[Copilot]

pr-sous-chef is the highest-traffic tracked workflow, making it the best candidate to test whether removing a redundant "Required skip rules per PR" backup table from the main-agent prompt reduces execution duration without regressing eval pass rates — the same three skip conditions are already enforced deterministically by the prefilter step and re-verified by the pr-processor sub-agent.

Changes

• Experiment: Added remove_redundant_context_v1 (control vs. candidate) to pr-sous-chef.md frontmatter, targeting grader:execution-duration as the primary metric with eval:comment-added / eval:pr-evaluated guardrails (≥0.90 pass rate), mann_whitney analysis, and a 15,000ms minimum effect threshold.
• Graders: Declared graders: execution-duration: {} — since any explicitly listed built-in grader causes the rest of the built-in registry to populate with defaults, this reliably surfaces execution-duration even after merging with imported custom graders.
• Shared graders import: Added shared/graders.md to imports:, pulling in the custom trajectory graders (recurrence-determinism, recurrence-laminarity, state-revisit-probability-rep).
• Prompt gating: Wrapped the "Required skip rules per PR" table in a template conditional so it's only rendered for the control variant, keeping candidate behavior otherwise identical:

{{#if experiments.remove_redundant_context_v1 == 'control' }}
## Required skip rules per PR
...
{{#endif}}

• Recompiled pr-sous-chef.lock.yml to reflect the frontmatter changes.

• Fixes [harness-experiment-proposal] pr-sous-chef — context assembly/remove_redundant_context A/B harness experiment #55719

[pelikhan]

@copilot resolve the merge conflicts in this pull request

[Copilot]

Pull request overview

Adds an A/B experiment to measure whether removing redundant skip-rule context improves PR Sous Chef execution duration without reducing evaluation pass rates.

Changes:

• Adds balanced control/candidate experiment configuration and grader instrumentation.
• Gates the redundant skip-rules table to the control variant.
• Regenerates the compiled workflow with experiment persistence and grader artifacts.

Show a summary per file

File	Description
.github/workflows/pr-sous-chef.md	Defines the experiment, graders, and prompt gating.
.github/workflows/pr-sous-chef.lock.yml	Compiles experiment assignment, grading, and state persistence.

Review details

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Balanced

[Copilot]

@copilot resolve the merge conflicts in this pull request

Merge conflicts resolved in 9c60c51 (merged main, recompiled pr-sous-chef.lock.yml).

[github-actions]

🎉 This pull request is included in a new release.

Release: v0.87.8