Integrate code-review skill into pr-review Pre-Flight and Try-Fix

.github/pr-review/pr-preflight.md

@@ -1,10 +1,10 @@
-# PR Pre-Flight — Context Gathering
+# PR Pre-Flight — Context Gathering & Code Review
 
-> **SCOPE:** Document only. No code analysis. No fix opinions. No running tests.
+> **SCOPE:** Gather context, classify files, and perform deep code review. No code changes. No fix selection. No test execution.
 
 ---
 
-## Steps
+## Part A: Context Gathering (Steps 1–6)
 
 1. **Read the issue** — full body + ALL comments via GitHub MCP tools
 2. **Find the PR** — read description, diff summary, review comments, inline feedback
@@ -35,7 +35,58 @@ gh pr view XXXXX --json comments --jq '.comments[] | select(.body | contains("Fi
 
 ---
 
-## Output File
+## Part B: Code Review (Step 7)
+
+> **Purpose:** Perform deep code analysis using the `code-review` skill to surface correctness issues, safety concerns, and MAUI convention violations BEFORE Try-Fix explores alternatives. These findings guide Try-Fix models toward higher-quality fixes.
+
+> **🚨 Independence-first requirement:** Step 7 MUST be invoked as a **separate sub-agent** (via the `task` tool with `agent_type: "general-purpose"`) so the code-review skill can form its assessment from the code BEFORE reading any PR narrative. The sub-agent receives ONLY the PR number — not the context gathered in Part A. This prevents anchoring bias.
+>
+> **Validation constraint:** The Step 7 prompt MUST NOT contain issue titles, root-cause descriptions, bug summaries, or any Part A content — only `PR #XXXXX`. If you find yourself adding context "to help" the sub-agent, you are violating independence-first.
+
+7. **Invoke the code-review skill as a sub-agent:**
+
+   Use the `task` tool to launch a separate agent. The prompt MUST NOT contain issue titles, root-cause descriptions, or any Part A context — only the PR number.
+
+   ```python
+   task(
+     name="code-review",
+     description="Code review for PR",
+     agent_type="general-purpose",
+     mode="sync",
+     prompt="""
+       Run the code-review skill for PR #XXXXX.
+       Follow the full 6-step workflow in .github/skills/code-review/SKILL.md.
+       Output the review in the format specified by that skill.
+     """
+   )
+   ```
+
+   The sub-agent internally follows the code-review skill's 6-step workflow:
+   1. Gather code context (independence-first — reads code BEFORE PR description)
+   2. Load MAUI review rules from `.github/skills/code-review/references/review-rules.md`
+   3. Form independent assessment
+   4. Reconcile with PR narrative and prior reviews
+   5. Check CI status
+   6. Blast radius, failure-mode probing, and verdict
+
+**If Step 7 fails, times out, or returns malformed output:**
+- Write `pre-flight/code-review.md` with: `## Code Review: SKIPPED\n\nReason: {failure description}`
+- Set verdict to `SKIPPED` in the Code Review Summary section of `content.md`
+- Omit `hints` from Try-Fix prompts (the `hints` field becomes optional when code review is unavailable)
+- Do NOT apply the code-review hard gate in Phase 3 (Report) — treat as if code review was not run
+
+**Store the sub-agent's full output** in `pre-flight/code-review.md` — use the exact output format from the code-review skill (do NOT reformat or summarize into a different template).
+
+**Extract key items for Try-Fix consumption** and add to `content.md`:
+- All ❌ Error findings (with file:line references)
+- All ⚠️ Warning findings (with file:line references)
+- Failure-mode probes and their answers
+- Blast radius assessment summary
+- The overall verdict and confidence level
+
+---
+
+## Output Files
 
 ```bash
 mkdir -p CustomAgentLogsTmp/PRState/{PRNumber}/PRAgent/pre-flight
@@ -52,16 +103,29 @@ Write `content.md`:
 - {Finding 1}
 - {Finding 2}
 
+### Code Review Summary
+**Verdict:** {LGTM / NEEDS_CHANGES / NEEDS_DISCUSSION / SKIPPED}
+**Confidence:** {high / medium / low / N/A}
+**Errors:** {count} | **Warnings:** {count} | **Suggestions:** {count}
+
+Key code review findings:
+- {❌/⚠️/💡} {Brief finding with file:line reference}
+- ...
+*(If SKIPPED: "Code review sub-agent failed or timed out. Reason: {details}")*
+
 ### Fix Candidates
 | # | Source | Approach | Test Result | Files Changed | Notes |
 |---|--------|----------|-------------|---------------|-------|
 | PR | PR #XXXXX | {approach} | ⏳ PENDING (Gate) | `file.cs` | Original PR |
 ```
 
