This skill turns the actionable workflow for social science master's and doctoral research topic selection (derived from the ideas in the article "Finding a PhD Topic with Codex") into an executable writing and self-check process, covering quantitative, qualitative, ethnographic, archival, documentary, corpus, and digital research.
Scope of application: Social sciences and adjacent humanities disciplines. Quantitative, qualitative, ethnographic, archival, documentary, corpus, and digital research can all use it. The criteria for judgment are always the same:Is the question clear, does the evidence match, is the claim honest, and can the project be completed by this researcher.
What it does not do: It does not generate a "correct topic" for you, does not guarantee novelty, admission, funding, publication, or thesis success, and does not treat any one method as the default answer for high-quality research.AI's confidence is not evidence; a filled-out form is not proof of passing.
I. First do routing: where is the user in the process?
| User request | Corresponding element | Which steps to take | Which references to read |
|---|---|---|---|
| "I want to research XX" / "Help me find a paper topic" | Topic ideation | Step 1 → 2 → 3 | 01-foundations.md、02-workflow-steps1-3.md |
| "Has anyone done this direction" / how to review the literature / how to draw the map | Literature review | Step 2 → 3 | 02-workflow-steps1-3.md |
| "I have data / I want to use method M" | Topic ideation (false starting point) | First return to Step 1 | 01-foundations.md §VI Two false starts |
| "Can this topic be done" / is the evidence enough | Argument organization + feasibility | Step 4 → 7 | 03-workflow-steps4-5.md、04-workflow-steps6-8.md |
| "Help me generate a few research questions" / how to write an RQ | Topic ideation | Step 5 | 03-workflow-steps4-5.md |
| "Help me think about how reviewers would attack it" / check novelty | Argument organization | Step 6 | 04-workflow-steps6-8.md |
| "Is the conclusion overstated" / expression grading | Academic expression | Step 4's claim ceiling + terminology discipline | 09-academic-expression.md、07-status-terminology.md |
| "Help me revise the draft / polish the argument paragraph" | Academic expression | Compare against the claim ceiling and Red flags | 09-academic-expression.md |
| "How to talk with the advisor" / how to write the proposal materials | Deliverable use | Part III | 08-usage-handoff.md |
| Need ready-made prompts | — | Take by step | 05-prompt-library.md(the only authoritative version) |
| Need form templates | — | Take by step | 06-worksheets.md |
Entry-point judgment (mandatory first step): Only interest → Mode A; already have a broad theme or relationship → Mode B; already have a tentative research question → Mode C (audit route, not advanced route). When there is only a dataset or method preference, first ask "what unresolved phenomenon can it help understand"; if that cannot be answered, start from Mode A.
II. Four iron rules (must be observed in every reply)
- Write "unknown" where you don't know, and don't let AI fill in guesses.
- Actively look for contradictions, zero results, inverse relationships, and alternative explanations—not for material that supports the starting point.
- Preserve evidence levels, evidence status, failed searches, and key unknowns, and don't flatten them into an optimistic conclusion.
- After each step, update
Current Research State, and don't carry the entire old conversation into the next step.
Supporting prohibitions (write them into every prompt you give):
- Do not generate topics, questions, or hypotheses (when that step has not yet been reached);
- Do not claim that a research gap or novelty exists;
- Do not default to quantitative, qualitative, causal, or any specific method;
- Do not write "not found in this round" as "no research";
- Do not write personal interest as already proven academic importance;
- Do not fabricate citations, sources, or results.
III. Twelve judgments (red lines for writing self-checks)
Any judgment about a topic or argument should first pass through this table. See the full version in 01-foundations.md §IV.
| Do not conflate them | Actionable check |
|---|---|
| Interest ≠ Topic | Have the phenomena, processes, and boundaries to be explored already been delineated? |
| Topic ≠ Research problem | Does it point out specifically where the field fails to explain things? |
| Research problem ≠ Research question | Does it turn a knowledge difficulty into a task that can be advanced by evidence? |
| Research gap ≠ "few studies" | What knowledge consequences does the lack of research cause? |
| Novelty ≠ nobody has studied it | Does it resolve a contradiction, mechanism, measurement, or newly answerable question? |
| Interesting ≠ Important | Besides personal interest, whose understanding of what would the answer change? |
| Important ≠ Answerable | Is there suitable, compliant, and time-matched evidence? |
| Available source ≠ Valid evidence | Do the fields, texts, or observations really correspond to the target concept? |
| Correlation ≠ Mechanism | Do you see the process, or only phenomena occurring at the same time? |
| Advanced method ≠ Strong contribution | Does the method resolve the original knowledge difficulty? |
| Feasible study ≠ PhD-worthy project | Can it support one core puzzle and coherent development? |
| AI judgement ≠ Researcher judgement | Which decisions require the user's values, experience, and advisor judgment? |
IV. One-page overview of the eight-step workflow
| Step | Main output | Decision question | Gate |
|---|---|---|---|
| 1. Research Search Space | One-page search space | Are the concepts and boundaries sufficient for searching? | PASS / REVISE / STOP |
| 2. Research Landscape | Field map | Do you see multiple problems, explanations, and counterevidence? | Same as above |
| 3. Unresolved Problems | 3–7 question cards | Is there consequential knowledge uncertainty? | Same as above |
| 4. Evidence Map | 1–3 evidence routes | Can the core problem be observed and the upper limit of claims be maintained? | Same as above |
| 5. Candidate RQs | 6–12 candidate cards | Are there substantively different, attackable questions? | Same as above (activate PARK) |
| 6. Academic Survival | Novelty Audit + Reviewer 2 | Do the candidates survive academic attack? | Same as above |
| 7. PhD Feasibility | Feasibility dossier | Are there no unhandled hard failures, and can it develop into a doctorate? | Same as above |
| 8. Final Shortlist | At most three Topic Cards | Can the trade-offs and the next item to verify be clearly explained? | Same as above |
Step 5 will briefly expand candidates, then narrow them,this is not a purely linear pipeline。
Fallback rules: If the design can only provide a weak proxy, the wrong unit, or an unsuitable time → return Question; if even the knowledge consequence of the question disappears → return Problem.
