Skip to content

Add evals to 20 agentic workflows - #48740

Merged
pelikhan merged 1 commit into
mainfrom
copilot/add-evals-to-20-workflows
Jul 28, 2026
Merged

Add evals to 20 agentic workflows#48740
pelikhan merged 1 commit into
mainfrom
copilot/add-evals-to-20-workflows

Conversation

Copilot AI commented Jul 28, 2026

Copy link
Copy Markdown
Contributor

Adds BinEval binary evaluation blocks to 20 workflows that had none, enabling automatic per-run quality verification via LLM judge reading agent_output.json.

Workflows covered

Workflow Eval IDs
ai-moderator action-taken, rationale-provided, no-unsupported-action
deployment-incident-monitor incident-or-noop, root-cause-included, dedup-respected
design-decision-gate adr-check-performed, action-taken, decision-justified
dev-hawk comment-posted, insights-provided
delight output-produced, ux-issues-identified, actionable-findings
cloclo output-produced, summary-comment-added
pr-sous-chef comment-added, nudge-targeted, pr-evaluated
pr-triage-agent labels-applied, triage-data-set, report-produced
research discussion-created, topic-addressed, findings-substantive
gpclean issue-or-noop, go-mod-analyzed, decision-explained
daily-assign-issue-to-user assignment-made, contributor-selected, comment-posted
discussion-task-miner output-produced, tasks-extracted, labels-applied
go-logger pr-created, logging-changes-made, validation-run
poem-bot issue-created, poem-on-theme, poem-present
constraint-solving-potd discussion-created, problem-stated, solution-provided
copilot-cli-deep-research issue-created, optimization-opportunities, evidence-based
daily-skill-optimizer issue-created, three-improvements, improvements-actionable
docs-noob-tester findings-reported, confusing-steps-identified, user-perspective-maintained
release release-updated, highlights-included, community-attribution
repo-audit-analyzer audit-findings-produced, actionable-recommendations, evidence-cited

Design notes

  • Each question covers exactly one observable property answerable YES/NO from agent_output.json alone — no external tool calls
  • All workflows already declare safe-outputs:, satisfying the prerequisite for agent_output.json artifact inclusion
  • Questions use the model: small default; no frontier model overrides needed (all factual checks)
  • All 265 workflows recompiled with zero errors

Adds BinEval binary evaluation questions to 20 workflows that previously
had no evals. Each workflow gets 2-3 observable YES/NO questions aligned
with its safe-output actions and success criteria.

Workflows updated:
- ai-moderator
- deployment-incident-monitor
- design-decision-gate
- dev-hawk
- delight
- cloclo
- pr-sous-chef
- pr-triage-agent
- research
- gpclean
- daily-assign-issue-to-user
- discussion-task-miner
- go-logger
- poem-bot
- constraint-solving-potd
- copilot-cli-deep-research
- daily-skill-optimizer
- docs-noob-tester
- release
- repo-audit-analyzer

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>
Copilot AI requested a review from pelikhan July 28, 2026 21:34
@pelikhan
pelikhan marked this pull request as ready for review July 28, 2026 21:45
Copilot AI review requested due to automatic review settings July 28, 2026 21:45
@pelikhan
pelikhan merged commit be3ac2d into main Jul 28, 2026
1 check passed
@pelikhan
pelikhan deleted the copilot/add-evals-to-20-workflows branch July 28, 2026 21:46

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Adds BinEval quality checks to 20 agentic workflows and regenerates affected compiled workflows.

