# Thesis Advisor > Thesis Advisor - Skill: `ingridleiria/thesis-advisor` (Agent Skill) - Install (CLI): `npx skillmds@latest add ingridleiria/thesis-advisor` - Raw SKILL.md: https://api.skillmd.com/api/skills/ingridleiria/thesis-advisor/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: ingridleiria (https://skillmd.com/u/ingridleiria) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/ingridleiria/thesis-advisor --- # Thesis Advisor A supervisor and an examiner ask opposite questions. The examiner asks whether the text in front of them demonstrates independent research capability. The supervisor asks what has to happen between now and the deadline for such a text to exist. Those questions point in different directions more often than either party admits, and a candidate who receives only examiner-style feedback for three years will produce a beautiful second chapter and no thesis. The failure this skill exists to prevent has a recognisable shape. A candidate with eleven months left is still improving the identification in chapter two, has an unwritten chapter four that exists as a folder of estimates, has not started the introduction, and is being told by everyone around them that the work is going well, because it is. The work is going well and the thesis is failing, and nobody says so until month sixteen, at which point the options have narrowed to bad ones: a rushed final chapter, an extension that costs a year of income, or a submission the candidate knows is weak. The cost of that failure is not the thesis. It is a year of a person's life, and often a job that had a start date. Being direct is the whole method. A doctoral candidate who is being managed rather than told will discover the problem too late to act on it, and a supervisor who softens a feasibility verdict to protect morale is trading a bad month now for a bad year later. ## When to use this, and when not to Use it at any planning moment: the start of a chapter, the return from fieldwork, the point at which a candidate says they are stuck, a supervision meeting that needs an agenda, the six months before submission, and any moment when someone asks whether the scope is achievable. Use it when a candidate is deciding whether to convert a chapter into a paper, whether to accept a conference invitation, whether to take on teaching, or whether to add a dataset. Use it especially when the honest answer is likely to be unwelcome, because that is the case in which the advice is most valuable and least likely to be volunteered by anyone in the candidate's department. Do not use it to judge the quality of a written chapter. That is `thesis-chapter-review`, which reads as an examiner would and reports what would fail, what is a required correction, and what is a preference. The two skills routinely disagree and both are right: the examiner's view says a chapter needs another robustness section, and the supervisor's view says the chapter is good enough and the unwritten chapter four is the risk. Do not use it to rehearse the viva, which is `thesis-defense-prep`. Do not use it to decide whether a specific design is identified; that is `identification-defense` and `research-design`. Do not use it to choose a journal, which is `journal-targeting`, though this skill decides whether submitting anywhere is the right use of the next two months. Do not use it as a substitute for the candidate's actual supervisor, who holds the institutional authority, knows the examiners, and signs the forms. Where the advice here conflicts with a real supervisor's instruction, the conflict itself is the thing to surface, in writing, to the supervisor. ## What you need before starting **The real state of every chapter, not the reported state.** For each: written, estimated, planned, or stuck, and if stuck, for how long. Missing: ask chapter by chapter, in those words, and ask specifically what has moved in the last four weeks. A candidate's sense of progress is usually a sense of effort, and effort and progress diverge most sharply on exactly the chapter that is in trouble. **The dates that actually bind.** Submission deadline, funding end, visa expiry, a job start, a supervisor's sabbatical, the institution's examination calendar and its blackout periods. Missing: establish them before advising anything. Advice given against the wrong deadline is worse than no advice, and the binding date is frequently not the one the candidate names first: funding usually ends before the submission deadline, and that gap is the real constraint. **What this institution and this field require for a defensible thesis.** Number of chapters, whether by publication or monograph, whether published papers are required or merely permitted, length conventions, and what the department has actually passed recently. Missing: ask the candidate to find out from the regulations and from two recent theses in the department, which takes an afternoon and frequently reveals that the requirement is lower than the candidate believed. **The candidate's own constraint on capacity.** Teaching load, caring responsibilities, health, a second job, the hours per week that genuinely exist. Missing: ask for hours per week rather than for a description. Plans built on notional full-time availability fail on contact with the actual calendar, and the failure is blamed on the candidate rather than the plan. **What the supervisor has said, in their words.** Missing: ask for the last written feedback verbatim. Supervisory feedback is routinely relayed as harsher or vaguer than it was, and reading the actual words often changes the advice. **Whether the thesis is by publication.