@in-the-loop-labs/pair-review 3.0.5 → 3.1.0

package/src/ai/prompts/baseline/consolidation/balanced.js

@@ -22,6 +22,7 @@
  * - {{reviewIntro}} - Review introduction line
  * - {{lineNumberGuidance}} - Line number guidance section
  * - {{customInstructions}} - Custom instructions section (optional)
+ * - {{dedupInstructions}} - Dedup instructions section (optional)
  * - {{reviewerSuggestions}} - Formatted reviewer suggestions input
  * - {{suggestionCount}} - Total number of input suggestions
  * - {{reviewerCount}} - Number of reviewers being consolidated
@@ -51,6 +52,23 @@ Multiple independent AI reviewers have analyzed the same code changes. Your job
 {{customInstructions}}
 </section>
 
+<section name="dedup-instructions" optional="true">
+{{dedupInstructions}}
+</section>
+
+<section name="reviewer-context-guidance" required="true">
+### Reviewer Context Awareness
+Each reviewer below may have been configured with custom instructions. These fall into two categories:
+
+- **Domain-focused reviewers**: Instructions that specify a code review focus area (e.g., "focus on security", "review error handling", "check performance"). Their findings *within that focus area* carry higher weight than generalist reviewers.
+- **General reviewers**: Either no custom instructions, or instructions about methodology/style/persona (e.g., "be thorough", "use a friendly tone"). Treat their suggestions at face value across all categories.
+
+**Weighting rules:**
+- Only boost a reviewer's findings when their instructions indicate domain expertise relevant to the finding's category
+- Cross-specialty findings from a domain-focused reviewer should be treated as general findings
+- In conflicts within a domain, prefer the domain-focused reviewer's analysis over a generalist's
+</section>
+
 <section name="input-suggestions" locked="true">
 ## Input: {{reviewerCount}} Reviewer(s), {{suggestionCount}} Total Suggestions
 
@@ -78,15 +96,39 @@ Multiple independent AI reviewers have analyzed the same code changes. Your job
 ### 4. Quality Filter
 - Drop suggestions with very low confidence (< 0.3) unless multiple reviewers agree
 - Boost confidence when multiple reviewers independently identify the same issue
+
+### 5. Severity Assessment
+Assess severity based on the evidence and reasoning across all reviewers. When reviewers assign different severities, apply the same conflict resolution principles above. When truly uncertain, preserve the highest severity. Omit severity for praise items.
+
+**Severity Definitions:**
+- **critical**: Production incidents, system failures, or security vulnerabilities — runtime crashes, data corruption or loss, race conditions, deadlocks, breaking changes, changes that will cause existing tests to fail
+- **medium**: Degraded functionality or reliability — missing error handling, N+1 queries, missing validation, missing or poor test coverage for new functionality
+- **minor**: Code quality concerns — documentation gaps, minor optimizations, style inconsistencies
 </section>
 
 <section name="consensus-handling" required="true">
-### 5. Consensus Handling
+### 6. Consensus Handling
 - **Agreement**: When multiple reviewers flag the same issue, increase confidence by 0.1-0.2 (cap at 1.0)
 - **Partial overlap**: Merge related but distinct observations into a richer suggestion
 - **Contradiction**: Use your judgment; prefer the more actionable analysis
 </section>
 
+<section name="balanced-output" required="true">
+### 7. Balanced Output
+- **Deduplicate, don't concatenate**: If two reviewers flagged the same issue, merge them into one suggestion. Distinct findings should each be preserved — the goal is to eliminate redundancy, not to reduce count.
+- **Limit praise** to 2–3 most noteworthy items across all reviewers
+- **Avoid suggestion overload** — aim for quality over quantity
+- **Include confidence scores** reflecting cross-reviewer agreement
+</section>
+
+<section name="summary-synthesis" required="true">
+## Summary Synthesis
+The summary field should synthesize the findings, not list them.
+- Synthesize the key findings into a single cohesive paragraph
+- **Draw on reviewer summaries**: Use these as evidence for your own synthesis — integrate their insights into a cohesive conclusion rather than listing them individually
+- Write as if a single reviewer produced this analysis — do not mention consolidation, merging, or multiple reviewers
+</section>
+
 <section name="output-schema" locked="true">
 ## Output Format
 
