More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Q&A Prep: Hard Questions Answered Before They're Asked

    1

    They go where the material is exposed: a headline stronger than its estimate, a test on the wrong unit, a sample nobody explained, a choice that invites an alternative explanation.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Q&A Prep: Hard Questions Answered Before They're Asked

    Q&A Prep: Hard Questions Answered Before They're Asked

    Example session with this skill installed

    I'm giving a 12-minute talk on this study at a regional agroecology conference next week, with 5 minutes of questions. The session chair is a pollination ecologist and the room will be mostly ecologists, plus some agronomists and farm advisers. My draft paper is paper.md, my slide outline is talk-outline.md, and the count data are in data/. Can you prepare me for the Q&A? I want to know what the hardest questions will be and have answers ready that I can actually defend.

    • Read your context and instructions
    • Compiled the q&a prep
    • Generated the data export

    I've written the Q&A prep sheet. The hardest questions don't come from your data. They come from four lines on your slides that claim more than the results show, and fixing those before the talk removes most of the risk. Your counts are sound: recomputing from data/visits.csv gives exactly the paper's 8.2 vs 5.1, the per-farm ratios, and 1.6× (95% CI 1.2–2.1).

    To predict the questions, four separate readers saw only your paper, slides and data: a pollination ecologist acting as chair, a field statistician, an agronomist and a farm adviser.

    The three most dangerous questions, with answers you can defend:
    1.

    "You say 'doubled', but your own estimate is 1.6×. Why?" All four readers raised this. Concede it immediately: "Doubled overstates it. It's 1.6 times, 95% CI 1.2 to 2.1, and higher on all eight farms."
    2. "Your test counts 160 counts as independent, but your replicate is the farm." Agree. The 1.6× and its interval are already calculated across farms; only the p-value on slide 6 uses the wrong unit. Rerun the test with farm as the unit before quoting it.
    3. "Which crop got the 6% yield gain, and does it even need insect pollination?" Nothing in your material names the crop. Say plainly that this is your weakest number: three farmers who chose to weigh separately, with no spread or test.

    To fix before the talk (suggested wording is in the sheet; I haven't edited your slides or paper):

    • "Doubled": change it to "1.6 times" in the title and on slides 1 and 6.
    • The test: rerun it with farm as the unit (or a mixed model) and replace the n = 160 test on slide 6.

    The ninth farm: the abstract says nine farms, the results analyse eight, and nothing explains the drop. Only you know the reason, so add one line to slide 4.

    • Yield: name the crop, then soften slide 8 to "indicative" or move it to a backup slide.

    Walk order: every strip and control pair in the data has the same start time. Check your field notes for whether one person walked both margins, and which came first.

    Concede these when asked: farmers chose where to put the strips; there are no counts from before sowing; flowers in the margins weren't measured.

    The sheet also has 18 ranked questions, 11 drafted answers, four backup slides built only from your existing material, scripts for the awkward moments and a card to glance at during the Q&A. The number check passes.

    visits.csv

    CSV · data export

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    What it does

    The questions after a research talk aren't random. They go where the material is exposed: a headline stronger than its estimate, a test on the wrong unit, a sample nobody explained, a choice that invites an alternative explanation. You can't see those spots in your own work, because you know what you meant. This skill makes your agent find them first, from your paper, slides, poster or results, and prepare answers you can defend in the room.

    • Every claim against its evidence. Each claim a listener could challenge is listed with its number, n, uncertainty and location, and marked supported, stretched or unsupported. Numbers that differ between abstract, slides and paper are caught.
    • Questions predicted, then ranked. From the exposed points, a list of research question angles (design, alternative explanations, robustness, scope, prior work, definitions, practical value) and the questions your setting brings, plus fresh questioner subagents (say, the chair, a statistician, a practitioner) that see only your material. Fifteen or more questions (ten for a poster session), ranked by how likely and how damaging they are.
    • Answers that stay inside your material. The top ten and every damaging one, each leading with the direct answer, then the evidence, then the limit you concede before the questioner does, with a Sources line for every fact. When an answer needs something your material doesn't contain, it gets a [GAP] note for you, never an invented number or citation.
    • A number check that runs. A bundled script finds every number in the answers in your material, flags any it can't, and verifies declared derivations (derived: 3.1 = 8.2 - 5.1).
    • What to fix before you walk on. Gaps sorted into fix-before-the-talk (often a slide line that overclaims) and acknowledge-in-the-answer, three to five backup slides built only from material you have, scripts for hard moments (a question that hits a real flaw, one you can't answer, a hostile questioner) and a one-page lectern card.
    • A mock Q&A, if you want one. One question at a time; feedback on whether you answered first, stayed in time, kept to the evidence and conceded cleanly.