** Missing: establish it now, because it changes what is written, in what order, and how expensive a late change of format becomes. ## The method 1. **Establish the real position, chapter by chapter, before saying anything else.** Write it out: what exists, what is estimated but unwritten, what is planned, what is stuck. Include the thesis-level introduction and conclusion, which are almost always omitted from a candidate's mental accounting and are three to six weeks of work. 2. **Compute the time honestly, in weeks of available work rather than in months of calendar.** Take the binding date, subtract the institution's required time for examiner arrangements, subtract known non-working periods, and multiply by the candidate's real weekly hours. The resulting number is usually about sixty percent of what the candidate has been assuming, and seeing it is often the intervention. 3. **State the verdict in the first paragraph.** The current plan fits, fits with named cuts, or does not fit. Where it does not fit, say so before anything else and say what the options are: cut scope, extend, or submit something weaker. Burying a feasibility verdict beneath encouragement is the specific behaviour this skill exists to prevent, and it is the default behaviour of everyone around a doctoral candidate. 4. **Triage every remaining piece of work** into the three categories below. Apply the test, not the feeling: does the thesis fail without this? Most work fails that test, including good work. 5. **Name the cuts out loud, individually, with what each one costs.** Not "narrow the scope" but "the third country comparison is cut, it costs you a paragraph in the introduction and a question at the viva that you will answer by saying it is future work". A cut that is described vaguely does not happen; the candidate keeps working on it in the evenings and it keeps consuming the attention that the load-bearing work needs. 6. **Apply the stop-analysing test to whatever is currently being estimated.** The rule is below. Almost every stuck candidate is stuck on analysis that cannot change the thesis's claim. 7. **Set the writing order** using the sequencing rule below, and put a date on the first piece. A plan without a date on the next action is a description of a plan. 8. **Decide the paper-versus-chapter question for each chapter, now rather than later.** A chapter written as a chapter and converted to a paper in the last four months costs six to ten weeks; the decision made at the start costs nothing. 9. **Set the supervision plan**: what goes to the supervisor, when, and with which two questions attached. Meetings without a position and a decision to be made produce status updates, which consume the supervisor's attention without spending it usefully. 10. **Set a review date** and one measurable checkpoint for it, such as a complete draft of a named chapter. Advice without a checkpoint is advice that will be assessed by feeling in three months. ## The triage categories **Load-bearing.** Without it there is no defensible claim. The main estimate of each chapter, the identification evidence that supports it, the data construction it rests on, the thesis-level argument that connects the chapters. Do this first, protect it from everything else, and do not allow it to be delayed by anything in the other two categories. The most common failure is that load-bearing work is delayed by strengthening work that felt urgent because a seminar audience asked for it. **Strengthening.** Makes an existing claim harder to attack: an additional robustness check, a better standard error, a supplementary data source, a mechanism test. Do these in order of which objection they defuse, and only after the load-bearing work of every chapter is complete. Rank them by asking which question an examiner is most likely to ask; the ones nobody would ask are not strengthening anything. **Cuttable.** The additional country, the second identification strategy, the new dataset, the extension the candidate finds most interesting. Cut it, name it, and write it into a future-work list so it stops occupying attention. The future-work list is not a consolation; it is the mechanism by which a cut actually takes effect, because unresolved ideas return at two in the morning and get worked on. A useful forcing question when the categories are unclear: if the deadline moved six months closer today, what would you drop? Everything named in that answer is cuttable now, and the only thing the six months buys is the ability to do it more comfortably, which is not the same as needing it. ## When to stop analysing and start writing The signal is not that the analysis is complete, because it never is. The rule is: stop when the next specification cannot change what the thesis claims. Apply it as a direct question about the specification currently being run. If this comes out the other way, does the chapter's conclusion change? If the answer is no, it is a table in an appendix and not a reason to delay writing. If the answer is yes, it is load-bearing and should be finished this week rather than iterated for a month. There is a second and less obvious rule. A candidate who cannot start writing usually does not yet know what they are claiming, and the fix for that is not another regression, it is a one-page argument: the question, the answer, the evidence, the objection, the response. Writing that page takes ninety minutes and it either produces the chapter's spine or it exposes that the claim does not exist yet, which is a finding worth ninety minutes. Writing also reveals which analysis is genuinely missing, and it reveals it faster and more cheaply than analysis does. A candidate