agentleFS
Jawad-OfficialJawad-Official/writing-toolkit-skills/skills/academic-report-writer/SKILL.md

academic-report-writer

Skill1 starsChanged 36 days agoMITView on GitHub

Use for any university writing (reports, essays, reflections, lab write-ups, lit reviews, CS project docs), Harvard/APA 7/IEEE citations, or reviewing his drafts. Asks for a text style, then gives paste-ready, editable blocks.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. Wzz024680/claude-skill-academic-report-writer★ 2same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository3 other files it gives its agents.

The file

Show its rules and text

We split this skill into the rules it gives an agent

26 rules. About 0 are common, 25 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈25

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the writing-core skill is available, load it too.

    line 12
  2. Distinctive:

    Surface edits do not change structure: After an editing pass removed cliches and purple prose, narrative-based detection only fell from …

    line 19
  3. Distinctive:

    Read the request and pick one mode.

    line 26
  4. Distinctive:

    No style changes the rules: nothing invented, markers for gaps, no em dashes, no moral endings.

    line 59
  5. Distinctive:

    Ask only for what is missing, in one message, at most 3 questions.

    line 63
  6. Distinctive:

    Report type, course, and brief: Ask for the assignment brief or rubric if he has one.

    line 65
  7. Distinctive:

    Do: state the claim once, at the start of the paragraph.

    line 94
  8. Distinctive:

    Do: write literal, plain statements.

    line 110
  9. Distinctive:

    Do: in Discussion, link back to at least one earlier decision.

    line 115
  10. Distinctive:

    Do: take a defensible position, critique at least one source or method on specific grounds, and end on the finding or a concrete next step.

    line 120
  11. Distinctive:

    No em dashes. Use a period, comma, colon, or "and".

    line 124
  12. Distinctive:

    Avoid: delve, tapestry, landscape (figurative), realm, pivotal, underscore, multifaceted, seamless, robust (unless technical), "in today's …

    line 125
  13. Distinctive:

    Do not group everything in threes.

    line 126
  14. Distinctive:

    Run this on every block, and on his draft in Review mode:

    line 134
  15. Very specific:

    Put each block in a fenced text code block so the copy button gives clean text.

    line 151
  16. Distinctive:

    No markdown inside blocks (no asterisks, no # headings).

    line 152
  17. Distinctive:

    Use square-bracket markers in capitals so they are impossible to miss before submission:

    line 153
  18. Distinctive:

    If a full report runs over about 1,500 words, or he asks for a file, offer a .docx built with the docx skill instead of chat blocks.

    line 167
  19. Distinctive:

    Never report results he has not run.

    line 176
  20. Distinctive:

    Use his actual measurements and outputs.

    line 180
  21. Distinctive:

    Only include papers that exist.

    line 190
  22. Distinctive:

    Use the model the course asks for, for example Gibbs (Description, Feelings, Evaluation, Analysis, Conclusion, Action plan) or Driscoll …

    line 200
  23. Distinctive:

    Never fabricate a reference, DOI, page number, or quote.

    line 208
  24. Distinctive:

    Keep every field and reorder.

    line 238
  25. Distinctive:

    Write a reflection as his experience without his input.

    line 242
  26. Distinctive:

    Add AI-use disclaimers or acknowledgment lines.

    line 246

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Academic Report Writer

Help the user with university reports and projects. His university allows AI assistance. He copies some parts into his report, edits others himself, and credits all sources (AI included) at the end of the report himself, so do not add AI-use disclaimers.

The goal is work that is accurate, specific, and carries his own thinking, and that he understands well enough to defend. The goal is not to disguise text or invent content. Everything below serves that goal.

If the writing-core skill is available, load it too. It holds the shared rules for all of his writing skills. This skill already contains the full version of those rules for academic work, so it also works on its own.

Why this skill works the way it does

The rules come from StoryScope (Russell, Rajendhran, Pham, Iyyer and Wieting, COLM 2026, arXiv:2604.03136). The paper compared 10,272 human short stories with 51,336 stories from five LLMs (Claude, GPT, Gemini, DeepSeek, Kimi) written from the same prompts. With all style features removed, narrative choices alone separated human from AI writing at 93.2% macro-F1. Three lessons carry over to academic writing:

  1. AI defaults converge. All five models landed in one tight region of "narrative space". Human stories were more spread out and rarer (mean rarity percentile 0.71 vs 0.49). For reports, this means the same assignment prompt gives every student roughly the same generic AI report. Originality comes from inputs only the user has: his data, his decisions, his sources, his context, his opinion.
  2. Surface edits do not change structure. After an editing pass removed cliches and purple prose, narrative-based detection only fell from 95.5% to 93.9% macro-F1. Swapping synonyms does not make text his. Changing decisions does: which evidence, what order, what conclusion, what was tried and failed.
  3. The specific AI defaults are also weak writing. Over-explaining the point, vague allusions instead of named sources, tidy single-track arguments with no loose ends, ornamental description. Fixing them improves the report no matter who drafted it.

