Deep Review UX improvements and flow-chat terminal card grouping

src/crates/core/src/agentic/agents/prompts/deep_review_agent.md

@@ -142,7 +142,7 @@ After the reviewer batch finishes, launch `ReviewJudge` with:
 - the same review target
 - the full reviewer outputs from every reviewer that ran, including timeout/cancel/failure notes
 - if file splitting was used, include outputs from **all** same-role instances and label each by group (e.g. "Security Reviewer [group 1/3]")
-- an instruction to validate, reject, merge, or downgrade findings, and to deduplicate any overlapping findings from same-role instances
+- an instruction to validate, reject, merge, or downgrade findings from a **third-party perspective** — the judge primarily examines reviewer reports for logical consistency and evidence quality, and only uses code inspection tools for targeted spot-checks when a specific claim needs verification
 
 If the execution policy says `judge_timeout_seconds > 0`, pass `timeout_seconds` with that value to the judge Task call.
 
@@ -152,6 +152,7 @@ The judge must explicitly call out:
 
 - likely false positives
 - optimization advice that is too risky or directionally wrong
+- findings where the reviewer's evidence does not support their conclusion
 - which findings should survive into the final report
 
 ### Phase 4: Report and wait for user approval

src/crates/core/src/agentic/agents/prompts/review_quality_gate_agent.md

@@ -2,7 +2,7 @@ You are the **Review Quality Inspector** for BitFun deep reviews.
 
 {LANGUAGE_PREFERENCE}
 
-You are not another broad reviewer. Your job is to validate the outputs from the specialist reviewers and prevent false positives, low-signal nitpicks, or directionally-wrong optimization advice from reaching the final report.
+Your primary role is an independent third-party arbiter that validates the **reports submitted by other reviewers**. You do not perform a broad independent code review from scratch. Instead, you examine each reviewer's findings from a logical and evidentiary standpoint, and use code inspection tools **only when necessary** to verify specific claims made by reviewers.
 
 ## Inputs
 
@@ -18,20 +18,24 @@ You will receive:
 For every candidate finding from the reviewers:
 
 1. decide whether it is **validated**, **downgraded**, or **rejected**
-2. verify it against the code/diff when needed
-3. check whether the suggested fix is actually safe and directionally correct
-4. if multiple same-role instances reported overlapping or duplicate findings, **merge them into a single finding** with the strongest severity and evidence
+2. evaluate the **internal consistency** of the reviewer's reasoning — does the evidence they cited actually support their conclusion?
+3. when a finding's validity is unclear from the reviewer's report alone, use read-only tools to **spot-check the specific code location** the reviewer referenced
+4. check whether the suggested fix direction is **logically sound** and **safe in principle**
+5. if multiple same-role instances reported overlapping or duplicate findings, **merge them into a single finding** with the strongest severity and evidence
+
+**Important**: Your code inspection should be targeted and minimal. Do not broadly re-review the codebase. Only inspect specific lines or files when a reviewer's claim needs verification or when you suspect a false positive / false negative.
 
 Be especially skeptical of:
 
 - speculative bugs with no evidence
 - "optimize this" advice without meaningful impact
 - recommendations that would widen scope or add risk without strong payoff
 - duplicated findings reported by multiple reviewers or multiple same-role instances
+- findings where the stated evidence does not logically lead to the stated conclusion
 
 ## Tools
 
-Use only read-only investigation:
+Use read-only investigation when needed:
 
 - `GetFileDiff`
 - `Read`

src/crates/core/src/agentic/agents/review_specialist_agents.rs

@@ -35,7 +35,7 @@ define_readonly_subagent!(
     ReviewJudgeAgent,
     REVIEW_JUDGE_AGENT_TYPE,
     "Review Quality Inspector",
-    r#"Independent read-only quality inspector that validates reviewer findings, removes false positives, checks whether optimization advice is directionally correct, and decides what should appear in the final deep-review report."#,
+    r#"Independent third-party arbiter that validates reviewer reports for logical consistency and evidence quality. It spot-checks specific code locations only when a claim needs verification, rather than re-reviewing the codebase from scratch."#,
     "review_quality_gate_agent",
     &["Read", "Grep", "Glob", "LS", "GetFileDiff", "Git"]
 );

src/web-ui/src/app/scenes/agents/components/ReviewTeamPage.scss

@@ -38,7 +38,7 @@
     width: 100%;
     padding: 16px;
     border: 0;
-    border-radius: 0;
+    border-radius: inherit;
     background: transparent;
     color: var(--color-text-primary);
     text-align: left;
@@ -79,7 +79,7 @@
 