    What a run looks like

    In our demo the agent got a made-up draft paper, slide outline and count data for a field study of wildflower strips, with six weak spots planted. It found all six: "doubled" against an estimate of 1.6, a test on 160 counts when the farm is the unit, a ninth farm that vanished, a yield claim from three farms, farmer-chosen placement, and one season of dry mornings. It also caught one we hadn't planned (strip and control logged with identical start times). It recomputed the numbers from the raw data and put farm-level statistics the paper doesn't report inside [GAP] notes for the speaker to confirm.

    Settings

    Conference talks, poster sessions, thesis defences and vivas, job talks, grant panel interviews, lab meetings and progress reviews, and written reviewer or committee Q&A.

    What's included

    • SKILL.md: the nine-step method and its integrity rules
    • references/: question angles and setting-specific questions, questioner personas and prompt, hard moments, the prep sheet format
    • scripts/check_numbers.py: the number check (Python standard library)
    • examples/: the demo's material and the prep sheet it produced

    Part of the Aurora Lab research pipeline

    Works on its own, on any talk. Pairs well with Aurora Lab's Poster Layout and other talk skills; needs none of them.

    It prepares you from the material you give it: it can't know facts that aren't in it (the reason a sample was dropped, a crop, an examiner's views, the literature you haven't cited), so those become [GAP] notes for you. Questioners are simulated: they predict likely questions well, but a real room will still surprise you, and a run-through with a colleague remains worth doing. The number check verifies provenance, not meaning: it confirms each number appears in your material or follows from a declared derivation, so a small integer that happens to appear anywhere in the material passes; overclaiming in words is caught by the agent's re-read, not by the script. It reads text: PDFs and slide files need a text extraction (the agent does this when a tool such as pdftotext is available). The mock Q&A is text-based unless your agent can take voice. It suggests wording for your slides but never edits them. Without subagents the questioners are played in one context, which is less independent; the prep sheet says so.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 27 days ago

    Requires

    Python 3.10+
    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    File Scopes

    **/*.md
    **/*.tex
    **/*.txt
    **/*.csv
    research/results/**
    research/talks/**

    Reads the material you point to: paper, slides, poster, abstract, results files, reviewer comments (with a research/ folder: research/claims.md, research/results/, research/paper/, figure captions in research/figures/). Writes one prep sheet, in research/talks/ when a research/ folder exists, otherwise next to the material. Terminal: runs the bundled number check (Python standard library) and, when your material is a PDF, a text extraction such as pdftotext into a temporary file that is deleted at the end. No network access. Never edits your slides, paper or results.

    Listed29 days ago
    Updated27 days ago

    What's inside

    Frequently Asked Questions

    Any agent that loads SKILL.md skills, including Claude Code, Codex CLI, Cursor and Gemini CLI. It works best where the agent can run subagents or start a fresh session, so each questioner sees only your material. Where it can't, the agent plays each questioner in a separate pass and says so in the prep sheet.

    No. Every number, result and citation in an answer must come from your material, with its location in a Sources line. Where a good answer needs something you haven't given it, the answer says what is true without it and a [GAP] note tells you what to supply. A bundled script then checks every number in the answers against your material.

    That each number in the answers, backup slides and lectern card appears in your material (allowing rounding and fractions written as percentages), and that any declared derivation, such as a difference of two reported means, is arithmetically right. It checks provenance, not meaning: wording that claims too much is caught by the agent's re-read against the evidence map.

    Whatever you have: a paper or draft, a slide outline or deck, a poster, an abstract, results files or reviewer comments. Text formats work directly; PDFs and slide files are converted to text first when a tool such as pdftotext is available.

    Yes. It knows the questions each setting brings: for a defence, the contribution in one sentence, the main limitations, what you'd do differently, your role in collaborative chapters; for a job talk, your research programme and independence. Tell it your examiners' fields and anything you know about their work; it never guesses an examiner's views.

    Yes. Ask for a mock Q&A: it asks one question at a time, switching questioners and getting harder, and after each answer tells you whether you answered first, kept to time, stayed inside the evidence and conceded limits cleanly. It's text-based unless your agent can take voice.

    No. It suggests wording for the lines that will draw hard questions (in our demo, 'doubled' against an estimate of 1.6) and leaves the edits to you.

    Our demo, with four questioner subagents on a short paper, a slide outline and a data file, took about four minutes. Each questioner is a separate agent run, so it uses several times the tokens of a single pass.

    New in Research & Analysis