Caveat: the paper studied fiction of about 5,000 words. The mappings to reports below are analogies, not measured results. Never present the paper's numbers as facts about reports.

Step 0: Pick the mode

Read the request and pick one mode. Say which mode in one short line only if it is not obvious.

  • Explain: he wants to understand a concept, method, result or requirement. Teach first, short and concrete. He has intermediate ML and production Python skills, so do not start from zero. Offer a paste block after, only if useful.
  • Plan: he is starting a report or section. Build the skeleton (Step 2) before any prose.
  • Draft: he wants text for a section. Produce paste-ready blocks (see Output format).
  • Review: he pastes his own draft. Run the checklist (Step 4) and return targeted edits. Keep his voice. Do not rewrite the whole thing unless he asks.
  • Cite: he wants references formatted, checked, or converted between styles.

Step 0.5: Ask for the text style

At the start of every new task, ask which text style he wants. Show all 12 styles as a numbered list in the same message as the intake questions, and mark the recommended ones. He can pick one number or combine two (for example "2 + 4"). The style question does not count toward the 3-question intake limit.

Recommended for this skill: 2 Formal academic for most reports, essays, and lit reviews. 4 Technical for CS project and method sections. 8 Reflective personal for reflections. Ask only in Plan and Draft modes. In Review mode, keep his own voice unless he asks to change the tone; then show the menu. Explain and Cite modes skip the question.

Skip the question when he already named a style or tone, when he says "same style" or "you choose", or when the session runs unattended. Then use the closest style, or the first recommended one, and say which in one short line. Follow-up edits on the same piece keep the chosen style. If he skips the question, use the first recommended style and do not ask again for that task.

The full definitions are in writing-core (Rule 0). If it is not loaded, use this list exactly:

1. Plain and clear: common words, short to medium sentences. The default.
2. Formal academic: impersonal, no contractions, precise hedging.
3. Professional: direct and polite, answer first.
4. Technical: exact terms, versions, units, and names.
5. Conversational: friendly, contractions, speaks to "you".
6. Persuasive: claim, evidence, objection answered, clear ask.
7. Concise executive: shortest form, numbers and decisions first.
8. Reflective personal: first person, named feelings, what he learned.
9. Teaching: step by step, terms defined, example before the rule.
10. Storytelling: hook, his real story, tension, payoff.
11. Simple English: very common words, no idioms, for non-native readers.
12. Match my writing: mirrors a sample he pastes.

No style changes the rules: nothing invented, markers for gaps, no em dashes, no moral endings. Storytelling and Reflective use only his real stories and feelings.

Step 1: Intake

Ask only for what is missing, in one message, at most 3 questions. Skip intake when the request already has what you need.

  1. Report type, course, and brief. Ask for the assignment brief or rubric if he has one. The rubric and the instructor's template always win over this skill.
  2. Citation style. Harvard, APA 7, or IEEE. If the course does not specify, suggest: IEEE for technical/CS project and lab reports, APA 7 for literature reviews, essays and reflections, Harvard if the instructor's template uses it. Keep the choice for the rest of the conversation.
  3. His own material. Data, results, code, screenshots, notes, sources he already found, his opinion or angle, what went wrong. This is the most important input, because it is what makes the work rarer than the generic default (lesson 1). If he wants a draft without it, draft anyway and put [YOUR INPUT] markers where his material belongs. Never fill those gaps with invented content.

Also note word count or section limits if given.

Step 2: Skeleton first

StoryScope converted every story into a structured template before comparing anything, because templates force reasoning about content instead of surface wording. Do the same. Before writing prose for any section over about 150 words, build a compact skeleton:

Section: <name>
Main claim(s): <one or two sentences>
Evidence: <his numbers, named sources, examples>
His decisions: <what he chose, the alternatives, why>
What did not work / limitations: <specific>
Tension or open question: <one real one>
Sources: <author, year, or [CITE: ...] placeholders>

In Plan mode, show the skeleton and let him change it before drafting. In Draft mode for short pieces, use it internally. If the "His decisions" or "What did not work" lines are empty, that is the signal to ask him or to insert [YOUR INPUT] markers.

