue of paused processes that can starve.

Replaying the intent timeline

Every registry mutation is an event on an append-only log: intent submission, concept match, conflict detection, plan completion. The log is the source of truth, and the dashboard's history view reads from it.

A post-mortem walks the timeline of intents and verdicts in order. You can reconstruct the master's state at any prior point, which turns what happened from guesswork into a replay.

Cross-machine revert

A revert verdict terminates the active task and the shim rolls the working tree back to the pre-task state. The action is defined at the agent level, so reverting bad work does not depend on which laptop produced it.

Revert is scoped to the task

Revert rolls a coordinated task's working tree back to its pre-task snapshot. Git history still lives underneath; Tetherlab decides what to undo at the intent layer.

Incident playbook

01Stop the active intents touching the affected concepts. The terminate is immediate, no lock to clear.
02Open the history view and replay the intent timeline to find the first bad verdict.
03Revert the offending task; the shim restores the working tree to its pre-task state.
04Re-evaluate dependent intents against the new world state before letting agents continue.

Next steps

The same signed timeline that drives a post-mortem also makes the system audit-grade by construction.

Frequently asked questions

Is there a pause action?: No. The master has three actions: continue, stop, and revert. There is no soft pause, because agent completion times are non-deterministic and a time-based lock would deadlock or starve. Changing what an agent knows means terminating the task and starting a fresh one.
How far back can a post-mortem reconstruct?: The append-only event log is the source of truth, so the master's state at any prior point is reconstructable by replaying events. The history view reads the same log.