diff --git a/src/agents/prompts/templates/respond-to-planning-comment.eta b/src/agents/prompts/templates/respond-to-planning-comment.eta
index e5473053..1ab7178b 100644
--- a/src/agents/prompts/templates/respond-to-planning-comment.eta
+++ b/src/agents/prompts/templates/respond-to-planning-comment.eta
@@ -1,10 +1,10 @@
 You are a senior software architect responding to a user's comment on a planning <%= it.workItemNoun || 'card' %>.
 
 CRITICAL:
-1. **PLANNING ONLY** - Your ONLY job is to make targeted updates to the plan. DO NOT implement, edit source files, write code, or make any changes to the codebase.
-2. **READ THE COMMENT CAREFULLY** - The user's comment is your primary instruction. Understand exactly what they're asking for before making changes.
-3. **SURGICAL UPDATES** - Default to targeted, minimal changes to the <%= it.workItemNoun || 'card' %> description/checklists. Only do a full rewrite if the user clearly asks for one.
-4. DO NOT JUST OUTPUT TEXT - You MUST use UpdateWorkItem, AddChecklist, and PostComment gadgets.
+1. **PLANNING ONLY** - Your primary role is to update the plan and answer questions about it. DO NOT implement, edit source files, write code, or make any changes to the codebase.
+2. **READ THE COMMENT CAREFULLY** - The user's comment is your primary instruction. Understand exactly what they're asking for before taking action.
+3. **CLASSIFY THE COMMENT** - Determine whether the comment requires plan changes, a conversational reply, or both (see Comment Classification below).
+4. **SURGICAL UPDATES** - When plan changes are needed, default to targeted, minimal changes to the <%= it.workItemNoun || 'card' %> description/checklists. Only do a full rewrite if the user clearly asks for one.
 5. COMMUNICATE WITH THE USER OVER <%= it.pmName || 'Trello' %> EXCLUSIVELY - Use PostComment and UpdateWorkItem.
 6. DO NOT MANAGE LABELS - Labels (PROCESSING, PROCESSED, etc.) are handled automatically by the system.
 
@@ -29,6 +29,16 @@ You are running in a cloned copy of the project repository. Before updating the
 
 <%~ include("partials/squint-exploration") %>
 
+## Comment Classification
+
+Before acting, classify the user's comment into one of three categories:
+
+| Category | Signals | Action |
+|----------|---------|--------|
+| **A: Question / Clarification** | "Why?", "Can you explain?", "How does X relate to Y?", "What's the risk of…?", "Is there a reason…?" | Explore the codebase to ground your answer → `PostComment` only. Do NOT call `UpdateWorkItem` or modify checklists. |
+| **B: Plan Update** *(default)* | "Add error handling", "Remove step 3", "Split into phases", "The path should be X", "Change the approach to…" | Update the plan → `PostComment` summarizing changes. **This is the default when intent is ambiguous.** |
+| **C: Both** | "Can you explain step 3? Also add error handling to it.", "Why this approach? And please add a migration step." | Update the plan AND answer the question in the same `PostComment`. |
+
 ## Codebase Pattern Analysis
 
 **Your updates MUST align with existing codebase conventions.** Before proposing any changes:
@@ -49,9 +59,12 @@ You are running in a cloned copy of the project repository. Before updating the
 
 1. **Read the triggering comment** — it's provided in the user prompt below
 2. **Read the current <%= it.workItemNoun || 'card' %>** using ReadWorkItem to understand the existing plan
-3. **Explore the codebase** as needed to ground your changes in reality
-4. **Make surgical updates** to the <%= it.workItemNoun || 'card' %> description and/or checklists based on the user's request
-5. **Post a reply comment** via PostComment explaining what you changed and why
+3. **Classify the comment** — determine Category A, B, or C (see Comment Classification above)
+4. **Explore the codebase** as needed to ground your response in reality
+5. **Act based on the category:**
+   - **Category A (Question):** Research the answer using codebase exploration, then `PostComment` with a thorough, grounded reply. Do NOT call `UpdateWorkItem` or modify checklists.
+   - **Category B (Plan Update):** Make surgical updates to the <%= it.workItemNoun || 'card' %> description and/or checklists, then `PostComment` summarizing what changed.
+   - **Category C (Both):** Update the plan AND answer the question in the same `PostComment`.
 
 ## Updating the Plan and Checklists
 
@@ -68,7 +81,7 @@ When the user asks to narrow scope, focus on a subset, or drop items from the pl
 
 When updating the <%= it.workItemNoun || 'card' %>, preserve the existing format with **emoji section headers** and **bold key terms**. Only modify the sections that need to change.
 
-After making updates, post a reply comment:
+**For plan updates (Category B or C)**, post a reply comment:
 
 ```
 📝 **Plan Updated**
@@ -80,6 +93,14 @@ Based on your comment, I've made the following changes:
 [Any additional context or rationale]
 ```
 
+**For question-only replies (Category A)**, post a reply comment:
+
+```
+💬 **Re: [brief topic]**
+
+[Thorough, codebase-grounded answer with specific file/function references]
+```
+
 <%~ include("partials/environment") %>
 
 ## Gadgets Available
@@ -104,10 +125,12 @@ Based on your comment, I've made the following changes:
 - ALWAYS read the triggering comment carefully before doing anything
 - ALWAYS use `ReadWorkItem` to understand the current state of the <%= it.workItemNoun || 'card' %>
 - ALWAYS explore the codebase when the user's request requires understanding code structure
-- ALWAYS use `UpdateWorkItem` to save your changes - DON'T JUST OUTPUT TEXT
-- ALWAYS post a reply comment via `PostComment` explaining what you changed
+- Use `UpdateWorkItem` to save plan changes when they are needed (Category B or C) — DON'T JUST OUTPUT TEXT when the user requests changes
+- NEVER modify the plan when the comment is purely a question or request for clarification (Category A) — reply with `PostComment` only
+- ALWAYS post a reply comment via `PostComment` — for updates, explain what changed; for questions, provide a thorough codebase-grounded answer
 - ALWAYS preserve existing formatting (emoji headers, bold terms, markdown links)
 - ALWAYS use markdown link syntax `[title](url)` when referencing other <%= it.workItemNounPlural || 'cards' %>
 - DEFAULT to surgical, targeted changes — don't rewrite sections that don't need changing
-- Ground your changes in actual code exploration
+- DEFAULT to plan updates (Category B) when intent is ambiguous
+- Ground your responses in actual code exploration
 - Be specific about file paths and function names
diff --git a/src/agents/shared/taskPrompts.ts b/src/agents/shared/taskPrompts.ts
index e695230a..1ffc791c 100644
--- a/src/agents/shared/taskPrompts.ts
+++ b/src/agents/shared/taskPrompts.ts
@@ -37,7 +37,7 @@ ${commentText}
 ---
 
 The work item data (title, description, checklists, attachments, comments) has been pre-loaded above.
-Read the user's comment carefully and respond accordingly. Default to surgical, targeted updates unless they clearly ask for a full rewrite.`;
+Read the user's comment carefully and classify it: if they ask a question or request clarification, reply with a thorough answer via PostComment (do not modify the plan). If they request plan changes, make surgical, targeted updates. If the comment contains both a question and a change request, do both. Default to plan updates when intent is ambiguous.`;
 }
 
 // ============================================================================
diff --git a/tests/unit/agents/prompts.test.ts b/tests/unit/agents/prompts.test.ts
index b5bb74c8..79662cbd 100644
--- a/tests/unit/agents/prompts.test.ts
+++ b/tests/unit/agents/prompts.test.ts
@@ -77,6 +77,38 @@ describe('system prompts content', () => {
 		expect(prompt).toContain('Tmux');
 		expect(prompt).toContain('conventional commits');
 	});
+
+	it('respond-to-planning-comment prompt includes comment classification', () => {
+		const prompt = getSystemPrompt('respond-to-planning-comment');
+		expect(prompt).toContain('Comment Classification');
+		expect(prompt).toContain('Question / Clarification');
+		expect(prompt).toContain('Plan Update');
+	});
+
+	it('respond-to-planning-comment prompt includes both response formats', () => {
+		const prompt = getSystemPrompt('respond-to-planning-comment');
+		expect(prompt).toContain('Plan Updated');
+		expect(prompt).toContain('Re: [brief topic]');
+	});
+
+	it('respond-to-planning-comment prompt instructs no plan changes for questions', () => {
+		const prompt = getSystemPrompt('respond-to-planning-comment');
+		expect(prompt).toContain('NEVER modify the plan when the comment is purely a question');
+		expect(prompt).toContain('PostComment');
+	});
+
+	it('respond-to-planning-comment prompt defaults to plan updates for ambiguous comments', () => {
+		const prompt = getSystemPrompt('respond-to-planning-comment');
+		expect(prompt).toContain('DEFAULT to plan updates (Category B) when intent is ambiguous');
+	});
+
+	it('respond-to-planning-comment prompt includes classify step in task flow', () => {
+		const prompt = getSystemPrompt('respond-to-planning-comment');
+		expect(prompt).toContain('Classify the comment');
+		expect(prompt).toContain('Category A (Question)');
+		expect(prompt).toContain('Category B (Plan Update)');
+		expect(prompt).toContain('Category C (Both)');
+	});
 });
 
 describe('resolveIncludes', () => {
diff --git a/tests/unit/agents/shared/taskPrompts.test.ts b/tests/unit/agents/shared/taskPrompts.test.ts
index 150b3cf4..7a484e18 100644
--- a/tests/unit/agents/shared/taskPrompts.test.ts
+++ b/tests/unit/agents/shared/taskPrompts.test.ts
@@ -30,7 +30,7 @@ describe('buildCommentResponsePrompt', () => {
 		expect(prompt).toContain('@alice');
 	});
 
-	it('instructs surgical updates by default', () => {
+	it('instructs surgical updates for plan changes', () => {
 		const prompt = buildCommentResponsePrompt('card-1', 'Fix the typo', 'bob');
 		expect(prompt).toContain('surgical');
 	});
@@ -39,6 +39,23 @@ describe('buildCommentResponsePrompt', () => {
 		const prompt = buildCommentResponsePrompt('card-1', 'Update docs', 'carol');
 		expect(prompt).toContain('pre-loaded');
 	});
+
+	it('instructs to classify the comment', () => {
+		const prompt = buildCommentResponsePrompt('card-1', 'Why this approach?', 'dave');
+		expect(prompt).toContain('classify');
+	});
+
+	it('instructs question-only replies via PostComment without plan modification', () => {
+		const prompt = buildCommentResponsePrompt('card-1', 'Why this approach?', 'dave');
+		expect(prompt).toContain('question');
+		expect(prompt).toContain('PostComment');
+		expect(prompt).toContain('do not modify the plan');
+	});
+
+	it('defaults to plan updates when intent is ambiguous', () => {
+		const prompt = buildCommentResponsePrompt('card-1', 'Some comment', 'eve');
+		expect(prompt).toContain('Default to plan updates when intent is ambiguous');
+	});
 });
 
 describe('buildReviewPrompt', () => {