fix(skills): make auto-optimize delivery evidence-bound - #1457
Merged
Qiong Wu (qiowu) (DingmaomaoBJTU) merged 1 commit intoSep 29, 2026
Merged
Conversation
Qiong Wu (qiowu) (DingmaomaoBJTU)
marked this pull request as ready for review
September 29, 2026 06:36
Qiong Wu (qiowu) (DingmaomaoBJTU)
requested a review
from a team
as a code owner
September 29, 2026 06:36
xieofxie
approved these changes
Sep 29, 2026
Qiong Wu (qiowu) (DingmaomaoBJTU)
enabled auto-merge (squash)
September 29, 2026 06:40
Qiong Wu (qiowu) (DingmaomaoBJTU)
deleted the
codex/auto-optimize-reliability
branch
September 29, 2026 06:41
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Auto-optimize could restart planning for supplied candidates, omit report metrics despite available raw samples, and claim delivery without consistently checking the full artifact chain.
This change adds a bounded workflow entry point for evidence-hash resume, deterministic report preparation, and final delivery through the existing finalizer and promotion helpers. It unifies role/PR ownership and stopping rules, displays native cycle metrics and explicit missing-data reasons, and computes percentiles from raw timed samples. Regression coverage includes duplicate/incompatible sessions, stale evidence, skipped gates, PR-after-stop attempts, report rendering and encoding.
Validation: 362 skill tests passed; Ruff check/format and all commit hooks passed. Existing live-agent trial records are retained with their scope limitations. Real QNN reproduction artifacts and full transcripts remain run-local and are not included. Evidence records are attestations with hash verification, not authenticated proof that hardware execution or independent review occurred.