Playbooks · Coding tasks
Agenta 0.128.13: tool budgets extend the watchdog
Written November 13, 2026
Agenta 0.128.13 lets a tool's own time budget extend the runner watchdog, gives workflow reference tools a long limit, and shows new automations live.
What changed
Agenta 0.128.13 is a small patch that follows 0.128.12 in the same fast-moving line. The 0.128.13 release notes list four changes, and three of them are fixes with a shared theme: long-running tool work and things an agent creates should not be cut off or hidden by the platform around them. The fourth change is a documentation update. There are no new headline features in this release, and nothing in the notes changes the platform's overall shape. What changes is how patient the runner is, how long one class of tools is allowed to run, and whether a new automation appears in the interface on its own.
That makes this a release for people who run agents that call slow tools. If your agents finish quickly, 0.128.13 will pass almost unnoticed. If a run has ever ended because a watchdog stopped a tool that was still working, or a workflow step timed out while reaching for a reference, the runner and API fixes are the reason to take this patch.
A tool's own budget can extend the watchdog
The runner fix, in the notes' own wording, lets a tool's own time budget extend the tool-call watchdog. A watchdog exists to stop a call that runs too long, which protects a session from a tool that hangs. The problem with a single fixed limit is that tools differ. A quick lookup and a long computation do not belong under the same ceiling. A tool that declares a longer budget is saying, in advance, that its work legitimately takes longer. Under 0.128.13, that declared budget can stretch the watchdog instead of the watchdog cutting the call at its own default point.
The practical effect is narrower than it may sound. This is not an open-ended removal of time limits, and the notes describe no new setting to tune. It is a change in which limit wins when the two disagree: the tool's stated budget now counts. Teams whose tools declare sensible budgets should see fewer premature cutoffs on their longest calls. Teams whose tools declare no budget get less from this fix, because there is nothing for the watchdog to honour.
Workflow reference tools get a long time limit
The API fix gives workflow reference tools a long time limit. Reference tools are the supporting calls a workflow makes behind its main task, rather than the task itself. A short limit on those calls creates an awkward failure: the main work is healthy, the agent is behaving, and the step still dies because a reference lookup took longer than the allowance around it. Raising that allowance to a long limit removes one brittle edge from workflow runs.
The notes give the change in a single line and do not publish the new limit's length. So the honest summary is directional rather than numerical: workflow reference tools have more time than they had in 0.128.12, by an amount the release list does not state. Anyone whose capacity planning depends on the exact number will need to look at the underlying change rather than plan from the release line alone.
Agent-created automations appear without a reload
The frontend fix shows agent-created automations without a reload. Before this patch, an automation an agent created during a session could exist in the platform while the open page still showed the older list. The person watching had to refresh before they could see, open, or trust what had just been made. Now the new automation appears on its own.
Small as it sounds, this is the change most users will actually notice. The two time-limit fixes prevent failures that, when they work, leave no trace. This one closes a visible gap between what the agent did and what the screen showed, and removes the doubt that comes with it — whether the automation failed to save, or saved somewhere else.
The remaining item is an update to the API reference documentation. It carries no behaviour change on its own, but it is the place teams building against the API should look first if a documented call shape and their integration disagree after the update.
What to do, and the limits
- Take the patch if your runs call slow tools. The watchdog and reference-tool fixes only matter where time limits have been biting, and that is exactly where they help.
- Look at the budgets your own tools declare. The watchdog change rewards a tool that states a realistic budget; a missing or sloppy budget leaves the old behaviour in charge.
- Expect an automation an agent creates to show up without refreshing the page. If it does not, that is now a bug worth reporting rather than normal behaviour.
- The notes are four single lines. They give no new limit values, no new settings, and no migration steps, so do not plan configuration work around this release.
- This is a patch in a quick line: 0.128.12 brought the session and channel changes covered in the earlier page, and those are not new here. Read 0.128.13 as a patience and visibility fix, not a feature release.
Related playbooks
- Agenta 0.128.12: sessions that keep up
- A critical flaw in gemini-mcp-tool, fixed in 1.1.6
- Build a small CLI tool end to end with an agent
- Claude Code 2.1.323: Code Mode writes the tool calls
Sources
Back to: Coding playbooks hub · Also: Agenta 0.128.12: sessions that keep up · Apify MCP Server 0.21.8 hardens the bridge