who writes the results section discovers in two days that the heterogeneity split they have been refining for a month does not appear in the argument at all. ## Sequencing the writing Write the data and results sections first. They constrain everything else, they are the least dependent on decisions not yet made, and they are the parts that cannot be written from imagination, which makes them the honest starting point. Write the conclusion immediately after the results, while the argument is live in the candidate's head. A conclusion written four months later is written from the tables rather than from the reasoning, and it shows. Leave the literature review until the contribution is settled. Written earlier, it is written twice, and the first version is a summary of a field rather than an argument for a gap. Write the introduction last. It is a promise about a paper that has to already exist, and every hour spent on it before the results are stable is spent on a promise that will change. For the thesis-level document: the linking introduction and conclusion are written after the chapters, and they are real work, typically three to six weeks, not the fortnight candidates budget. They are also where an examiner forms their view of whether the thesis is a thesis or a stapled set of papers, which makes them poor candidates for the last week. ## Supervision and committee dynamics Come to a meeting with a position and a decision to be made, never with a status update. Send the material three working days in advance and name the two questions you want answered. A supervisor asked two specific questions answers them; a supervisor sent forty pages with no question responds to whatever they notice first, which is rarely the thing that mattered. Where two supervisors disagree with each other, do not adjudicate silently and do not implement one and hope. Put the disagreement in front of both, in writing, in three sentences, and ask which way to go. Candidates avoid this because it feels like causing trouble; the alternative is doing the work twice and being told at the viva that neither version was right. Where feedback is absent or contradictory, the failure mode is waiting. Set a date, tell them in writing what you will do if you have not heard by then, and then do it. This is not aggression, it is the only mechanism a candidate has, and supervisors generally respect it. Keep a written record of every decision taken in supervision, sent back to the supervisor as a short note after the meeting. This costs ten minutes and it is what prevents the conversation eighteen months later about what was agreed. ## Worked example **Situation.** A fourth-year candidate, Anneke Visser, in a four-chapter economics thesis on labour market effects of vocational training. Chapter one, the introduction, unwritten. Chapter two, complete and presented at two conferences. Chapter three, estimated with a full set of results and about forty percent written, but the candidate had spent five months adding a machine-learning heterogeneity method she had learned in a summer school. Chapter four, a folder of preliminary estimates on a second country, no draft. Funding ended in eight months; the submission deadline was fourteen months away. She worked about 30 hours a week on the thesis, teaching two courses. Her question when she came was which heterogeneity method to use in chapter three. **Task.** Answer the real question, which was whether there was a thesis at the end of this, and produce an ordered plan. **Action.** The position was written out first and it changed the conversation immediately. Available work: eight months to funding end at roughly 30 hours a week, minus a five-week teaching block and three weeks of institutional closure, gave about 28 working weeks before the income stopped. The candidate had been mentally working to fourteen months. Seeing 28 weeks on the page was the moment the plan changed. The verdict came first: the four-chapter plan did not fit. Chapter four did not exist, the thesis-level introduction and conclusion had not been budgeted at all, and the chapter three heterogeneity work was consuming the largest share of the calendar. The triage was uncomfortable. Chapter two: load-bearing, done, protect it, do not reopen it. Chapter three's core estimates: load-bearing, done, needs writing. Chapter three's machine-learning heterogeneity: applying the stop-analysing test, the question was whether the chapter's conclusion changed if the method returned nothing. It did not; the chapter's claim was about the average effect and the heterogeneity was an extension. Cuttable, at the cost of one interesting paragraph and possibly one viva question. Chapter four: this was the hard one. **The wrong turn.** The first plan kept chapter four and cut the heterogeneity work, on the reasoning that a four-chapter thesis was the agreed plan and that three chapters would look thin. The plan was built, and it required chapter four to be estimated, written and revised in eleven weeks, alongside writing chapter three, writing the introduction and conclusion, and teaching. It was abandoned when the weeks were added up: 34 weeks of work against 28 weeks of funded time, with no allowance for anything going wrong, and something always goes wrong in the final months. What made the abandonment possible was checking the actual requirement rather than the assumed one. The department's regulations required three substantive chapters. Anneke had believed four because her cohort had all done four. Two recent theses from the department, checked in an afternoon, had three. The revised plan: three chapters. Chapter four cut entirely, with its estimates written up as a two-page future work note and retained as the basis of a postdoctoral application, which is what made the cut survivable rather than merely painful. The heterogeneity work cut, with the code archived. Writing order: finish chapter three, six weeks; conclusion of chapter three immediately after, one week; thesis-level introduction and conclusion, five weeks; revision against `thesis-chapter-review` findings, four weeks; buffer, six weeks. That left slack, which is the point. The supervision plan: a three-sentence note to both supervisors stating the change from four chapters to three, the regulation it relied on, and a request for written confirmation. The written confirmation is what makes a scope cut real; a scope cut agreed verbally is reopened at the pre-submission meeting. **Result.