@@ -100,6 +142,7 @@ Output JSON with this structure:
       "line": 42,
       "old_or_new": "NEW",
       "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+      "severity": "critical|medium|minor (omit for praise)",
       "title": "Brief title",
       "description": "Detailed explanation",
       "suggestion": "How to fix/improve (omit for praise)",
@@ -110,13 +153,14 @@ Output JSON with this structure:
   "fileLevelSuggestions": [{
     "file": "path/to/file",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title describing file-level concern",
     "description": "Explanation of the file-level observation",
     "suggestion": "How to address the file-level concern (omit for praise items)",
     "confidence": 0.0-1.0,
     "reasoning": ["Step-by-step reasoning explaining why this issue was flagged (optional)"]
   }],
-  "summary": "Brief consolidation summary. Write as if a single reviewer produced this analysis — do NOT mention 'consolidation', 'merging', or 'multiple reviewers'."
+  "summary": "Single cohesive paragraph summarizing key findings. Write as a single reviewer."
 }
 
 ### GitHub Suggestion Syntax
@@ -155,9 +199,13 @@ const sections = [
   { name: 'critical-output', locked: true },
   { name: 'role-description', required: true },
   { name: 'custom-instructions', optional: true },
+  { name: 'dedup-instructions', optional: true },
+  { name: 'reviewer-context-guidance', required: true },
   { name: 'input-suggestions', locked: true },
   { name: 'consolidation-rules', required: true },
   { name: 'consensus-handling', required: true },
+  { name: 'balanced-output', required: true },
+  { name: 'summary-synthesis', required: true },
   { name: 'output-schema', locked: true },
   { name: 'diff-instructions', required: true },
   { name: 'guidelines', required: true }
@@ -173,9 +221,13 @@ const defaultOrder = [
   'critical-output',
   'role-description',
   'custom-instructions',
+  'dedup-instructions',
+  'reviewer-context-guidance',
   'input-suggestions',
   'consolidation-rules',
   'consensus-handling',
+  'balanced-output',
+  'summary-synthesis',
   'output-schema',
   'diff-instructions',
   'guidelines'

package/src/ai/prompts/baseline/consolidation/fast.js

@@ -23,6 +23,7 @@
  * - {{reviewIntro}} - Review introduction line
  * - {{lineNumberGuidance}} - Line number guidance section
  * - {{customInstructions}} - Custom instructions section (optional)
+ * - {{dedupInstructions}} - Dedup instructions section (optional)
  * - {{reviewerSuggestions}} - Formatted reviewer suggestions input
  * - {{suggestionCount}} - Total number of input suggestions
  * - {{reviewerCount}} - Number of reviewers being consolidated
@@ -52,6 +53,15 @@ Merge suggestions from multiple AI reviewers. Deduplicate. Resolve conflicts. Ke
 {{customInstructions}}
 </section>
 
+<section name="dedup-instructions" optional="true">
+{{dedupInstructions}}
+</section>
+
+<section name="reviewer-context-guidance" required="true" tier="fast">
+### Reviewer Context
+Reviewers may have custom instructions below. Only boost findings when instructions indicate domain expertise relevant to the finding (e.g., "focus on security" boosts security findings). Persona/methodology instructions (e.g., "be thorough") don't confer specialist weight. In domain conflicts, prefer the domain-focused reviewer.
+</section>
+
 <section name="input-suggestions" locked="true">
 ## Input: {{reviewerCount}} Reviewer(s), {{suggestionCount}} Total Suggestions
 
