From f91358b0fbe836c7501e678590e1e5068436c12b Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Wed, 19 Aug 2026 04:59:07 +0000
Subject: [PATCH] improve(27): add agent-first path for evals step

Restructure step 27 to follow the Copilot-CLI-first guideline:
- Primary path: /agentic-workflows skill prompt to add evals block
- Manual terminal steps moved into a <details> collapse
- Added commit+push instructions before triggering a run
- Added iterative refinement example with agent prompt
- Removed stray TIP callout (folded into inline guidance)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---
 workshop/27-evaluate-workflow-quality.md | 41 +++++++++++++++++++-----
 1 file changed, 33 insertions(+), 8 deletions(-)

diff --git a/workshop/27-evaluate-workflow-quality.md b/workshop/27-evaluate-workflow-quality.md
index 4fe5ffdb..b774574e 100644
--- a/workshop/27-evaluate-workflow-quality.md
+++ b/workshop/27-evaluate-workflow-quality.md
@@ -28,9 +28,20 @@ You'll add an `evals:` block to your workflow, define binary quality questions,
 
 ## Steps
 
-### Add an `evals:` block
+### Add an `evals:` block with the skill
 
-Open `.github/workflows/daily-status.md` and add binary questions to frontmatter.
+In your Copilot CLI session in the terminal, paste:
+
+```prompt
+/agentic-workflows add three binary eval questions to daily-status.md: one checking that a status issue was created, one checking the output includes a summary of repository activity, and one checking that no writes happened outside declared safe outputs.
+```
+
+The skill adds the `evals:` block to your frontmatter, compiles the lock file, and shows you the diff.
+
+<details>
+<summary>:desktop_computer: Terminal path — add the evals block directly</summary>
+
+Open `.github/workflows/daily-status.md` and add binary questions to frontmatter:
 
 ```markdown .github/workflows/daily-status.md
 ---
@@ -48,17 +59,25 @@ evals:
 ---
 ```
 
-Each question should test one observable claim and be answerable from agent output alone.
+Each question should test one observable claim and be answerable from agent output alone. Compile after editing:
+
+```bash
+gh aw compile daily-status
+```
+
+</details>
 
-### Compile and run
+### Commit and trigger a run
 
-Compile after editing:
+Commit both the workflow source and the recompiled lock file, then trigger a run from the Actions UI:
 
 ```bash
-gh aw compile daily-status
+git add .
+git commit -m "feat: add evals to daily-status workflow"
+git push
 ```
 
-Trigger a run from the Actions UI (or use `gh aw run daily-status` if you prefer CLI).
+Go to **Actions → Daily Status Report → Run workflow** and click **Run workflow**.
 
 ### Inspect evaluation results
 
@@ -78,8 +97,14 @@ Example record:
 
 When you update your prompt or tools, rerun the workflow and compare answers across runs. A question that flips from `YES` to `NO` is a fast signal that quality regressed and needs investigation.
 
+To refine your questions, return to Copilot CLI and describe what you observed:
+
+```prompt
+/agentic-workflows the eval question "includes_summary" is too broad — update it to check that the output includes at least one open issue or pull request from the last 24 hours.
+```
+
 > [!TIP]
-> Prefer using an agent with `/agentic-workflows` to add or refine `evals:` questions, then run `gh aw compile --watch` while iterating.
+> Use `gh aw compile --watch` while iterating on eval questions to get instant feedback on compile errors.
 
 ## ✅ Checkpoint