+Write `code-review.md` — the exact output from the code-review sub-agent, in the format specified by `.github/skills/code-review/SKILL.md` (Review Output Format section). Do NOT reformat or create a custom template — preserve the skill's native output verbatim.
+
 ---
 
 ## Common Mistakes
 
-- ❌ Researching root cause — save for Try-Fix phase
-- ❌ Looking at implementation code — just gather context
+- ❌ Skipping the code-review step — it provides critical findings for Try-Fix
+- ❌ Reading the PR description before code in Step 7 — independence-first prevents anchoring bias
 - ❌ Running tests — that's the Gate phase
+- ❌ Proposing fixes — save fix ideas for Try-Fix phase

.github/pr-review/pr-report.md

@@ -12,18 +12,26 @@
 
 - Phases 1-2 (Pre-Flight, Try-Fix) must be complete before starting
 - Gate result is available from the prompt (ran separately before this skill)
+- **Read `pre-flight/content.md`** to get the code-review summary (verdict, confidence, error/warning counts)
+- Optionally read `pre-flight/code-review.md` for full findings if needed for the recommendation
 
 ---
 
 ## Steps
 
-1. **Determine recommendation:**
+1. **Determine recommendation** (rows evaluated in order — first match wins):
 
-   | Condition | Recommendation |
-   |-----------|----------------|
-   | PR's fix selected and Gate passed | `✅ APPROVE` |
-   | Alternative fix found via Try-Fix | `⚠️ REQUEST CHANGES` — suggest alternative |
-   | Gate failed | `⚠️ REQUEST CHANGES` — fix doesn't work |
+   | Priority | Condition | Recommendation |
+   |----------|-----------|----------------|
+   | 1 | Code review verdict is `NEEDS_CHANGES` (any ❌ errors) | `⚠️ REQUEST CHANGES` — code review found errors |
+   | 2 | Gate failed (tests fail with fix) | `⚠️ REQUEST CHANGES` — fix doesn't work |
+   | 3 | Alternative fix found via Try-Fix that is simpler/better | `⚠️ REQUEST CHANGES` — suggest alternative |
+   | 4 | Code review verdict is `NEEDS_DISCUSSION` | `⚠️ REQUEST CHANGES` — include code review concerns |
+   | 5 | PR's fix selected AND Gate passed AND code review LGTM or SKIPPED | `✅ APPROVE` |
+
+   **🚨 Hard gate:** If the code review (from Pre-Flight) has verdict `NEEDS_CHANGES`, the final recommendation MUST be `REQUEST CHANGES` regardless of Gate or Try-Fix results. Code-review ❌ Errors cannot be overridden by passing tests alone.
+
+   **Code review SKIPPED:** If the code-review sub-agent failed or timed out (verdict = `SKIPPED`), the hard gate does NOT apply. Proceed as if code review was not available — base the recommendation on Gate and Try-Fix results only. Note in the report that code review was unavailable.
 
 2. **Write output files** — Save recommendation to `content.md`
 
@@ -47,18 +55,22 @@ Write `content.md`:
 | Phase | Status | Notes |
 |---|---|---|
 | Pre-Flight | ✅ COMPLETE | {notes} |
+| Code Review | {verdict} ({confidence}) | {error_count} errors, {warning_count} warnings |
 | Gate | ✅ PASSED | {platform} |
 | Try-Fix | ✅ COMPLETE | {N} attempts, {M} passing |
 | Report | ✅ COMPLETE | |
 
+### Code Review Impact on Try-Fix
+{Brief description of how code-review findings influenced try-fix exploration. Did any model specifically address a code review ❌ Error? Did failure-mode probes reveal issues that guided fix approaches?}
+
 ### Summary
 {Brief summary of the review}
 