@@ -64,6 +74,16 @@ Merge suggestions from multiple AI reviewers. Deduplicate. Resolve conflicts. Ke
 - When reviewers disagree, keep the more specific analysis.
 - Preserve unique insights from individual reviewers.
 - Drop very low confidence (< 0.3) items unless multiple reviewers agree.
+- Assess severity using evidence across reviewers; preserve highest when uncertain. Omit for praise.
+- Severity: critical (crashes, security, data loss, race conditions, breaking changes, test failures), medium (degraded functionality, missing error handling/validation/test coverage), minor (code quality, docs, optimizations).
+</section>
+
+<section name="balanced-output" required="true" tier="fast">
+Deduplicate, don't concatenate: merge duplicate findings, preserve distinct ones. Max 2 praise items.
+</section>
+
+<section name="summary-synthesis" required="true" tier="fast">
+Draw on reviewer summaries as evidence for your own synthesis. Write one cohesive paragraph — no mention of consolidation/merging.
 </section>
 
 <section name="output-schema" locked="true">
@@ -74,6 +94,7 @@ Merge suggestions from multiple AI reviewers. Deduplicate. Resolve conflicts. Ke
     "line": 42,
     "old_or_new": "NEW",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title",
     "description": "Detailed explanation",
     "suggestion": "How to fix/improve (omit for praise)",
@@ -82,12 +103,13 @@ Merge suggestions from multiple AI reviewers. Deduplicate. Resolve conflicts. Ke
   "fileLevelSuggestions": [{
     "file": "path/to/file",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title describing file-level concern",
     "description": "Explanation of the file-level observation",
     "suggestion": "How to address the file-level concern (omit for praise items)",
     "confidence": 0.0-1.0
   }],
-  "summary": "Key findings as if from single reviewer (no mention of consolidation/merging)"
+  "summary": "Single cohesive paragraph of key findings. Write as single reviewer."
 }
 
 ### GitHub Suggestion Syntax
@@ -121,8 +143,12 @@ const sections = [
   { name: 'critical-output', locked: true },
   { name: 'role-description', required: true, tier: ['fast'] },
   { name: 'custom-instructions', optional: true, tier: ['fast', 'balanced', 'thorough'] },
+  { name: 'dedup-instructions', optional: true },
+  { name: 'reviewer-context-guidance', required: true, tier: ['fast'] },
   { name: 'input-suggestions', locked: true },
   { name: 'consolidation-rules', required: true, tier: ['fast'] },
+  { name: 'balanced-output', required: true, tier: ['fast'] },
+  { name: 'summary-synthesis', required: true, tier: ['fast'] },
   { name: 'output-schema', locked: true },
   { name: 'diff-instructions', required: true, tier: ['fast'] },
   { name: 'guidelines', required: true, tier: ['fast'] }
@@ -139,8 +165,12 @@ const defaultOrder = [
   'critical-output',
   'role-description',
   'custom-instructions',
+  'dedup-instructions',
+  'reviewer-context-guidance',
   'input-suggestions',
   'consolidation-rules',
+  'balanced-output',
+  'summary-synthesis',
   'output-schema',
   'diff-instructions',
   'guidelines'

package/src/ai/prompts/baseline/consolidation/thorough.js

@@ -27,6 +27,7 @@
  * - {{reviewIntro}} - Review introduction line
  * - {{lineNumberGuidance}} - Line number guidance section
  * - {{customInstructions}} - Custom instructions section (optional)
+ * - {{dedupInstructions}} - Dedup instructions section (optional)
  * - {{reviewerSuggestions}} - Formatted reviewer suggestions input
  * - {{suggestionCount}} - Total number of input suggestions
  * - {{reviewerCount}} - Number of reviewers being consolidated
@@ -69,6 +70,23 @@ Take your time to analyze the reviewer findings thoroughly. For each cluster of
 {{customInstructions}}
 </section>
 
