agentleFS
169884902hzl169884902hzl/engineering-paper-skills/skills/engineering-writing/SKILL.md

engineering-writing

Skill3 starsChanged 40 days agoMITView on GitHub

Draft, restructure, or plan English engineering research paper sections from author-provided claims, methods, results, figures, notes, outlines, LaTeX files, or manuscript drafts. Use for section logic, contribution-evidence mapping, abstract/introduction/related-work/methods/experiments/discussion/conclusion drafting, and evidence-bounded manuscript prose. Do not use for pure language polishing, figure/table-only work, reviewer responses, or final build/readiness validation unless the user explicitly asks for those workflows.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 6 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill to write the paper's argument from the author's evidence.

    line 8
  2. Distinctive:

    Never invent experiments, results, numbers, baselines, citations, datasets, or completed work.

    line 15
  3. Very specific:

    If a fact that would change the scientific claim is missing, mark a concise placeholder ([RESULT NEEDED], [CITATION NEEDED], [VALUE?]) or …

    line 17
  4. Distinctive:

    Use mechanisms the author supplies.

    line 20
  5. Very specific:

    Use engineering-polishing for wording once the section logic is stable.

    line 30
  6. Very specific:

    Use engineering-figure-table for captions, table design, and visual roles.

    line 31
  7. Very specific:

    Use engineering-response for reviewer, editor, or co-author comments.

    line 32
  8. Very specific:

    Use engineering-paper-auditor for a pre-submission reviewer simulation.

    line 33
  9. Very specific:

    Use engineering-validation for builds, references, and readiness claims.

    line 34
  10. Distinctive:

    Check P1 (handicapped baseline), P2 (unstated operating assumption), P3 (unmeasured internal component), and P4 (added cost versus a …

    line 45
  11. Distinctive:

    WRITE (draft, compose, turn notes or results into prose): return usable manuscript prose first.

    line 90
  12. Distinctive:

    Follow the user's requested length and format.

    line 99

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Engineering Writing

Use this skill to write the paper's argument from the author's evidence. Most of what makes a strong engineering paper is decided before prose: which claim, which comparison, which table, which figure. This skill puts that experience first and leaves sentence craft to the model.

Hard Floor

  • Never invent experiments, results, numbers, baselines, citations, datasets, or completed work. Planned work stays planned.
  • If a fact that would change the scientific claim is missing, mark a concise placeholder ([RESULT NEEDED], [CITATION NEEDED], [VALUE?]) or ask, and keep writing the parts that are supported.
  • Use mechanisms the author supplies. Do not add a physical or causal mechanism the author did not give.
  • For an existing paper, read the live source and local project instructions before rewriting. Preserve LaTeX labels, citations, macros, and terminology.
  • Final manuscript prose defaults to English. Non-English notes are source material; rebuild the argument in English rather than translating sentence by sentence.

Boundaries

  • Use engineering-polishing for wording once the section logic is stable.
  • Use engineering-figure-table for captions, table design, and visual roles.
  • Use engineering-response for reviewer, editor, or co-author comments.
  • Use engineering-paper-auditor for a pre-submission reviewer simulation.
  • Use engineering-validation for builds, references, and readiness claims.

Experience First

  • What to write first, and how to write while results are incomplete: writing-process.md. In short: lock one sentence, allocate the page budget, design result tables before the remaining experiments, design figures from the tables, then write Methods, Experiments, Introduction, and the Abstract last.
  • What reviewers will attack, so the draft can pre-empt it: reviewer-attack-patterns.md. Check P1 (handicapped baseline), P2 (unstated operating assumption), P3 (unmeasured internal component), and P4 (added cost versus a simpler alternative) whenever drafting Introduction, Methods, or Experiments.
  • How senior authors restructure a draft: revision-moves.md.

Section Notes

  • Abstract: the concrete failure mode, the method core, the evaluation scope, and the strongest numbers. Write it last.
  • Introduction: task difficulty, the core bottleneck, why existing routes fall short, the design requirement that follows, and the proposed formulation. Let the problem drive the story, not the tool. State the key idea at principle level; keep mechanism detail for Methods. Each contribution is a testable sentence; their number follows the independent, supported advances.
  • Related Work: organize by technical axis; give the nearest neighbors the detail the positioning needs; state only differences and limitations the sources support.
  • Methods: a reader path (input, intermediate object, transformation, handoff, execution and failure handling), not a module directory. Say how every decision rule is computed; label heuristics as heuristics.
  • Results: ranking, key number, the interpretation the evidence supports, per-category results, and the operating boundary. Turn numbers into findings (bottleneck, stage attribution, method-family verdict, closing lesson; see "From Numbers To Findings" in revision-moves.md). Present an unmeasured mechanism as an interpretation, not a cause. Keep setup short and parameters in a table.
  • Ablation: relate each major delta to a component role or failure mode, and remember that cumulative ablations depend on the order of addition.
  • Robustness and failure: the stress axis, the trend, the route by which the factor affects the method, what still works, and the hardest regime.
  • Conclusion: the method, the strongest evidence, a bounded takeaway, and future work derived from the observed failure regime.