 ### Root Cause
 {Root cause analysis}
 
 ### Fix Quality
-{Assessment of the fix}
+{Assessment of the fix — informed by both gate results and code review findings}
 ```
 
 ---

.github/scripts/Review-PR.ps1

@@ -499,8 +499,11 @@ The agent will analyze the issue, determine the appropriate test type (UI test,
     }
 }
 
-# Post gate result as a separate PR comment
-$postGateScript = Join-Path $PSScriptRoot "post-gate-comment.ps1"
+# Post gate result by updating (or creating) the unified AI Summary comment.
+# The same script is called again in STEP 3 once the review phases finish; here
+# it runs early so the PR author sees the gate outcome without waiting for the
+# full review.
+$postGateScript = Join-Path $PSScriptRoot "post-ai-summary-comment.ps1"
 if (Test-Path $postGateScript) {
     try {
         if ($DryRun) {
@@ -509,10 +512,10 @@ if (Test-Path $postGateScript) {
             & $postGateScript -PRNumber $PRNumber
         }
     } catch {
-        Write-Host "  ⚠️ Failed to post gate comment (non-fatal): $_" -ForegroundColor Yellow
+        Write-Host "  ⚠️ Failed to post gate section (non-fatal): $_" -ForegroundColor Yellow
     }
 } else {
-    Write-Host "  ⚠️ post-gate-comment.ps1 not found" -ForegroundColor Yellow
+    Write-Host "  ⚠️ post-ai-summary-comment.ps1 not found" -ForegroundColor Yellow
 }
 
 # Apply gate result label

.github/scripts/post-ai-summary-comment.ps1

@@ -13,9 +13,16 @@
     After posting, the PR author is @-mentioned so they know to review.
 
     Content is auto-loaded from PRAgent phase files:
+    CustomAgentLogsTmp/PRState/<PRNumber>/PRAgent/gate/content.md          (always shown first, open)
     CustomAgentLogsTmp/PRState/<PRNumber>/PRAgent/{pre-flight,try-fix,report}/content.md
+    CustomAgentLogsTmp/PRState/<PRNumber>/PRAgent/pre-flight/code-review.md
 
-    Gate is posted separately by post-gate-comment.ps1.
+    Gate is included as a section inside this unified comment — the script may
+    be called by Review-PR.ps1 twice per run: once after the gate completes
+    (gate-only update) and once after the review phases finish (full update).
+
+    Any standalone legacy "<!-- AI Gate -->" comment from older versions of
+    the script is deleted after a successful post to avoid duplicates.
 
 .PARAMETER PRNumber
     The pull request number (required)
@@ -60,9 +67,34 @@ if (-not (Test-Path $PRAgentDir)) {
 }
 
 $phases = [ordered]@{
-    "pre-flight" = @{ File = "pre-flight/content.md"; Icon = "🔍"; Title = "Pre-Flight — Context & Validation" }
-    "try-fix"    = @{ File = "try-fix/content.md";    Icon = "🔧"; Title = "Fix — Analysis & Comparison" }
-    "report"     = @{ File = "report/content.md";     Icon = "📋"; Title = "Report — Final Recommendation" }
+    "pre-flight"  = @{ File = "pre-flight/content.md";     Icon = "🔍"; Title = "Pre-Flight — Context & Validation" }
+    "code-review" = @{ File = "pre-flight/code-review.md"; Icon = "🔬"; Title = "Code Review — Deep Analysis" }
+    "try-fix"     = @{ File = "try-fix/content.md";        Icon = "🔧"; Title = "Fix — Analysis & Comparison" }
+    "report"      = @{ File = "report/content.md";         Icon = "📋"; Title = "Report — Final Recommendation" }
+}
+
+# ─── Gate content (rendered first, always open) ───
+$gateSection = $null
+$gateFilePath = Join-Path $PRAgentDir "gate/content.md"
+if (Test-Path $gateFilePath) {
+    $gateContent = Get-Content $gateFilePath -Raw -Encoding UTF8
+    if (-not [string]::IsNullOrWhiteSpace($gateContent)) {
+        Write-Host "  ✅ gate ($((Get-Item $gateFilePath).Length) bytes)" -ForegroundColor Green
+        $gateSection = @"
+<details open>
+<summary>🚦 <strong>Gate — Test Before & After Fix</strong></summary>
+
+---