+<section name="dedup-instructions" optional="true">
+{{dedupInstructions}}
+</section>
+
+<section name="reviewer-context-guidance" required="true" tier="thorough">
+### Reviewer Context Awareness
+Each reviewer below may have been configured with custom instructions. These fall into two categories:
+
+- **Domain-focused reviewers**: Instructions that specify a code review focus area (e.g., "focus on security", "review error handling", "check performance"). Their findings *within that focus area* carry higher weight than generalist reviewers.
+- **General reviewers**: Either no custom instructions, or instructions about methodology/style/persona (e.g., "be thorough", "use a friendly tone"). Treat their suggestions at face value across all categories.
+
+**Weighting rules:**
+- Only boost a reviewer's findings when their instructions indicate domain expertise relevant to the finding's category
+- Cross-specialty findings from a domain-focused reviewer should be treated as general findings
+- In conflicts within a domain, prefer the domain-focused reviewer's analysis over a generalist's
+</section>
+
 <section name="input-suggestions" locked="true">
 ## Input: {{reviewerCount}} Reviewer(s), {{suggestionCount}} Total Suggestions
 
@@ -111,10 +129,18 @@ When reviewers disagree about an issue:
 ### 4. Quality Filter
 - Drop suggestions with very low confidence (< 0.3) unless multiple reviewers agree
 - Elevate suggestions where reviewers independently converge
+
+### 5. Severity Assessment
+Assess severity based on the evidence and reasoning across all reviewers. When reviewers assign different severities, apply the same conflict resolution principles above. When truly uncertain, preserve the highest severity. Omit severity for praise items.
+
+**Severity Definitions:**
+- **critical**: Production incidents, system failures, or security vulnerabilities — runtime crashes, data corruption or loss, race conditions, deadlocks, breaking changes, changes that will cause existing tests to fail
+- **medium**: Degraded functionality or reliability — missing error handling, N+1 queries, missing validation, missing or poor test coverage for new functionality
+- **minor**: Code quality concerns — documentation gaps, minor optimizations, style inconsistencies
 </section>
 
 <section name="consensus-handling" required="true" tier="thorough">
-### 5. Consensus Handling and Confidence Calibration
+### 6. Consensus Handling and Confidence Calibration
 
 **Cross-Reviewer Agreement:**
 - **Strong consensus** (3+ reviewers): Boost confidence by 0.2 (cap at 1.0)
@@ -131,6 +157,14 @@ When reviewers disagree about an issue:
 Note: Confidence is about certainty of value, not severity.
 </section>
 
+<section name="balanced-output" required="true" tier="thorough">
+### 7. Balanced Output
+- **Deduplicate, don't concatenate**: If two reviewers flagged the same issue, merge them into one suggestion. If three reviewers each found one style nit, consider whether one representative example suffices. Distinct findings should each be preserved — the goal is to eliminate redundancy, not to reduce count.
+- **Limit praise** to 2–3 most noteworthy items across all reviewers. Praise should highlight genuinely commendable patterns, not routine correctness.
+- **Avoid suggestion overload** — a review with 30 suggestions is harder to act on than one with 12 well-chosen ones. Aim for quality over quantity.
+- **Include confidence scores** that reflect cross-reviewer agreement: consensus findings get boosted, lone findings keep their original confidence.
+</section>
+
 <section name="summary-synthesis" required="true" tier="thorough">
 ## Summary Synthesis Guidance
 The summary field should synthesize the findings, not list them.
@@ -139,7 +173,8 @@ The summary field should synthesize the findings, not list them.
 - **Lead with the most important insight**: What should the reviewer focus on first?
 - **Connect the dots**: How do individual findings relate to each other?
 - **Calibrate severity**: Is this code fundamentally sound with minor issues, or are there structural problems?
-- **Write as a single reviewer**: Do NOT mention consolidation, merging, or multiple reviewers
+- **Draw on reviewer summaries**: Use these as evidence for your own synthesis — integrate their insights into a cohesive conclusion rather than listing them individually.
+- **Write as a single reviewer**: Do not mention consolidation, merging, or multiple reviewers -- unless specifically requested
 </section>
 