V. Five state systems (do not merge into one score)
| What is being asked | Which system to use | Value |
|---|---|---|
| Can I proceed to the next step? | Workflow Gate | PASS / REVISE / STOP（+ PARK） |
| How far has this evidence route been verified? | Evidence State | NAMED / DOCUMENTED / ACCESS-CONFIRMED / TESTED |
| How does this candidate stand after academic attack? | Academic Survival | SURVIVE / REDESIGN / PARK / REJECT |
| What is the current status of this hard condition? | PhD Feasibility Dependency | CONFIRMED / RESOLVABLE DEPENDENCY / UNKNOWN / FAIL |
| Can this candidate enter the final comparison? | PhD Feasibility Disposition | PROCEED TO COMPARISON / REDESIGN / PARK / REJECT |
Six inequalities that must not be flattened：
UNKNOWN ≠ PASS— Unknown requires a verification action, not an optimistic interpretation.DOCUMENTED ≠ ACCESS-CONFIRMED— An official document saying the source exists does not mean the user has satisfied the eligibility, permission, location, cost, and ethics conditions.SURVIVE ≠ novel— Passing one bounded attack does not amount to proving that no similar research exists anywhere in the world.PARK ≠ bad idea— What is missing is decisive evidence or conditions, not the idea itself.PASS ≠ permanent approval— New literature, changes in access, ethics opinions, or empirical results may all turn it into REVISE or STOP.REJECT ≠ wasted work— The rejection record preserves the closest work, failure conditions, and boundaries, and can prevent the same mistake from being repeated later.
Do not create a "composite status". A candidate can perfectly well be simultaneously: Step 6 SURVIVE, evidence DOCUMENTED, access dependency UNKNOWN, Step 7 disposition PARK, this step's Gate REVISE.
These labels are not contradictory; they answer different questions. See 07-status-terminology.md。
VI. Working rhythm (repeat these five beats at each step)
- Read the current chapter and confirm this step'spurpose and Gate；
- In
05-prompt-library.mdfind the corresponding prompt, prepare onlyRequired input； - Write the output into
06-worksheets.mdin the corresponding table,do not treat a long conversation as the final record； - Review the results with the quality check and Red flags;stop when human judgment is needed；
- Update
Current Research State, then follow08-usage-handoff.mdto proceed to the next step or temporarily exit.
Context control: at each transition carry only the content of Carry forward , and actively discard Leave behind。
08-usage-handoff.md gives what each of the nine handoffs should carry, what to discard, which judgment must be made by the user,
and which type of unknown will block progress.
VII. When you must stop and find a human
When you cannot judge the weight of a debate within this discipline, when key full texts or source descriptions cannot be obtained, when the research involves sensitive participants or high-risk materials, when the cost of learning the method is unclear, or when the project depends on a particular supervisor, institution, or source provider—hand the current card to the appropriate person。
AI can help you prepare questions,but cannot authorize on their behalf。
For typical scenarios and methods that need to be handed to a human, see 08-usage-handoff.md and 01-foundations.md §V "When to find a human".
VIII. Usage conventions
- Forms of address in the prompts：
05-prompt-library.mdThe prompts inpreserve verbatimthe original, and in the main text refer to yourself as "I am using Codex". Just copy directly; if using in another agent, you may replaceCodexwith the current assistant's name, and change nothing else. - The only authoritative version of the prompts is
05-prompt-library.md. The "short previews" in the step files are for explanation only; do not create a parallel version to maintain separately. - Quick / Deep tiers: Steps 2, 4, 6, and 7 provide both Quick Run and Deep Run tiers. Quick Run reduces the first-round workload,without lowering the standards for evidence, ethics, access, or hard failure。
- Do not treat long conversations as records: all conclusions must go into the corresponding tables of
06-worksheets.mdandCurrent Research State。 - write "unknown" for anything uncertain, and do not accept optimistic completion by any party (including yourself).
IX. Reference File Index
The following nine files are all located under this skill's references/ directory; read only the one needed for the current step, as required.
| File | Content |
|---|---|
01-foundations.md |
Core concepts, six easily confused concepts, twelve judgments, capability boundaries, evidence levels, three entry modes, Gate definitions, Funnel |
02-workflow-steps1-3.md |
Step 1 Search Space / Step 2 Domain Landscape (core literature review) / Step 3 Open Questions |
03-workflow-steps4-5.md |
Step 4 Evidence Map and claim ceiling / Step 5 Candidate Research Questions |
04-workflow-steps6-8.md |
Step 6 Academic Attack (Novelty Audit + Reviewer 2) / Step 7 Feasibility / Step 8 Final Shortlist |
05-prompt-library.md |
All copyable prompts (single authoritative version） |
06-worksheets.md |
11 worksheet templates |
07-status-terminology.md |
Five sets of status systems + glossary + Current Research State |
08-usage-handoff.md |
Shortlist use, advisor communication, nine handovers, recovery rules |
09-academic-expression.md |
Academic expression discipline: claim ceilings, matching verbs to evidence, honest phrasing sentence patterns, Red flags for each step (required reading when drafting and revising) |
When using it, do only four things: write "unknown" for what you do not know; require counterevidence to be sought; retain evidence levels and failure records; update the status table at each step.