- Home
- Skills
- Writing & Content
- Paper Write: LaTeX Drafting from Outlines
More screenshots
Works with the AI Agents you already use
Paper Write: LaTeX Drafting from Outlines
Draft LaTeX papers evidence-first: assemble a claims-evidence matrix before prose, write results before the introduction, verify citations, and mark data gaps honestly instead of fabricating them.
$5
Paper Write: LaTeX Drafting from Outlines
Example session with this skill installed
I have an outline for a 9-page NLP conference paper on retrieval-augmented QA. The main results table is done — my method wins by 3.2 points averaged over seeds. The ablation over retriever depth hasn't been run yet, and two of my related-work citations are ones I typed from memory. Deadline is in three weeks. What do I write, in what order, and what do I do about the gaps?
- Read your context and instructions
- Compiled the paper write
Assembly gate first — you do not pass it yet, and that determines the order of work.
Two gate failures to fix before or during drafting: (1) the retriever-depth ablation is missing. It goes in the draft as a DATA_NEEDED marker, worded exactly: the experiment implied by the outline, no runs yet, claim held. Until it runs, your paper may claim the 3.2-point win ("shows" — direct measurement, multiple seeds) but may not attribute the gain to retrieval depth; attribution wording stays out of the prose entirely, not softened to "suggests," because you have no evidence either way. (2) The two from-memory citations get verified before the related-work section is drafted — titles, authors, venue, year, and that the cited version supports the sentence attached to it. If either fails verification, it is replaced or the sentence is cut. From-memory citations are how fabricated references get into camera-readies.
Matrix excerpt: C1 "method beats baselines by 3.2 points" — Table 1 — supported. C2 "gain comes from retriever depth" — ablation — DATA_NEEDED. C3 (background claims) — cited work — pending verification.
Drafting order for your three weeks: Methods this week, written from the actual training and evaluation code (seeds, splits, retriever settings — the boring details are what reviewers request). Results next: main table first, existing ablations ordered by what they rule out. Run the depth ablation in parallel with drafting, not after — if it lands, C2 enters with real wording; if it doesn't run by the marker sweep, C2 is cut and the paper stands on C1. Related Work after Results, organized by approach family. Introduction in the final week, written from the finished Results — including the limitation, stated in the preview paragraph. Abstract last, in the final pass.
Final gate before submission: grep DATA_NEEDED returns nothing, the drift check confirms 3.2 reads identically in abstract, intro, results, and conclusion, and the readiness rubric scores 9/12+ with no zero on claim support.
Connects securely to your tools. The creator never sees your data.
About this skill
Most weak paper drafts share one defect: the prose was written first and the evidence retrofitted, so the introduction promises more than the results deliver and the citations were typed from memory. This skill runs the drafting process in the order that prevents fiction. An assembly gate comes first — an outline with a claims-evidence matrix mapping every planned claim to the figure, table, or log that supports it; figures finalized or marked draft; a bibliography in which every entry is verified as real; and the venue's constraints known before they become deadline surgery. The drafting protocol then writes Methods from the code and logs (not from memory of the plan), Results as plain description, Discussion with limitations stated early and negative results reported, Related Work organized by approach family, the Introduction once the evidence's story is known, and the Abstract last of all — a summary, not a promise. Supporting machinery includes the matrix template, a claim-strength decision table that ties verbs to evidence ("shows" for direct multi-setting measurement, "suggests" for indirect support, silence for unmeasured claims), citation hygiene rules, and the DATA_NEEDED marker convention: searchable, precise, and never containing an invented number — each marker is filled with real results before submission or its claim is cut. A worked example shows an introduction written from finished results with the limitation in the preview paragraph, plus edge cases (results changing late, page-limit surgery order, negative headline results, rebuttal revisions) and a consistency-pass quality gate.
- It will not invent results, citations, or numbers: gaps become markers and cut claims, which is the point, but it means the skill can't rescue a paper whose experiments aren't done.
- Venue specifics (page limits, anonymity, reference style) must be supplied by you; the skill enforces them once known.
- The drafting order assumes a results-driven paper (systems, ML, empirical work); a purely theoretical paper needs the matrix adapted to proofs and lemmas.
- It improves honesty and structure; it cannot make a weak result strong, and its rubric will say so.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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 24 days ago
Permissions
No special permissions declared or detected