 <section name="output-schema" locked="true">
@@ -155,6 +190,7 @@ Output JSON with this structure:
       "line": 42,
       "old_or_new": "NEW",
       "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+      "severity": "critical|medium|minor (omit for praise)",
       "title": "Brief title",
       "description": "Detailed explanation",
       "suggestion": "How to fix/improve (omit for praise)",
@@ -165,13 +201,14 @@ Output JSON with this structure:
   "fileLevelSuggestions": [{
     "file": "path/to/file",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title describing file-level concern",
     "description": "Explanation of the file-level observation",
     "suggestion": "How to address the file-level concern (omit for praise items)",
     "confidence": 0.0-1.0,
     "reasoning": ["Step-by-step reasoning explaining why this issue was flagged"]
   }],
-  "summary": "Brief summary of the key findings and their significance. Write as if a single reviewer produced this analysis — do NOT mention 'consolidation', 'merging', or 'multiple reviewers'."
+  "summary": "Single cohesive paragraph synthesizing key findings and their significance. Write as a single reviewer — do not mention consolidation, merging, or multiple reviewers."
 }
 
 ### GitHub Suggestion Syntax
@@ -227,9 +264,12 @@ const sections = [
   { name: 'role-description', required: true, tier: ['thorough'] },
   { name: 'reasoning-encouragement', required: true, tier: ['thorough'] },
   { name: 'custom-instructions', optional: true, tier: ['balanced', 'thorough'] },
+  { name: 'dedup-instructions', optional: true },
+  { name: 'reviewer-context-guidance', required: true, tier: ['thorough'] },
   { name: 'input-suggestions', locked: true },
   { name: 'consolidation-rules', required: true, tier: ['thorough'] },
   { name: 'consensus-handling', required: true, tier: ['thorough'] },
+  { name: 'balanced-output', required: true, tier: ['thorough'] },
   { name: 'summary-synthesis', required: true, tier: ['thorough'] },
   { name: 'output-schema', locked: true },
   { name: 'diff-instructions', required: true, tier: ['thorough'] },
@@ -247,9 +287,12 @@ const defaultOrder = [
   'role-description',
   'reasoning-encouragement',
   'custom-instructions',
+  'dedup-instructions',
+  'reviewer-context-guidance',
   'input-suggestions',
   'consolidation-rules',
   'consensus-handling',
+  'balanced-output',
   'summary-synthesis',
   'output-schema',
   'diff-instructions',

package/src/ai/prompts/baseline/level1/balanced.js

@@ -99,6 +99,15 @@ Identify the following in changed code:
 Do NOT modify files or run write commands. Analyze and report only.
 </section>
 
+<section name="severity-classification" required="true">
+### Severity Classification
+Assign a severity to each suggestion (except praise):
+- **critical**: Production incidents, system failures, or security vulnerabilities — runtime crashes, data corruption or loss, race conditions, deadlocks, breaking changes, changes that will cause existing tests to fail
+- **medium**: Degraded functionality or reliability — missing error handling, N+1 queries, missing validation, missing or poor test coverage for new functionality
+- **minor**: Code quality concerns — documentation gaps, minor optimizations, style inconsistencies
+Omit severity for praise items.
+</section>
+
 <section name="output-schema" locked="true">
 ## Output Format
 
@@ -111,6 +120,7 @@ Do NOT modify files or run write commands. Analyze and report only.
     "line": 42,
     "old_or_new": "NEW",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title",
     "description": "Detailed explanation",
     "suggestion": "How to fix/improve (omit for praise items)",
@@ -183,6 +193,7 @@ const sections = [
   { name: 'initial-setup', required: true },
   { name: 'focus-areas', required: true },
   { name: 'available-commands', required: true },
+  { name: 'severity-classification', required: true },
   { name: 'output-schema', locked: true },
   { name: 'diff-instructions', required: true },
   { name: 'category-definitions', required: true },
@@ -204,6 +215,7 @@ const defaultOrder = [
   'initial-setup',
   'focus-areas',
   'available-commands',
+  'severity-classification',
   'output-schema',
   'diff-instructions',
   'category-definitions',

package/src/ai/prompts/baseline/level1/fast.js

@@ -79,6 +79,14 @@ ONLY suggest for files in this list:
 Annotated diff tool, \`cat -n\`, ls, find, grep. Do NOT modify files.
 </section>
 
