Add A/B experiment for rendering verifier prompt context

[Copilot]

daily-rendering-scripts-verifier has the fleet’s highest average AI-credit cost, partly due to four redundant closing guidelines. This change measures whether removing that duplicated context improves execution duration without reducing verification correctness.

• Experiment
  • Adds balanced control/candidate variants via remove_redundant_context_v1.
  • Uses grader:execution-duration as the primary metric.
  • Guards correctness with eval:correct_verification_outcome >= 0.90.
• Prompt mutation
  • control preserves the existing Guidelines section.
  • candidate retains only the unique safety rule.
• Generated workflow
  • Recompiles the corresponding .lock.yml with experiment assignment, eval, and grader wiring.

experiments:
  remove_redundant_context_v1:
    variants: [control, candidate]
    metric: "grader:execution-duration"
    guardrail_metrics:
      - name: "eval:correct_verification_outcome"
        threshold: ">=0.90"

• Fixes [harness-experiment-proposal] daily-rendering-scripts-verifier — context assembly/remove_redundant_context A/B harness experiment #56330

[Copilot]

🟡 Changes recommended

The control arm is never selected, the correctness guardrail is inadequate, and the router references a missing file.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Adds an A/B experiment to evaluate whether trimming redundant verifier prompt guidance reduces execution time without harming correctness.

Changes:

• Adds control/candidate prompt variants with duration and correctness metrics.
• Regenerates experiment, grader, and evaluation workflow wiring.
• Adds compatibility guidance routing.

File summaries

File	Description
.github/workflows/daily-rendering-scripts-verifier.md	Defines the experiment and conditional prompt mutation.
.github/workflows/daily-rendering-scripts-verifier.lock.yml	Compiles experiment, grader, evaluation, and state jobs.
.github/skills/agentic-workflows/SKILL.md	Adds a reference to a nonexistent compatibility document.

Review details

Suppressed comments (1)

.github/skills/agentic-workflows/SKILL.md:27

• The PR description and its validation explicitly scope this change to the verifier source and generated lock file, but this unrelated router update adds a third changed file. Remove it from this PR, or document why compatibility-advisory guidance is part of the experiment change and update the stated validation scope.

- `.github/aw/compat.md`

• Files reviewed: 3/3 changed files
• Comments generated: 3
• Review effort level: Balanced

[github-actions]

🎉 This pull request is included in a new release.

Release: v0.88.2