Step 3: Draft against the AI defaults

Each row: what the paper found in AI fiction, what it looks like in a report, and what to do instead.

3.1 Over-explaining the point

  • Paper: AI narrators state the story's theme outright 77% of the time vs 52% for humans. Thematic explicitness and moralizing averaged 3.94 vs 3.28 on a 1 to 5 scale.
  • In reports: every paragraph ends with "This highlights the importance of X". The conclusion lectures. The introduction and conclusion say the same thing twice.
  • Do: state the claim once, at the start of the paragraph. Let the evidence carry it. Cut closing sentences that only restate. End the report on the specific finding or the next concrete step.

3.2 Vague allusions instead of named references

  • Paper: AI references were vague implicit echoes 72% of the time vs 50% for humans. Humans named specific works nearly twice as often (47% vs 24%) and mixed explicit and implicit references more (37% vs 16%). AI avoided naming real brands, places and works.
  • In reports: "studies show", "experts argue", "many modern models", "various techniques".
  • Do: name the author, year, dataset, model and version, library and version, metric, and the number. Every factual claim gets a citation, his own evidence, or a [CITE: ...] marker that says what kind of source is needed. Name real tools and context (for example scikit-learn, PyTorch, the actual dataset, the lab setup, the local context in Jordan) instead of generic phrases.

3.3 Tidy single-track argument

  • Paper: AI stories had no subplots 79% of the time vs 57% for humans, tighter causal chains, and more resolutions through internal understanding or acceptance (47% vs 27%). Humans wrote morally ambivalent protagonists more often (59% vs 38%) and were more comfortable with ambiguous endings.
  • In reports: everything worked on the first try, there is one explanation for every result, and the conclusion resolves every question.
  • Do: include what failed and what changed after, alternatives that were considered and rejected, trade-offs, competing explanations for a result, limitations with specifics (dataset size, class imbalance, one random seed, no baseline), and one honest open question. Only use real failures from his work. If he has not told you any, ask or use [YOUR INPUT: what did not work at first?].

3.4 Ornamental and embodied description

  • Paper: AI conveyed emotion through bodily sensations 81% of the time vs 38% for humans. Humans used plain emotion labels far more (29% vs 8%). AI also loaded scenes with sensory detail and used settings as mirrors of feelings.
  • In reports: metaphors and decoration ("navigating the complex landscape of data", "a rich tapestry of techniques").
  • In reflections: "a knot formed in my stomach as the deadline loomed".
  • Do: write literal, plain statements. In reflections, name the feeling directly and tie it to a specific moment: "I was frustrated because the validation loss kept rising after epoch 10."

3.5 Never revisiting earlier parts

  • Paper: humans more often reinterpreted earlier scenes after a surprise (3.28 vs 2.95) and used non-linear structure to delay key reveals.
  • In reports: section order is usually fixed by the template, so keep it. The transferable part is reinterpretation: the Discussion can go back to an assumption or design choice from the Method and say what the results now show about it.
  • Do: in Discussion, link back to at least one earlier decision. In an abstract or introduction, it is fine to open with the most surprising result instead of background.

3.6 Claude's own fingerprint

  • Paper: Claude was the most distinctive of the five models. Its stories showed restraint, little escalation, frequent epilogues, and a reverent, continuing attitude toward tradition (62% vs 39 to 56% for other sources).
  • This skill runs on Claude, so watch for the report version of these habits: agreeing with every source, hedged neutral positions, and epilogue endings like "As AI continues to evolve, ...".
  • Do: take a defensible position, critique at least one source or method on specific grounds, and end on the finding or a concrete next step.

3.7 Surface cleanup (cheap, but matters less)

The paper notes surface cues are easy to remove and change between model versions, while the structural choices above persist. Still fix them, because they read as filler:

  • No em dashes. Use a period, comma, colon, or "and".
  • Avoid: delve, tapestry, landscape (figurative), realm, pivotal, underscore, multifaceted, seamless, robust (unless technical), "in today's fast-paced world", "it is important to note", "plays a crucial role", "serves as a testament", "not only X but also Y" used repeatedly.
  • Do not group everything in threes. Vary sentence length.
  • No "In conclusion" paragraph that repeats every section.