+<section name="severity-classification" required="true" tier="fast">
+### Severity
+Assign severity to each suggestion (omit for praise):
+- **critical**: Production incidents, crashes, security vulnerabilities, data corruption/loss, race conditions, breaking changes, test failures
+- **medium**: Degraded functionality, missing error handling, missing validation, missing/poor test coverage
+- **minor**: Code quality, docs gaps, minor optimizations
+</section>
+
 <section name="output-schema" locked="true">
 ## Output Format
 
@@ -91,6 +99,7 @@ Annotated diff tool, \`cat -n\`, ls, find, grep. Do NOT modify files.
     "line": 42,
     "old_or_new": "NEW",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title",
     "description": "Explanation",
     "suggestion": "How to fix (omit for praise)",
@@ -139,6 +148,7 @@ const sections = [
   { name: 'initial-setup', required: true, tier: ['fast'] },
   { name: 'focus-areas', required: true, tier: ['fast'] },
   { name: 'available-commands', required: true, tier: ['fast'] },
+  { name: 'severity-classification', required: true, tier: ['fast'] },
   { name: 'output-schema', locked: true },
   { name: 'diff-instructions', required: true, tier: ['fast'] },
   { name: 'guidelines', required: true, tier: ['fast'] }
@@ -159,6 +169,7 @@ const defaultOrder = [
   'initial-setup',
   'focus-areas',
   'available-commands',
+  'severity-classification',
   'output-schema',
   'diff-instructions',
   'guidelines'

package/src/ai/prompts/baseline/level1/thorough.js

@@ -140,6 +140,15 @@ Your role is strictly to analyze and report findings.
 Note: You may optionally use parallel read-only Tasks to analyze different parts of the changes if that would be helpful for a thorough analysis.
 </section>
 
+<section name="severity-classification" required="true">
+### Severity Classification
+Assign a severity to each suggestion (except praise):
+- **critical**: Production incidents, system failures, or security vulnerabilities — runtime crashes, data corruption or loss, race conditions, deadlocks, breaking changes, changes that will cause existing tests to fail
+- **medium**: Degraded functionality or reliability — missing error handling, N+1 queries, missing validation, missing or poor test coverage for new functionality
+- **minor**: Code quality concerns — documentation gaps, minor optimizations, style inconsistencies
+Omit severity for praise items.
+</section>
+
 <section name="output-schema" locked="true">
 ## Output Format
 
@@ -153,6 +162,7 @@ Output JSON with this structure:
     "line": 42,
     "old_or_new": "NEW",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title",
     "description": "Detailed explanation",
     "suggestion": "How to fix/improve (omit this field for praise items - no action needed)",
@@ -260,6 +270,7 @@ const sections = [
   { name: 'initial-setup', required: true, tier: ['thorough'] },
   { name: 'focus-areas', required: true, tier: ['thorough'] },
   { name: 'available-commands', required: true, tier: ['thorough'] },
+  { name: 'severity-classification', required: true },
   { name: 'output-schema', locked: true },
   { name: 'diff-instructions', required: true, tier: ['thorough'] },
   { name: 'confidence-guidance', required: true, tier: ['thorough'] },
@@ -283,6 +294,7 @@ const defaultOrder = [
   'initial-setup',
   'focus-areas',
   'available-commands',
+  'severity-classification',
   'output-schema',
   'diff-instructions',
   'confidence-guidance',

package/src/ai/prompts/baseline/level2/balanced.js

@@ -97,6 +97,15 @@ Do NOT modify files. Your role is strictly to analyze and report findings.
 Note: You may use parallel read-only Tasks to examine multiple files simultaneously.
 </section>
 
+<section name="severity-classification" required="true">
+### Severity Classification
+Assign a severity to each suggestion (except praise):
+- **critical**: Production incidents, system failures, or security vulnerabilities — runtime crashes, data corruption or loss, race conditions, deadlocks, breaking changes, changes that will cause existing tests to fail
+- **medium**: Degraded functionality or reliability — missing error handling, N+1 queries, missing validation, missing or poor test coverage for new functionality
+- **minor**: Code quality concerns — documentation gaps, minor optimizations, style inconsistencies
+Omit severity for praise items.
+</section>
+
 <section name="output-schema" locked="true">
 ## Output Format
 
@@ -109,6 +118,7 @@ Note: You may use parallel read-only Tasks to examine multiple files simultaneou
     "line": 42,
     "old_or_new": "NEW",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title",
     "description": "Explanation mentioning why full file context was needed",
     "suggestion": "How to fix/improve (omit for praise items)",
@@ -118,6 +128,7 @@ Note: You may use parallel read-only Tasks to examine multiple files simultaneou
   "fileLevelSuggestions": [{
     "file": "path/to/file",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "File-level concern",
     "description": "File-level observation (architecture, organization, naming, etc.)",
     "suggestion": "How to address (omit for praise items)",
@@ -194,6 +205,7 @@ const sections = [
   { name: 'analysis-process', required: true },
   { name: 'focus-areas', required: true },
   { name: 'available-commands', required: true },
+  { name: 'severity-classification', required: true },
   { name: 'output-schema', locked: true },
   { name: 'diff-instructions', required: true },
   { name: 'file-level-guidance', optional: true },
@@ -214,6 +226,7 @@ const defaultOrder = [
   'analysis-process',
   'focus-areas',
   'available-commands',
+  'severity-classification',
   'output-schema',
   'diff-instructions',
   'file-level-guidance',

package/src/ai/prompts/baseline/level2/fast.js

@@ -80,6 +80,14 @@ ONLY suggest for files in this list:
 Annotated diff tool (preferred), \`cat -n <file>\`, ls, find, grep. Do NOT modify files.
 </section>
 
+<section name="severity-classification" required="true" tier="fast">
+### Severity
+Assign severity to each suggestion (omit for praise):
+- **critical**: Production incidents, crashes, security vulnerabilities, data corruption/loss, race conditions, breaking changes, test failures
+- **medium**: Degraded functionality, missing error handling, missing validation, missing/poor test coverage
+- **minor**: Code quality, docs gaps, minor optimizations
+</section>
+
 <section name="output-schema" locked="true">
 ## Output Format
 
@@ -92,6 +100,7 @@ Annotated diff tool (preferred), \`cat -n <file>\`, ls, find, grep. Do NOT modif
     "line": 42,
     "old_or_new": "NEW",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title",
     "description": "Why file context was needed",
     "suggestion": "How to fix (omit for praise)",
@@ -100,6 +109,7 @@ Annotated diff tool (preferred), \`cat -n <file>\`, ls, find, grep. Do NOT modif
   "fileLevelSuggestions": [{
     "file": "path/to/file",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "File-level concern",
     "description": "File-level observation",
     "suggestion": "How to address (omit for praise)",
@@ -146,6 +156,7 @@ const sections = [
   { name: 'analysis-process', required: true, tier: ['fast'] },
   { name: 'focus-areas', required: true, tier: ['fast'] },
   { name: 'available-commands', required: true, tier: ['fast'] },
+  { name: 'severity-classification', required: true, tier: ['fast'] },
   { name: 'output-schema', locked: true },
   { name: 'diff-instructions', required: true, tier: ['fast'] },
   { name: 'guidelines', required: true, tier: ['fast'] }
@@ -166,6 +177,7 @@ const defaultOrder = [
   'analysis-process',
   'focus-areas',
   'available-commands',
+  'severity-classification',
   'output-schema',
   'diff-instructions',
   'guidelines'

package/src/ai/prompts/baseline/level2/thorough.js

@@ -171,6 +171,15 @@ Your role is strictly to analyze and report findings.
 Note: You may optionally use parallel read-only Tasks to examine multiple files simultaneously if that would be helpful for a thorough analysis.
 </section>
 
