[Plugin] dsh-session-integrity: diagnose and safely continue poisoned sessions #3555
DON738110198
started this conversation in
Show Your Plugins!
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
I built dsh-session-integrity, a zero-dependency Cordis plugin and offline CLI for diagnosing provider-visible Session defects and continuing from a provider-safe boundary without rewriting history.
The original target is the failure family discussed in #3524: an assistant tool request becomes durable, but the model-visible surface has no matching tool result. A closed error turn can then survive tail repair, so later provider requests keep rejecting the same transcript. The bilingual README now explains that full failure chain, why dispatched calls are outcome-unknown, and why blind retries can duplicate external side effects:
v0.2.1: safe recovery MVP
recovercommand finds the latest completed provider-safe prefix and emits a redacted plan.This deliberately does not implement in-place rollback, hard delete, automatic archive, tool retry, or rollback of external side effects. Hard deletion needs an explicit core persistence/API policy rather than a browser-only workaround.
Install
Offline diagnosis and recovery planning:
Verification
dsh-v0.1.0-rc.8Web profile.dsh-session-integritywith its declared dependencies, and the served client bundle registers the recovery action.Repository and bilingual docs: https://github.com/DON738110198/dsh-session-integrity
Release: https://github.com/DON738110198/dsh-session-integrity/releases/tag/v0.2.1
The separate core patch for the new-session write path remains linked from #3524. Detection/recovery and prevention are intentionally kept as distinct review surfaces; that core proposal has not been merged upstream.
All reactions