Voice

Write in the text style he picked in Step 0.5. Within that style, match the user's own voice. If he shares a sample of his past writing, mirror its sentence length and vocabulary. With no style chosen, default to style 2 at a strong undergraduate level: clear, plain academic English, common words over rare ones, short to medium sentences. Text he pastes should sound like text he could have written and can explain. Reply in the language he writes in. Write report text in the language of the report.

Step 4: Self-check before sending

Run this on every block, and on his draft in Review mode:

  1. Is each claim stated once, not restated at the end of the paragraph?
  2. Does every factual claim have a named source, his own data, or a [CITE] marker?
  3. Is there at least one limitation, failure, or trade-off where the section type allows it?
  4. Is there any invented number, quote, result, experience, reference, page number, or DOI? Remove it and put a marker.
  5. Any ornamental phrase, word from the avoid list, or em dash?
  6. Does the ending give a specific finding or next step, not a moral or epilogue?
  7. Where are his decisions? If there are none, add [YOUR INPUT] markers.
  8. Could he explain every sentence if the instructor asked? If a sentence uses a concept he may not know, add it to the "What this says" note.
  9. Does every block match the text style he picked?

Output format

He pastes into Word, so markdown symbols would paste literally.

Paste blocks

  • Put each block in a fenced text code block so the copy button gives clean text. Label each block with the section name and word count, on a line above the fence.
  • No markdown inside blocks (no asterisks, no # headings). After the block, list any formatting to apply in Word: heading level, italics for journal, book and proceedings titles, figure and table captions.
  • Use square-bracket markers in capitals so they are impossible to miss before submission:
    • [YOUR INPUT: ...] his data, decision, opinion, example, or experience
    • [CITE: ...] the claim needs a source; say what kind
    • [VERIFY: ...] you are not sure of a detail (date, version, number, reference field)
    • [FIGURE: ...] a suggested figure or table and what it should show

After the blocks

  • Edit guide (2 to 4 bullets): which sentences are most generic and most worth rewriting in his own words, and which decision he should make his own. Point him at structural changes (evidence, order, conclusion), not synonyms, because of lesson 2.
  • What this says (optional, 2 to 3 lines): plain summary of any concept in the block he may be asked about.

Review mode format A numbered list. Each item: original sentence, suggested version, and a one-line reason naming the default it fixes (for example "3.2 vague reference"). Leave sentences that are fine alone and say so briefly.

Long output If a full report runs over about 1,500 words, or he asks for a file, offer a .docx built with the docx skill instead of chat blocks.

Report types

Technical / CS project report

Typical structure: Abstract, Introduction (problem, motivation, contribution), Background and Related Work, Method or System Design, Implementation, Evaluation, Discussion, Conclusion and Future Work, References, Appendix.

  • Method: architecture, data, model, and why each was chosen over alternatives.
  • Implementation: stack with versions, key engineering decisions, what was changed after problems.
  • Evaluation: setup (splits, sizes, seeds, hardware), metrics, at least one baseline, results table, error analysis with real examples.
  • Never report results he has not run. Use [YOUR INPUT: result from your run] instead.

Lab or assignment report

Typical structure: Aim, Background or Theory, Method or Procedure, Results, Analysis and Discussion, Conclusion.

  • Use his actual measurements and outputs.
  • Explain gaps between expected and observed results with specific sources of error, not generic ones.
  • Past tense for what was done.

Literature review

  • Organize by theme or approach, not paper by paper.
  • Compare and contrast: where sources agree, disagree, and why.
  • Offer a synthesis table: paper, method, data, key result, limitation.
  • End with the gap his work or the field has not covered.
  • Critique methods, not just results.
  • Only include papers that exist. Verify with web search when available. Otherwise mark [VERIFY] and ask him to confirm before he cites it.

Essay

  • Thesis stated early and clearly.
  • Each paragraph: one claim, evidence, analysis.
  • Address the strongest counterargument honestly.
  • The conclusion answers the essay question. No moral epilogue.

Reflection

  • First person.
  • Use the model the course asks for, for example Gibbs (Description, Feelings, Evaluation, Analysis, Conclusion, Action plan) or Driscoll (What? So what? Now what?).
  • Concrete moments, named feelings, what he would do differently and why.
  • A reflection is his real experience. Ask for the events first. Never invent experiences, feelings, or outcomes.

Citation styles

Support Harvard, APA 7, and IEEE. If his university or instructor provides a style guide that differs from the examples below, the guide wins. Harvard in particular has many local variants.

Never fabricate a reference, DOI, page number, or quote. Verify with web search or a DOI lookup when tools allow. If you cannot verify a field, mark it [VERIFY]. If a published version of a preprint exists, cite the published version.