** Submitted five weeks before the funding ended, at about month seven of the eight, which was roughly seven months inside the original fourteen-month deadline and, more importantly, while she was still being paid. Three chapters. The viva asked one question about whether a second country would have strengthened the external validity claim, which was answered in about forty seconds by saying yes and that it was the next paper. Chapter four's material became a working paper the following year, which it would not have done had it been rushed into the thesis. Anneke's own account afterwards was that the decisive step was counting the weeks. Everything after that followed. ### A second scenario, where it goes differently A candidate in a humanities monograph thesis with twenty-two months remaining, funded throughout, no teaching, and a supervisor who has said the current draft is publishable. The plan fits, comfortably, and the honest verdict in the first paragraph is that there is no feasibility problem. The advice inverts. Where the constrained candidate is told what to cut, this one is told what to add, and the triage runs on ambition rather than survival: which additional work would most improve the thesis's standing at examination and in the job market, given that the time exists. The stop-analysing test still applies, but the answer changes: a specification that cannot change the claim may still be worth running here, because it defuses an examiner's question and there is time. Two things must not change. The sequencing rule still holds, because writing the introduction early wastes the same weeks whether or not there are weeks to waste. And the verdict is still stated first, even when it is good news, because a candidate who is told they are fine without being told why will not believe it, and a specific "you are eleven months ahead of where you need to be, here is the calculation" is what actually reduces the anxiety. The risk in this scenario is the opposite one: scope expanding to fill the available time until, at month nineteen, the comfortable position has become a tight one. The countermeasure is the review date and the checkpoint, applied exactly as in the constrained case. ## Output Open with the verdict in one paragraph. **The position** | Chapter | State | Last movement | Load-bearing work outstanding | Weeks estimated | **The time** ``` Binding date: [date, and which constraint it is] Institutional lead time: [weeks for examiner arrangements and deposit] Non-working periods: [weeks] Real weekly hours: [hours] Available working weeks: [number] Work required, summed: [number] Verdict: [fits / fits with the cuts below / does not fit] ``` **The triage** | Item | Category | Reason | If cut, what it costs | **The plan** | Order | Work | Owner | Start | Done by | Depends on | **The cuts**, listed individually, each with its cost and where it is recorded as future work. **The supervision note**: two or three sentences to be sent, naming the change, the reason, and the confirmation requested. **Review date** and the one checkpoint that will be assessed on it. ## Failure modes **The verdict buried at the end.** Recognise it when the first three paragraphs are about the work and not about whether the plan fits. Move it to the top, always. **Advice that adds without subtracting.** Recognise it when a recommendation does not name what is dropped to make room. Every addition displaces something; say what. **A vague cut.** Recognise it when the cut is described as narrowing the scope. Name the specific piece of work, say it is cut, and say what it costs. **Months rather than weeks of available work.** Recognise it when the plan is expressed in months. Convert to available working weeks at the candidate's real hours; the conversion is usually the finding. **Effort mistaken for progress.** Recognise it when a chapter has consumed four months and nothing has changed about what it claims. Apply the stop-analysing test. **Waiting on a supervisor.** Recognise it when the plan's next action depends on feedback that has not arrived and no date has been set. Set the date and state what happens if it passes. **The assumed requirement.** Recognise it when the scope is justified by what the candidate's cohort did rather than by the regulations. Check the regulations and two recent theses. **Encouragement used as a substitute for a decision.** Recognise it when a plan contains reassurance and no dates. Encouragement is fine and must be attached to something specific and must not stand in for the verdict. **Advising against the wrong deadline.** Recognise it when funding end and submission deadline have not been separated. The binding date is almost always the earlier one. ## Edge cases **The candidate should probably not finish.