Remove unactionable info from reviewer agent

[JanKrivanek]

Context

@jankratochvilcz was pointing two friction points with pr-reviewer:

• Non actionable comments - e.g. Migrate FormatUrl to multithreaded execution model #13573 (comment)
• The overviw table containing numerous 'LGTM' rows - which reduces 'signal-to-noise' ratio

Changes

Guiding to reduce nonactionable outputs

cc @jankratochvilcz

[github-actions]

🔍 Skill Validator Results

⚠️ Warnings or advisories found

Scope	Checked
Skills	13
Agents	1
Total	14
Severity	Count
---	---:
❌ Errors	0
⚠️ Warnings	6
ℹ️ Advisories	0

Summary

Level	Finding
ℹ️	Found 13 skill(s)
ℹ️	[assessing-breaking-changes] 📊 assessing-breaking-changes: 992 BPE tokens [chars/4: 1,154] (detailed ✓), 8 sections, 2 code blocks
ℹ️	[merge-dependency-updates] 📊 merge-dependency-updates: 2,030 BPE tokens [chars/4: 1,887] (detailed ✓), 16 sections, 7 code blocks
ℹ️	[reviewing-msbuild-code] 📊 reviewing-msbuild-code: 166 BPE tokens [chars/4: 216] (compact ✓), 1 sections, 0 code blocks
ℹ️	[reviewing-msbuild-code] ⚠ Skill is only 166 BPE tokens (chars/4 estimate: 216) — may be too sparse to provide actionable guidance.
ℹ️	[reviewing-msbuild-code] ⚠ No code blocks — agents perform better with concrete snippets and commands.
ℹ️	[reviewing-msbuild-code] ⚠ No numbered workflow steps — agents follow sequenced procedures more reliably.
ℹ️	[maintaining-binary-log-compatibility] 📊 maintaining-binary-log-compatibility: 1,271 BPE tokens [chars/4: 1,509] (detailed ✓), 12 sections, 2 code blocks
ℹ️	[release] 📊 release: 1,901 BPE tokens [chars/4: 1,889] (detailed ✓), 11 sections, 0 code blocks
ℹ️	[release] ⚠ No code blocks — agents perform better with concrete snippets and commands.

Full validator output

```text Found 13 skill(s) [assessing-breaking-changes] 📊 assessing-breaking-changes: 992 BPE tokens [chars/4: 1,154] (detailed ✓), 8 sections, 2 code blocks [merge-dependency-updates] 📊 merge-dependency-updates: 2,030 BPE tokens [chars/4: 1,887] (detailed ✓), 16 sections, 7 code blocks [reviewing-msbuild-code] 📊 reviewing-msbuild-code: 166 BPE tokens [chars/4: 216] (compact ✓), 1 sections, 0 code blocks [reviewing-msbuild-code] ⚠ Skill is only 166 BPE tokens (chars/4 estimate: 216) — may be too sparse to provide actionable guidance. [reviewing-msbuild-code] ⚠ No code blocks — agents perform better with concrete snippets and commands. [reviewing-msbuild-code] ⚠ No numbered workflow steps — agents follow sequenced procedures more reliably. [maintaining-binary-log-compatibility] 📊 maintaining-binary-log-compatibility: 1,271 BPE tokens [chars/4: 1,509] (detailed ✓), 12 sections, 2 code blocks [release] 📊 release: 1,901 BPE tokens [chars/4: 1,889] (detailed ✓), 11 sections, 0 code blocks [release] ⚠ No code blocks — agents perform better with concrete snippets and commands. [multithreaded-task-migration] 📊 multithreaded-task-migration: 3,319 BPE tokens [chars/4: 3,600] (standard ~), 29 sections, 9 code blocks [multithreaded-task-migration] ⚠ Skill is 3,319 BPE tokens (chars/4 estimate: 3,600) — approaching "comprehensive" range where gains diminish. [use-bootstrap-msbuild] 📊 use-bootstrap-msbuild: 709 BPE tokens [chars/4: 751] (detailed ✓), 14 sections, 5 code blocks [pipelines-health-check] 📊 pipelines-health-check: 5,025 BPE tokens [chars/4: 5,098] (comprehensive ✗), 40 sections, 10 code blocks [pipelines-health-check] ⚠ Skill is 5,025 BPE tokens (chars/4 estimate: 5,098) — "comprehensive" skills hurt performance by 2.9pp on average. Consider splitting into 2–3 focused skills. [authoring-errors-and-warnings] 📊 authoring-errors-and-warnings: 1,287 BPE tokens [chars/4: 1,393] (detailed ✓), 13 sections, 4 code blocks [changewaves] 📊 changewaves: 901 BPE tokens [chars/4: 1,007] (detailed ✓), 9 sections, 1 code blocks [optimizing-msbuild-performance] 📊 optimizing-msbuild-performance: 1,182 BPE tokens [chars/4: 1,342] (detailed ✓), 10 sections, 1 code blocks [integrating-sdk-and-msbuild] 📊 integrating-sdk-and-msbuild: 1,250 BPE tokens [chars/4: 1,440] (detailed ✓), 14 sections, 2 code blocks [deploy-msbuild-to-vs] 📊 deploy-msbuild-to-vs: 2,312 BPE tokens [chars/4: 2,363] (detailed ✓), 24 sections, 12 code blocks ✅ All checks passed (13 skill(s)) Found 1 agent(s) Validated 1 agent(s)

✅ All checks passed (1 agent(s))

</details>

[Copilot]

Pull request overview

Updates the expert-reviewer agent guidance to reduce non-actionable review output and improve signal-to-noise in the final summary.

Changes:

• Adds explicit instruction that inline comments must be actionable (no praise/“looks good” inline notes).
• Changes the summary guidance to omit LGTM dimensions from the summary table and instead report a single “X/Y dimensions clean” line.

Show a summary per file

File	Description
.github/agents/expert-reviewer.agent.md	Tightens reviewer-agent guidance to avoid non-actionable inline comments and reduce LGTM noise in summary output.

Copilot's findings

• Files reviewed: 1/1 changed files
• Comments generated: 1

[github-actions]

Expert Review — PR #13578

#	Dimension	Verdict
22	Correctness & Edge Cases	🟡 1 MODERATE, 1 NIT

✅ 23/24 dimensions clean.

• Correctness — contradictory wording between "record it as LGTM in the summary table" (line 649) and "Omit all LGTM dimensions from the table" (line 657); all-clear case not covered by example

---

Overall: Good change that meaningfully reduces review noise. The core idea — actionable-only inline comments and findings-only summary tables — is sound. The one real issue is the self-contradictory instruction at line 649 which references the old "record in table" behavior while the new line 657 says to omit from the table. A quick wording fix resolves it.

Note

🔒 Integrity filter blocked 1 item

The following item were blocked because they don't meet the GitHub integrity level.

• #13578 pull_request_read: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".

To allow these resources, lower min-integrity in your GitHub frontmatter:

tools:
  github:
    min-integrity: approved  # merged | approved | unapproved | none

Generated by Expert Code Review (on open) for issue #13578 · ● 2.3M

[jankratochvilcz]

Let one nit, otherwise looks good, thanks for quick action!