Scope belongs in the prose as a concrete operating condition, not as a disclaimer. Keep it wherever a passage is read on its own (Abstract, headline result, Conclusion) and avoid repeating it in adjacent sentences. Process language ("the supplied notes show", "pending verification") and audit language ("without claiming") do not belong in manuscript prose.

Request Modes

  • WRITE (draft, compose, turn notes or results into prose): return usable manuscript prose first. Add a short note after it only for placeholders, missing evidence, or claims the author should check.
  • PLAN (outline, paper plan, contribution design, writing order): start from the one-sentence thesis, the section job map, the table slots, and the contribution-evidence map.
  • AUDIT: hand off to engineering-paper-auditor.
  • POLISH: hand off to engineering-polishing.

Follow the user's requested length and format. If the user asks for prose only, return prose only.

When to Open Extra Files

FileOpen when
references/paper-workflow.mdStarting from scratch, rebuilding a draft, or deciding paper-writing order
references/contribution-evidence.mdDefining the paper thesis, contributions, claim boundaries, or evidence map
references/title.mdDrafting or auditing the title
references/abstract.mdDrafting or revising the abstract
references/introduction.mdDrafting Introduction, positioning, gap, formulation, or contribution list
references/related-work.mdDrafting Related Work, taxonomy, nearest-neighbor distinction, or folded literature positioning
references/methods.mdWriting Methods, overview, formulas, system roles, algorithms, execution, or safety/deployment logic
references/methods-worksheet.mdA Methods draft risks becoming a module directory or formula dump
references/experiments.mdPlanning or writing Experiments/Results, baselines, metrics, main results, ablations, failure analysis
references/experiments-worksheet.mdResults need setup, baseline, metric, ablation, category, stress, or failure-envelope structure
references/discussion.mdWriting Discussion, limitations, interpretation, or implications
references/conclusion.mdWriting a bounded conclusion and future work
references/section-boundaries.mdAuditing whether content belongs in the right section or when a section is drifting
references/section-budget.mdChecking whether a section is too thin, too dense, or taking space from evidence
references/page-budget-war-plan.mdCutting manuscript length without damaging evidence anchors
references/source-learning.mdLearning structure from 3-5 neighboring papers without copying wording or surface format
references/ral-style-writing-guide.mdThe user explicitly wants RA-L-style section skeletons or sentence patterns
references/positive-patterns.mdChoosing a paper-level or section-level pattern before drafting, or a draft reads structurally generic
references/bad-sentence-repairs.mdRepairing common bad manuscript sentences and section-level failure symptoms
manifest.yamlLooking up which references belong to a section or request type
references/examples.mdNeeding concrete prompt and output behavior examples
references/failure-modes.mdHandling thin evidence, invented-citation requests, or overclaim pressure
references/venue-aware-writing.mdTarget venue family may change abstract, limitation, reproducibility, or contribution framing
references/paper-level-narrative-map.mdFull-paper or multi-section writing needs story dependency checks
../_shared/writing-process.mdDeciding what to write first, or writing while results are incomplete
../_shared/reviewer-attack-patterns.mdDrafting Introduction, Methods, or Experiments that a reviewer will probe
../_shared/revision-moves.mdRestructuring an existing draft
../_shared/story-spine.mdBuilding problem-to-evidence-to-boundary logic before drafting
../_shared/evidence-boundary.mdAny task asks for stronger claims, missing evidence, or manuscript facts
../_shared/citation-boundary.mdRelated Work or citations are requested without provided sources
../_shared/citation-verification-workflow.mdThe user supplies externally discovered citations (deep research, scholarly APIs) that need a verification gate before Related Work positioning
../_shared/claim-strength.mdCalibrating verbs, novelty, robustness, generalization, or causal language
../_shared/ai-assisted-writing-policy.mdAI-assisted prose or venue disclosure risk is relevant
../_shared/list-to-argument.mdSource material is a bullet list, module list, result-row list, or contribution list
../_shared/non-english-source-notes.mdNon-English notes must become English manuscript prose
../_shared/output-mode.mdThe user asks for output only
../_shared/sentence-role-and-story-flow.mdDrafting, shortening, or reordering paragraphs where every sentence must justify its role
../_shared/terminology-ledger.mdA writing task may rename methods, metrics, categories, or baselines

Open only what the task needs; none of these files is mandatory.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.