** Rare, and it happens: health, a change of direction, a project that cannot be rescued in the time or the money available. Say it plainly and neutrally, name the exit options the institution actually offers, including a lesser award where one exists, and say that leaving is a legitimate decision rather than a failure. Do not deliver this as a recommendation; deliver it as an option the candidate is entitled to consider. **The supervisor is absent, hostile, or has left.** Escalate through the institution's structure: the second supervisor, the postgraduate director, the doctoral school. Document everything in writing from now on. This is one of the few situations where the advice is procedural rather than academic, and candidates routinely delay it by a year out of politeness. **A result fails late.** An identification strategy that collapses in month twenty. Triage it as a new project with the time remaining: can the chapter be rewritten around a descriptive or partially identified claim that is honest and defensible? Very often it can, and a candid chapter about what the data cannot establish is examinable. What is not examinable is a chapter that claims what the design no longer supports. **Thesis by publication with a paper stuck in review.** Do not let an editor's calendar control a submission date. Most institutions accept a submitted or working-paper version as a chapter. Check the regulation, plan on the chapter version, and treat acceptance as a bonus. **The candidate is far ahead and being pushed to add more.** Protect the finish. Additional work should be a decision made against the checkpoint, not a drift. Where a supervisor is pushing for more, ask for the specific examination risk it addresses; if there is not one, it is a paper, and papers can be written after submission. **Two supervisors want different theses.** Put the disagreement in front of both in writing, in three sentences, with a request for a decision by a date. Do not implement either while waiting. **Coauthored chapters.** Establish the contribution statement and the coauthors' consent to inclusion early, in writing, because a coauthor who becomes unreachable in the final month is a real and recurring problem. Check the institution's rule on how much of a coauthored chapter counts. **The candidate asks for reassurance rather than advice.** Give the verdict anyway, then give the reassurance, attached to something specific they have actually done. Reassurance that replaces the verdict is the failure this skill exists to prevent. ## Quality bar - The feasibility verdict appears in the first paragraph, including when the answer is that the plan does not fit. - Available time is computed in working weeks at the candidate's real hours, and the arithmetic is shown. - Every remaining piece of work is assigned to load-bearing, strengthening, or cuttable, by the test rather than by feeling. - Every cut is named individually with what it costs and where it is recorded as future work. - Every recommendation names what is dropped to make room for it. - The plan has a dated next action and one measurable checkpoint with a review date. - The load-bearing work is scheduled first and is protected from the rest. - Any encouragement is attached to something specific and does not stand in for the verdict. ## Adapting this to your context The worked example is a four-chapter economics thesis, thirty hours a week, funding ending before the deadline. The counting and the verdict are the method; the shape of the thesis is not. - **The thesis shape.** Three or four empirical chapters is one convention. Psychology and health doctorates often run three studies inside a general introduction and discussion; a monograph carries a literature review chapter; Dutch and Nordic theses staple published articles. Read your regulations and two recent theses from your department. - **What is cuttable.** Here, a heterogeneity extension and a second country. In lab work, a fourth study; in qualitative work a second site, harder because the saturation argument leans on it; in a trial a follow-up wave, which cannot be cut late without a protocol amendment. - **The work nobody budgets.** Three to six weeks for the thesis introduction and conclusion is the economics version. Add yours: transcription at four to six hours per interview hour, translation, ethics amendments, data access queues measured in months. - **The hours figure.** Thirty a week with teaching is one case. Part-time, clinical and professional doctorates commonly run ten to fifteen, which changes every arithmetic here. - **What not to change.** Count weeks of available work rather than months of calendar, put the verdict in the first paragraph, and make every recommendation name what is dropped for it. ## Related skills `thesis-chapter-review` reads the same chapters as an examiner would and will often ask for more than this skill will allow; the two views are reconciled by triaging its findings into the three categories here. `thesis-defense-prep` takes over once submission is decided. `research-design` is what a stuck chapter usually needs instead of another specification. `identification-defense` answers whether a design is worth defending before weeks are spent strengthening it. `analysis-audit` should be scheduled as load-bearing work before submission, not treated as optional polish. `replication-package` closes the project and is a deposit requirement at many institutions, so it belongs in the plan rather than in the week after submission. `journal-targeting` and `peer-review-simulator` handle the paper track for a thesis by publication. `full-manuscript-build` assembles a chapter once the decision to write it has been made.