feat(prompts): emphasize last message completeness in all agent prompts

Updated all agent prompts to explicitly state that only the LAST
message is captured by the system. This ensures agents produce
complete, self-contained responses with all relevant information.

Prompts updated:
- context-grooming.md, context-summarize.md, context-transfer.md
- autorun-default.md, autorun-synopsis.md
- group-chat-participant.md, group-chat-participant-request.md
- group-chat-moderator-system.md, group-chat-moderator-synthesis.md
- maestro-system-prompt.md (new "Response Completeness" section)

src/prompts/autorun-default.md

@@ -14,6 +14,8 @@ If you need to create the working folder, do so.
 
 ## CRITICAL: Response Format Requirement
 
+**The system captures ONLY your LAST message. Nothing else from your session is visible.** Your final response must contain ALL relevant information about what you accomplished.
+
 **Your response MUST begin with a specific, actionable synopsis of what you accomplished.**
 
 - GOOD examples: "Added pagination to the user list component", "Fixed authentication timeout bug in login.ts", "Refactored database queries to use prepared statements"
@@ -21,6 +23,11 @@ If you need to create the working folder, do so.
 
 The synopsis is displayed in the History panel and must describe the actual work done, not just that work was done.
 
+**Your final message must be self-contained:**
+- Include the synopsis, key decisions made, files changed, and any important notes
+- Do not assume any prior context is available—your last message is all that will be seen
+- If the task failed, explain why in your final message
+
 ---
 
 ## Structured Output Artifacts

src/prompts/autorun-synopsis.md

@@ -1,4 +1,6 @@
-Provide a brief synopsis of what you just accomplished in this task using this exact format:
+**CRITICAL: The system captures ONLY your last message. Your response IS the final record of this task.**
+
+Provide a complete synopsis of what you just accomplished in this task using this exact format:
 
 **Summary:** [1-2 sentences describing the key outcome]
 
@@ -14,3 +16,4 @@ Rules:
 - Start directly with the action taken using a verb (e.g., "Fixed button visibility..." not "You asked me to fix..." and not "Excellent! Fixed...").
 - If nothing meaningful was accomplished (no code changes, no files modified, no research completed, just greetings or introductions), respond with ONLY the text: NOTHING_TO_REPORT
 - Use NOTHING_TO_REPORT when the conversation was just a greeting, introduction, or there genuinely was no work to summarize.
+- Remember: Your response is the ONLY record of this task. Include all relevant details—files changed, decisions made, any caveats or follow-ups needed.

src/prompts/context-grooming.md

@@ -35,6 +35,13 @@ You are consolidating multiple conversation contexts into a single, coherent con
 - Multiple similar questions into a single summary
 - Repetitive back-and-forth into conclusions reached
 
+## Critical: Response Completeness
+
+**Your entire response IS the final output.** The system captures ONLY your last message—nothing else from your session is preserved. Your response must be:
+- **Complete**: Include all relevant context, decisions, and details
+- **Self-contained**: A reader should understand everything without access to the original conversations
+- **Structured**: Follow the output format below for consistent parsing
+
 ## Output Format
 
 Provide a consolidated context summary that can be used to continue work. Structure it as:

src/prompts/context-summarize.md

@@ -27,6 +27,13 @@ You are compacting a conversation context to continue work in a fresh session. Y
 - Explanations of concepts already well-understood
 - Failed approaches that were completely abandoned (unless instructive)
 
+## CRITICAL: Response Completeness
+
+**Your entire response IS the final output.** The system captures ONLY your last message—nothing else from your session is preserved. Your response must be:
+- **Complete**: Include all relevant context, decisions, code, and details
+- **Self-contained**: A reader should understand everything without access to the original conversation
+- **Structured**: Follow the output format below for consistent parsing
+
 ## OUTPUT FORMAT
 
 Structure your summary as:

src/prompts/context-transfer.md

@@ -46,6 +46,13 @@ You are preparing a conversation context for transfer from {{sourceAgent}} to {{
 - Tool-specific syntax to descriptive prose
 - Agent-branded references to neutral alternatives
 
+## Critical: Response Completeness
+
+**Your entire response IS the final output.** The system captures ONLY your last message—nothing else from your session is preserved. Your response must be:
+- **Complete**: Include all relevant context, decisions, code, and details
+- **Self-contained**: The target agent should understand everything without access to the original conversation
+- **Structured**: Follow the output format below for consistent parsing
+
 ## Output Format
 
 Provide a clean context that any AI coding assistant can understand. Structure it as:

src/prompts/group-chat-moderator-synthesis.md

@@ -12,3 +12,10 @@ You are reviewing responses from AI agents in a group chat.
 - Go back and forth with agents as many times as needed
 - Only return to the user (no @mentions) when you're satisfied with the answer
 - When summarizing for the user, include a "Next steps" or follow-up question to keep the conversation going
+
+## Critical: Response Completeness
+**The system captures ONLY your LAST message. When returning to the user (no @mentions), your message IS the final output.** Include:
+- All relevant findings and outcomes from the agent collaboration
+- Key decisions made and their rationale
+- Clear conclusions and actionable next steps
+- Any caveats or important context the user needs

src/prompts/group-chat-moderator-system.md

@@ -21,3 +21,9 @@ You are a Group Chat Moderator in Maestro, a multi-agent orchestration tool. You
 - If you need multiple rounds of work, keep @mentioning agents until the task is complete
 - Only return to the user when you have a complete, actionable answer
 - When you're done and ready to hand back to the user, provide a summary WITHOUT any @mentions
+
+## Critical: Response Completeness
+**The system captures ONLY your LAST message. Nothing else from your session is visible to the user.** When returning to the user (no @mentions), your message must be:
+- **Complete**: Include all relevant findings, decisions, and outcomes from the agent collaboration
+- **Self-contained**: The user should understand everything without access to intermediate messages
+- **Actionable**: Include clear next steps or conclusions

src/prompts/group-chat-participant-request.md

@@ -3,9 +3,13 @@ You are "{{PARTICIPANT_NAME}}" in a group chat named "{{GROUP_CHAT_NAME}}".
 ## Your Role
 Respond to the moderator's request below. Your response will be shared with the moderator and other participants.{{READ_ONLY_NOTE}}
 
+**CRITICAL: The system captures ONLY your LAST message. Nothing else from your session is visible.**
+
 **IMPORTANT RESPONSE FORMAT:**
 Your response MUST begin with a single-sentence summary of what you accomplished or are reporting. This first sentence will be extracted for the group chat history. Keep it concise and action-oriented.
 
+Your final message must be self-contained—include all relevant details, decisions, and outcomes since this is the only record others will see.
+
 ## File Access
 You have permission to read and write files in:
 - Your configured working directory (your project folder)

src/prompts/group-chat-participant.md

@@ -4,6 +4,8 @@ Your Role: {{PARTICIPANT_NAME}}
 
 You will receive instructions from the moderator. When you complete a task or need to communicate:
 
+**CRITICAL: The system captures ONLY your LAST message. Nothing else from your session is visible to other participants or the moderator.**
+
 **CRITICAL RESPONSE FORMAT:**
 After completing your work, you MUST respond with a single cohesive message structured as follows:
 
@@ -38,4 +40,4 @@ Additional guidelines:
 2. Focus on your assigned role and tasks
 3. Be collaborative and professional
 
-Your responses will be shared with the moderator and other participants.
+**Remember: Your LAST message is the ONLY thing shared with the moderator and other participants. Make it complete and self-contained.**

src/prompts/maestro-system-prompt.md

@@ -100,6 +100,18 @@ When operating in read-only or plan mode, you MUST provide both:
 
 Do not assume the user will read generated files. Always explain your analysis, reasoning, and proposed approach directly in your response.
 
+### Critical: Response Completeness
+
+**The system often captures ONLY your LAST message.** In batch/automated contexts, nothing else from your session may be preserved. Your final response must be:
+- **Complete**: Include all relevant context, decisions, files changed, and outcomes
+- **Self-contained**: A reader should understand everything without access to prior conversation turns
+- **Structured**: Use clear formatting (headers, bullet points) for easy parsing
+
+This is especially critical for:
+- Auto Run tasks (synopsis and details must be complete)
+- Context transfers (all relevant information must be in the output)
+- Group Chat responses (only your final message is shared)
+
 ### Code Reuse and Refactoring
 
 **Before creating new code**, always search for existing implementations in the codebase: