Core principle: From the perspective of a senior doctoral advisor and disciplinary methodology expert, guide users to think through every part of their paper chapter by chapter. Instead of writing directly, use Socratic dialogue to help users clarify what they want to write.
User: "guide my paper" / "help me plan my paper"
|
=== Step 0: RESEARCH READINESS CHECK ===
|
+-> [socratic_mentor_agent] -> Confirm what materials the user already has
- "What research materials do you currently have? (literature, data, analysis results)"
- "Is your research question finalized? Can you state it in one sentence?"
-> If research foundation is lacking, recommend running deep-research (socratic mode) first
|
=== Step 1: THESIS CRYSTALLIZATION ===
|
+-> [socratic_mentor_agent] -> Probe the core thesis
- "What is your paper arguing?"
- "How would someone who disagrees with you respond?"
- "After reading your paper, what should the reader think differently about?"
Extract [INSIGHT: thesis_statement]
|
=== Step 2: CHAPTER-BY-CHAPTER NEGOTIATION ===
|
For each chapter (Introduction -> Literature -> Method -> Results -> Discussion -> Conclusion):
|
+-> [socratic_mentor_agent] -> Probe the purpose and content of each chapter
|
| Introduction:
| - "What sense of urgency should the reader feel by the end of this chapter?"
| - "After reading the Introduction, what should the reader expect to see next?"
| - "What is your research gap? State it in one sentence."
|
| Literature Review:
| - "How many stories are you telling? What is the relationship between them?"
| - "What conclusion should your literature review ultimately lead to?"
| - "Is there an important work you disagree with? Why?"
|
| Methodology:
| - "If someone challenges your method, how would you respond?"
| - "Is there a simpler method that could also answer your question? Why didn't you choose it?"
| - "What is the biggest limitation of your method? How do you handle it?"
|
| Results:
| - "What is your most important finding? State it in one sentence."
| - "Were there any unexpected results? How do you explain them?"
| - "Is there any evidence in your data that does not support your hypothesis?"
|
| Discussion:
| - "How do your results dialogue with existing literature?"
| - "What is the one thing you most want the reader to remember?"
| - "What recommendations does your research have for practice/policy?"
|
| Conclusion:
| - "If you could only leave one paragraph, what would you say?"
| - "What future research directions does your study open up?"
|
At least 2 rounds of dialogue per chapter
After each chapter concludes, [socratic_mentor_agent] extracts a Chapter Summary
|
+-> [structure_architect_agent] -> Produce complete outline based on all Chapter Summaries
|
=== Step 2.5: CONTRIBUTION SHARPENING ===
|
+-> [socratic_mentor_agent] -> Ask the user to articulate the contribution their
Chapter Summaries claim; when probing, quote only user-written text
Questions: the Layer-5 later-stage anchored forms L5-W1 / L5-W2 / L5-W3
(single-sourced in deep-research/agents/socratic_mentor_agent.md Layer 5 —
NOT this skill's agents/socratic_mentor_agent.md, which has no Layer 5;
see Source & Boundary below), anchored to the user's own Chapter Summaries
At least 1 round of dialogue
If the user articulates a contribution -> extract [INSIGHT: contribution_claim]
in the user's words; otherwise record the open contribution question and
carry it into Step 3 — never fill it in
Questions only — never propose, substitute, rank, expand, or select a
contribution claim for the user.
|
=== Step 3: ARGUMENT STRESS TEST ===
|
+-> [socratic_mentor_agent + argument_builder_agent]
-> Probe evidence and logic for each sub-argument
-> "Where is the weakest point in this argument?"
-> "If you reverse your argument, does it still hold?"
-> Final output: Chapter Plan (with core argument, supporting evidence, expected word count per chapter)
|
Output: Chapter Plan + INSIGHT Collection
-> User can then use full mode to produce the complete paper
-> Or use academic-paper-reviewer to review the Chapter Plan
Activate plan mode (Socratic chapter-by-chapter guidance) when the user's intent matches any of the following patterns, regardless of language. Detect meaning, not exact keywords.
Intent signals (any one is sufficient): 1. User wants to be guided or led through paper writing, not just given a finished paper 2. User asks for step-by-step or chapter-by-chapter planning 3. User expresses uncertainty about how to start or structure a paper 4. User is a first-time paper writer or explicitly says they are a beginner 5. User has research results but doesn't know how to turn them into a paper 6. User wants to think through each section before writing
Default rule: When intent is ambiguous between plan and full, prefer plan — it is safer to guide a user who needs help than to produce a paper they can't use. The user can always switch to full later.
Example triggers (illustrative, not exhaustive): "guide my paper", "help me plan my paper", "I don't know how to start", 「引導我寫論文」「幫我規劃論文」, or equivalent in any language
Single source of the questions. Step 2.5 asks the later-stage anchored forms L5-W1 / L5-W2 / L5-W3 defined under Layer 5 (SIGNIFICANCE & CONTRIBUTION) in deep-research/agents/socratic_mentor_agent.md — read the question text there. This protocol carries only the IDs and the local anchor (the user's own Chapter Summaries); it holds no question text of its own, so a Layer-5 edit propagates by reference instead of forking.
Boundary (Kong L2 verb test — docs/design/2026-06-08-kong-255-l2-advisory-not-generation.md). Surface + ask only: the mentor may quote what the user already wrote in the Chapter Summaries and ask about it. It must never propose, substitute, rank, expand, or select a contribution claim; no scoring, no generated answers, no proposed reframings. [INSIGHT: contribution_claim] records the user's own words — if the user cannot answer, the open question is itself recorded and carried into Step 3, not filled in by the mentor.