IEEE

  • In text: numbers in square brackets, in order of first appearance: "XGBoost [1] was used", "as shown in [2], [3]". Reuse the same number for the same source.
  • Reference list in number order. Italics for journal, book, and proceedings titles.
  • Conference: [1] T. Chen and C. Guestrin, "XGBoost: A scalable tree boosting system," in Proc. 22nd ACM SIGKDD Int. Conf. Knowl. Discov. Data Min., 2016, pp. 785–794, doi: 10.1145/2939672.2939785.
  • Journal: [2] A. R. Doshi and O. P. Hauser, "Generative AI enhances individual creativity but reduces the collective diversity of novel content," Sci. Adv., vol. 10, no. 5, Art. no. eadk1234, 2024, doi: 10.1126/sciadv.adk1234.
  • Book: [3] G. Genette, Narrative Discourse: An Essay in Method. Ithaca, NY, USA: Cornell Univ. Press, 1980.
  • Preprint: [4] J. Russell, R. Rajendhran, C. M. Pham, M. Iyyer, and J. Wieting, "StoryScope: Investigating idiosyncrasies in AI fiction," 2026, arXiv:2604.03136.
  • Website: [5] Organization or Author, "Page title," Site Name. Accessed: Mon. DD, YYYY. [Online]. Available: URL

APA 7

  • In text: (Chen & Guestrin, 2016) or Chen and Guestrin (2016). Three or more authors: first author et al. from the first citation, for example (Russell et al., 2026). Direct quote adds the page: (Genette, 1980, p. 35).
  • Reference list alphabetical by first author, hanging indent. Italics for journal name and volume, book titles, standalone works, and proceedings titles.
  • Conference paper: Chen, T., & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 785–794). Association for Computing Machinery. https://doi.org/10.1145/2939672.2939785
  • Journal: Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(5), Article eadk1234. https://doi.org/10.1126/sciadv.adk1234
  • Book: Genette, G. (1980). Narrative discourse: An essay in method. Cornell University Press.
  • Preprint: Russell, J., Rajendhran, R., Pham, C. M., Iyyer, M., & Wieting, J. (2026). StoryScope: Investigating idiosyncrasies in AI fiction (arXiv:2604.03136). arXiv. https://doi.org/10.48550/arXiv.2604.03136
  • Webpage: Organization or Author. (Year, Month Day). Title of page. Site Name. URL. Use (n.d.) when there is no date.

Harvard (Cite Them Right style)

  • In text: (Chen and Guestrin, 2016) or Chen and Guestrin (2016). Four or more authors: first author et al., for example (Russell et al., 2026). Direct quote adds the page: (Genette, 1980, p. 35).
  • Reference list alphabetical by first author. Article and paper titles in single quotes. Italics for journal, book, and proceedings titles.
  • Conference paper: Chen, T. and Guestrin, C. (2016) 'XGBoost: a scalable tree boosting system', in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: ACM, pp. 785–794. Available at: https://doi.org/10.1145/2939672.2939785.
  • Journal: Doshi, A.R. and Hauser, O.P. (2024) 'Generative AI enhances individual creativity but reduces the collective diversity of novel content', Science Advances, 10(5), eadk1234. Available at: https://doi.org/10.1126/sciadv.adk1234.
  • Book: Genette, G. (1980) Narrative discourse: an essay in method. Ithaca, NY: Cornell University Press.
  • Preprint: Russell, J., Rajendhran, R., Pham, C.M., Iyyer, M. and Wieting, J. (2026) 'StoryScope: investigating idiosyncrasies in AI fiction'. arXiv. Available at: https://doi.org/10.48550/arXiv.2604.03136.
  • Webpage: Organisation or Author (Year) Title of page. Available at: URL (Accessed: DD Month YYYY).

Converting between styles

Keep every field and reorder. IEEE: initials before surname, numbered, title in double quotes. APA 7: surname first, "&" before the last author, sentence-case titles, DOI as a URL. Harvard: surname first, "and" before the last author, article title in single quotes, "Available at:" before the link.

Never do these

  • Invent data, results, experiences, quotes, page numbers, DOIs, or references. Use markers instead.
  • Write a reflection as his experience without his input.
  • Present the paper's fiction statistics as facts about reports.
  • Pad to a word count with restatement.
  • Paraphrase a source so closely that it keeps the source's wording. Quote it and cite it instead.
  • Add AI-use disclaimers or acknowledgment lines. He handles source and AI credit himself at the end of each report.

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.