Changes:

  • Adds workflow-specific binary evaluation questions.
  • Integrates eval execution, artifact collection, and state persistence.
  • Regenerates affected lock workflows.
Show a summary per file
File Description
.github/workflows/research.md Adds research-result evaluations.
.github/workflows/research.lock.yml Compiles research eval jobs.
.github/workflows/repo-audit-analyzer.md Adds audit-quality evaluations.
.github/workflows/repo-audit-analyzer.lock.yml Compiles audit eval jobs.
.github/workflows/release.md Adds release-highlight evaluations.
.github/workflows/pr-triage-agent.md Adds PR-triage evaluations.
.github/workflows/pr-triage-agent.lock.yml Compiles triage eval jobs.
.github/workflows/pr-sous-chef.md Adds PR-nudge evaluations.
.github/workflows/pr-sous-chef.lock.yml Compiles sous-chef eval jobs.
.github/workflows/poem-bot.md Adds poem-output evaluations.
.github/workflows/poem-bot.lock.yml Compiles poem eval jobs.
.github/workflows/gpclean.md Adds GPL-analysis evaluations.
.github/workflows/gpclean.lock.yml Compiles GPL eval jobs.
.github/workflows/go-logger.md Adds logging-change evaluations.
.github/workflows/docs-noob-tester.md Adds documentation-testing evaluations.
.github/workflows/docs-noob-tester.lock.yml Compiles documentation eval jobs.
.github/workflows/discussion-task-miner.md Adds task-mining evaluations.
.github/workflows/dev-hawk.md Adds PR/CI analysis evaluations.
.github/workflows/design-decision-gate.md Adds ADR-gate evaluations.
.github/workflows/deployment-incident-monitor.md Adds incident-response evaluations.
.github/workflows/delight.md Adds UX-finding evaluations.
.github/workflows/daily-skill-optimizer.md Adds skill-improvement evaluations.
.github/workflows/daily-skill-optimizer.lock.yml Compiles skill eval jobs.
.github/workflows/daily-assign-issue-to-user.md Adds assignment evaluations.
.github/workflows/copilot-cli-deep-research.md Adds research-quality evaluations.
.github/workflows/constraint-solving-potd.md Adds problem-quality evaluations.
.github/workflows/cloclo.md Adds output and summary evaluations.
.github/workflows/ai-moderator.md Adds moderation evaluations.

Review details

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

  • Files reviewed: 40/40 changed files
  • Comments generated: 10
  • Review effort level: Medium

run-success: "✅ [{workflow_name}]({run_url}) finished PR sous-chef nudges."
run-failure: "⚠️ [{workflow_name}]({run_url}) {status} while preparing PRs."
timeout-minutes: 25
evals:
Comment on lines +298 to +302
question: Did the agent add a comment to at least one pull request?
- id: nudge-targeted
question: Does the agent output show a specific reason why the selected PR needs a nudge toward maintainer investigation?
- id: pr-evaluated
question: Does the agent output confirm that it evaluated at least one open PR for nudge eligibility?
sandbox:
agent:
sudo: false
evals:
diff_available: $diff_available
}' > /tmp/gh-aw/agent/adr-prefetch-summary.json

evals:
sandbox:
agent:
sudo: false
evals:
Comment on lines +38 to +42
question: Did the agent assign an issue to a user?
- id: contributor-selected
question: Does the agent output show that a relevant contributor was selected from recent merged PRs?
- id: comment-posted
question: Did the agent post a comment explaining the assignment decision?
Comment on lines +51 to +53
question: Did the agent post a comment with PR or CI analysis?
- id: insights-provided
question: Does the agent output include specific insights about pull request status or CI results?
Comment on lines +92 to +96
question: Did the agent create an issue or post a message with UX findings?
- id: ux-issues-identified
question: Does the agent output identify at least one specific UX improvement opportunity?
- id: actionable-findings
question: Does the agent output include actionable recommendations rather than only generic observations?
- id: highlights-included
question: Does the agent output confirm that release highlights were generated and prepended to the release notes?
- id: community-attribution
question: Does the agent output include attribution to community contributors in the release highlights?

evals:
- id: labels-applied
question: Did the agent apply triage labels to at least one pull request?
@github-actions

Copy link
Copy Markdown
Contributor

🎉 This pull request is included in a new release.

Release: v0.83.5

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants