Repository navigation
docs: a proposal for resuming a run, from its failure or from a step - #37
Merged
Merged
Conversation
A run that fails on its fortieth firing is run again from its first. This proposes resuming instead: atomic firings, so a failed firing keeps its inputs; a checkpoint written when a run fails, restored by `--resume-from`; and the queue trace the IDE already steps through, extended to record what each firing consumed, produced and changed, so any step it shows can be resumed from with `--at-step` and a button beside the stepper. Claude-Session: https://claude.ai/code/session_015VK7fH1c4aKbexnq2QcuKU
This was referenced Oct 1, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A run that fails on its fortieth firing is run again from its first: every agent asked again, every tool run again.
docs/proposals/resume.mdproposes resuming instead, in four phases:--resume-from. It captures the queues (with values), actor and control-node state, scopes, context and sub-plans as JSON with type tags (no pickle). The resume refuses a changed graph and accepts changed action bodies.--at-step. It builds on the trace the IDE's ◀ step ▶ stepper already reads. Each step gainsconsumed,producedand the actor's state, the trace is written on failure and appended as the run goes, and any step can be replayed into a state without running anything.Nothing is implemented by this PR.
https://claude.ai/code/session_015VK7fH1c4aKbexnq2QcuKU