+<section name="severity-classification" required="true">
+### Severity Classification
+Assign a severity to each suggestion (except praise):
+- **critical**: Production incidents, system failures, or security vulnerabilities — runtime crashes, data corruption or loss, race conditions, deadlocks, breaking changes, changes that will cause existing tests to fail
+- **medium**: Degraded functionality or reliability — missing error handling, N+1 queries, missing validation, missing or poor test coverage for new functionality
+- **minor**: Code quality concerns — documentation gaps, minor optimizations, style inconsistencies
+Omit severity for praise items.
+</section>
+
 <section name="output-schema" locked="true">
 ## Output Format
 
@@ -184,6 +193,7 @@ Output JSON with this structure:
     "line": 42,
     "old_or_new": "NEW",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title",
     "description": "Detailed explanation mentioning why full file context was needed",
     "suggestion": "How to fix/improve based on file context (omit for praise items)",
@@ -193,6 +203,7 @@ Output JSON with this structure:
   "fileLevelSuggestions": [{
     "file": "path/to/file",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title describing file-level concern",
     "description": "Explanation of the file-level observation (architecture, organization, naming, etc.)",
     "suggestion": "How to address the file-level concern (omit for praise items)",
@@ -307,6 +318,7 @@ const sections = [
   { name: 'analysis-process', required: true, tier: ['thorough'] },
   { name: 'focus-areas', required: true, tier: ['thorough'] },
   { name: 'available-commands', required: true, tier: ['thorough'] },
+  { name: 'severity-classification', required: true },
   { name: 'output-schema', locked: true },
   { name: 'diff-instructions', required: true, tier: ['thorough'] },
   { name: 'confidence-guidance', required: true, tier: ['thorough'] },
@@ -331,6 +343,7 @@ const defaultOrder = [
   'analysis-process',
   'focus-areas',
   'available-commands',
+  'severity-classification',
   'output-schema',
   'diff-instructions',
   'confidence-guidance',

package/src/ai/prompts/baseline/level3/balanced.js

@@ -120,6 +120,15 @@ You may use parallel read-only Tasks to explore different areas of the codebase
 {{sparseCheckoutGuidance}}
 </section>
 
+<section name="severity-classification" required="true">
+### Severity Classification
+Assign a severity to each suggestion (except praise):
+- **critical**: Production incidents, system failures, or security vulnerabilities — runtime crashes, data corruption or loss, race conditions, deadlocks, breaking changes, changes that will cause existing tests to fail
+- **medium**: Degraded functionality or reliability — missing error handling, N+1 queries, missing validation, missing or poor test coverage for new functionality
+- **minor**: Code quality concerns — documentation gaps, minor optimizations, style inconsistencies
+Omit severity for praise items.
+</section>
+
 <section name="output-schema" locked="true">
 ## Output Format
 
@@ -133,6 +142,7 @@ Output JSON with this structure:
     "line": 42,
     "old_or_new": "NEW",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title",
     "description": "Detailed explanation mentioning why codebase context was needed",
     "suggestion": "How to fix/improve based on codebase context (omit for praise items)",
@@ -142,6 +152,7 @@ Output JSON with this structure:
   "fileLevelSuggestions": [{
     "file": "path/to/file",
     "type": "bug|improvement|praise|suggestion|design|performance|security|code-style",
+    "severity": "critical|medium|minor (omit for praise)",
     "title": "Brief title describing file-level concern",
     "description": "Explanation of the file-level observation from codebase perspective",
     "suggestion": "How to address the file-level concern (omit for praise items)",
@@ -230,6 +241,7 @@ const sections = [
   { name: 'focus-areas', required: true, tier: ['balanced'] },
   { name: 'available-commands', required: true, tier: ['balanced'] },
   { name: 'sparse-checkout', optional: true, tier: ['fast', 'balanced', 'thorough'] },
+  { name: 'severity-classification', required: true },
   { name: 'output-schema', locked: true },
   { name: 'diff-instructions', required: true, tier: ['balanced'] },
   { name: 'file-level-guidance', optional: true, tier: ['balanced', 'thorough'] },
@@ -252,6 +264,7 @@ const defaultOrder = [
   'focus-areas',
   'available-commands',
   'sparse-checkout',
+  'severity-classification',
   'output-schema',
   'diff-instructions',
   'file-level-guidance',