Repository navigation
Add a fixed native SDK recovery diagnostic - #143
Open
dfrysinger wants to merge 9 commits into
Open
dfrysinger wants to merge 9 commits into
dfrysinger wants to merge 9 commits into
Conversation
Stage an optional instruction carrier in an isolated home, verify enabled sources before import and each resumed phase, and pass retained tool logs to judges. Keep default suppression and tool grants unchanged. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: cf89240b-a65d-414a-9123-5c1dc9c40c7d
Separate reconstructed replay controls from exact installation claims while retaining materiality and execution gates. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: cf89240b-a65d-414a-9123-5c1dc9c40c7d
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: cf89240b-a65d-414a-9123-5c1dc9c40c7d
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: cf89240b-a65d-414a-9123-5c1dc9c40c7d
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: cf89240b-a65d-414a-9123-5c1dc9c40c7d
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: cf89240b-a65d-414a-9123-5c1dc9c40c7d
Exercise native todo and schedule recovery under ordinary permission mediation. Validate correlated SDK completion, preserve first outcomes, reject output overlap with original inputs and require serial schedule reconciliation. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: cf89240b-a65d-414a-9123-5c1dc9c40c7d
dfrysinger
changed the base branch from
main
to
dfrysinger/replay-reconstruction-policy
November 7, 2026 18:04
dfrysinger
marked this pull request as draft
November 7, 2026 18:05
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: cf89240b-a65d-414a-9123-5c1dc9c40c7d
dfrysinger
marked this pull request as ready for review
November 7, 2026 18:55
dfrysinger
marked this pull request as draft
November 7, 2026 18:59
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: cf89240b-a65d-414a-9123-5c1dc9c40c7d
dfrysinger
marked this pull request as ready for review
November 7, 2026 19:10
dfrysinger
force-pushed
the
dfrysinger/replay-reconstruction-policy
branch
from
November 10, 2026 00:48
18c890a to
604c55f
Compare
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
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.
Adds a fixed, isolated native recovery diagnostic to the existing evaluator. It uses CLI97's paired public SDK and normal permission mediation to verify restored context, successful skill loading, native todo and schedule recovery, a matching final response followed by idle, independent cold state and cleanup. Ordinary prose and repository evaluation keep their existing behavior.
The diagnostic rejects output inside original inputs, including externally resolved frozen revisions. Candidate messages use immediate delivery. Schedule recovery requires each list/stop/re-list call to finish before the next starts. Native calls use the remaining owner budget; abort, graceful stop and forced stop have bounded fallbacks. The caller does not clean a fixture after setup failed to establish ownership. Owner preflight reads cannot count as model actions, and workflow authentication stays in the candidate SDK option rather than arguments or retained files.
The affected Python modules pass 126 tests, the native helper passes 46 checks, and an isolated Actions candidate passes all 13 native checkpoints without judges or retries. Both independent source-review families closed the material findings across the implementation and its SDK foundation.
This is synthetic execution-route proof, not historical reproduction, skill improvement or shared-suite qualification. Private fixtures, transcripts, runtime archives and evidence are not included. Merge, product release and installation remain separate.
The failed setup producer attempts cleanup only of its proven-owned fixture. If that cleanup itself fails, the failed run has no guaranteed second caller retry and cannot qualify or start a candidate. Original-prefix correspondence and historical profiles remain separate admission gates.