   &__policy-metrics {
     display: grid;
-    grid-template-columns: repeat(5, minmax(86px, 1fr));
+    grid-template-columns: repeat(auto-fit, minmax(86px, 1fr));
     gap: 8px;
     min-width: 0;
   }
@@ -113,6 +113,14 @@
     }
   }
 
+  /* Sections that render their own cards should not show the default body frame. */
+  &__section--no-body-frame .bitfun-config-page-section__body {
+    background: transparent;
+    border: 0;
+    border-radius: 0;
+    overflow: visible;
+  }
+
   &__policy-action {
     display: inline-flex;
     align-items: center;
@@ -283,16 +291,12 @@
 
   &__member-card {
     display: flex;
-    align-items: flex-start;
-    gap: $size-gap-3;
+    flex-direction: column;
     width: 100%;
-    min-height: 148px;
-    padding: 14px;
     border-radius: 8px;
     border: 1px solid var(--border-subtle);
     background: color-mix(in srgb, var(--element-bg-soft) 78%, transparent);
     text-align: left;
-    cursor: pointer;
     transition:
       transform $motion-fast $easing-standard,
       border-color $motion-fast $easing-standard,
@@ -310,6 +314,30 @@
         0 0 0 1px color-mix(in srgb, var(--member-accent, #64748b) 30%, transparent),
         0 12px 24px color-mix(in srgb, var(--shadow-color, #0f172a) 10%, transparent);
     }
+
+    &.is-expanded {
+      grid-column: 1 / -1;
+    }
+  }
+
+  &__member-card-header {
+    display: flex;
+    align-items: flex-start;
+    gap: $size-gap-3;
+    width: 100%;
+    min-height: 148px;
+    padding: 14px;
+    border: 0;
+    border-radius: inherit;
+    background: transparent;
+    color: var(--color-text-primary);
+    text-align: left;
+    cursor: pointer;
+
+    &:focus-visible {
+      outline: none;
+      box-shadow: inset 0 0 0 2px color-mix(in srgb, var(--member-accent, #64748b) 50%, transparent);
+    }
   }
 
   &__member-card-icon {
@@ -385,11 +413,29 @@
     color: color-mix(in srgb, #f59e0b 82%, var(--color-text-muted));
   }
 
+  &__member-card-chevron {
+    display: inline-flex;
+    align-items: center;
+    align-self: flex-start;
+    color: var(--color-text-muted);
+    margin-top: 2px;
+  }
+
+  &__member-card-detail {
+    overflow: hidden;
+    animation: review-team-card-expand $motion-base $easing-standard forwards;
+    border-top: 1px solid var(--border-subtle);
+  }
+
+  &__member-card-detail-inner {
+    padding: 16px;
+  }
+
   &__detail-hero {
     display: flex;
     align-items: flex-start;
     gap: $size-gap-4;
-    padding: 16px;
+    padding: 0 0 16px;
     border-bottom: 1px solid var(--border-subtle);
   }
 
@@ -454,8 +500,7 @@
     display: flex;
     flex-direction: column;
     gap: 10px;
-    padding: 16px;
-    border-bottom: 1px solid var(--border-subtle);
+    padding: 16px 0 0;
   }
 
   &__block-label {
@@ -510,6 +555,17 @@
   }
 }
 
+@keyframes review-team-card-expand {
+  from {
+    opacity: 0;
+    max-height: 0;
+  }
+  to {
+    opacity: 1;
+    max-height: 800px;
+  }
+}
+
 @media (max-width: 960px) {
   .review-team-page {
     &__summary-grid,

src/web-ui/src/app/scenes/agents/components/ReviewTeamPage.test.tsx

@@ -299,7 +299,7 @@ describeWithJsdom('ReviewTeamPage', () => {
       memberButton!.dispatchEvent(new dom.window.MouseEvent('click', { bubbles: true }));
     });
 
-    expect(container.textContent).toContain('Member Detail');
+    expect(container.textContent).toContain('Responsibilities');
     expect(container.textContent).toContain('Logic');
   });
 });