+
+$gateContent
+
+</details>
+"@
+    } else {
+        Write-Host "  ⏭️  gate (empty)" -ForegroundColor Gray
+    }
+} else {
+    Write-Host "  ⏭️  gate (not found)" -ForegroundColor Gray
 }
 
 $phaseSections = @()
@@ -93,8 +125,8 @@ $content
     }
 }
 
-if ($phaseSections.Count -eq 0) {
-    throw "No phase content found. Ensure at least one content.md exists in $PRAgentDir."
+if (-not $gateSection -and $phaseSections.Count -eq 0) {
+    throw "No gate or phase content found. Ensure at least one of gate/content.md or {phase}/content.md exists in $PRAgentDir."
 }
 
 # ============================================================================
@@ -123,7 +155,13 @@ $timestamp = (Get-Date).ToUniversalTime().ToString("yyyy-MM-dd HH:mm UTC")
 # BUILD NEW SESSION BLOCK
 # ============================================================================
 
-$phaseContent = $phaseSections -join "`n`n---`n`n"
+# Combine gate (always first, open) with phases (collapsed). When only one
+# kind of content is available, the session still renders cleanly.
+$sessionParts = @()
+if ($gateSection)            { $sessionParts += $gateSection }
+if ($phaseSections.Count -gt 0) { $sessionParts += ($phaseSections -join "`n`n---`n`n") }
+$phaseContent = $sessionParts -join "`n`n---`n`n"
+
 $sessionMarkerStart = "<!-- SESSION:$commitSha7 START -->"
 $sessionMarkerEnd = "<!-- SESSION:$commitSha7 END -->"
 
@@ -176,12 +214,17 @@ function Merge-Sessions {
     foreach ($sha in $orderedKeys) {
         $block = $sessions[$sha]
         if ($isFirst) {
-            # Ensure latest session has <details open>
-            $block = $block -replace '<details(?:\s+open)?>', '<details open>'
+            # Ensure ONLY the outer (session-wrapping) details tag is open. Inner
+            # phase tags must keep their original open/collapsed state — we used
+            # to re-open all of them via a global regex replace, which forced
+            # every phase to expand on each new session.
+            $rx = [regex]::new('<details(?:\s+open)?>')
+            $block = $rx.Replace($block, '<details open>', 1)
             $isFirst = $false
         } else {
-            # Collapse older sessions
-            $block = $block -replace '<details\s+open>', '<details>'
+            # Collapse the outer details of older sessions; leave inner phases alone.
+            $rx = [regex]::new('<details\s+open>')
+            $block = $rx.Replace($block, '<details>', 1)
         }
         $allSessions += $block
     }
@@ -300,3 +343,25 @@ try {
 } finally {
     Remove-Item $tempFile -ErrorAction SilentlyContinue
 }
+
+# ============================================================================
+# CLEAN UP LEGACY STANDALONE GATE COMMENTS
+# ============================================================================
+# Earlier versions of this workflow posted gate results in a separate comment
+# marked with <!-- AI Gate -->. Now that the gate is included as a section in
+# this unified comment, those legacy comments are duplicates and should go.
+
+try {
+    $legacyMarker = "<!-- AI Gate -->"
+    $allRaw = gh api "repos/dotnet/maui/issues/$PRNumber/comments" --paginate 2>$null
+    if ($allRaw) {
+        $allComments = $allRaw | ConvertFrom-Json
+        $legacy = @($allComments | Where-Object { $_.body -and $_.body.Contains($legacyMarker) })
+        foreach ($lc in $legacy) {
+            Write-Host "🧹 Deleting legacy gate comment (ID: $($lc.id))..." -ForegroundColor Gray
+            gh api --method DELETE "repos/dotnet/maui/issues/comments/$($lc.id)" 2>&1 | Out-Null
+        }
+    }
+} catch {
+    Write-Host "⚠️ Legacy gate-comment cleanup failed (non-fatal): $_" -ForegroundColor Yellow
+}