by thefooliman
A multi-agent academic paper workflow for structured writing, iterative polishing, fact-checking and rule-based academic compliance optimization. Suitable for 0–30 pages course papers and seminar theses.
# Add to your Claude Code skills
git clone https://github.com/thefooliman/multi-agent-academic-paper-workflowGuides for using ai agents skills like multi-agent-academic-paper-workflow.
See how multi-agent-academic-paper-workflow compares with popular alternatives.
multi-agent-academic-paper-workflow is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by thefooliman. A multi-agent academic paper workflow for structured writing, iterative polishing, fact-checking and rule-based academic compliance optimization. Suitable for 0–30 pages course papers and seminar theses. It has 51 GitHub stars.
multi-agent-academic-paper-workflow's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/thefooliman/multi-agent-academic-paper-workflow" and add it to your Claude Code skills directory (see the Installation section above). multi-agent-academic-paper-workflow ships a SKILL.md manifest, so compatible agents can discover and load it automatically.
multi-agent-academic-paper-workflow is primarily written in Python. It is open-source under thefooliman on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh multi-agent-academic-paper-workflow against similar tools.
No comments yet. Be the first to share your thoughts!
Unlocks once the catalog security scan passes (runs nightly).
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
multi-agent-academic-paper-workflow vs everything claude code
See comparison
A multi-agent academic writing workflow for Claude Code: structured chapter generation, iterative polishing, double review (self-review + blind fact-check), guard rule enforcement, human-in-the-loop confirmation, and breakpoint resume.
State is saved to state.json + workflow-log.json in your working directory after every step. After /paper-stop (or an interrupt), run /paper-write or /paper-check again; it will offer to resume exactly where you left off.
The coordinator (this session) dispatches worker sub-agents: writer, polishing, self-review, blind-fact-check, guard, io-worker. All agent-to-agent communication is JSON, following schemas/.
A Claude Code plugin for writing and humanizing academic papers. It's aimed at 0–30 page course papers and seminar theses: you bring a topic and an outline, it produces a structured draft that doesn't read like a ai wrote it.
Three entry points, plus two control commands:
/paper-write: generate a paper from scratch, chapter by chapter./paper-check: humanize an existing .md / .tex / .docx manuscript./paper-radical: aggressive de-AIGC pass for papers that need to get past Turnitin AI, GPTZero, 知网, or 万方./paper-stop: save and pause. /paper-reset: wipe state and start over.A coordinator drives the whole thing from your session and hands work to a set of worker agents: writer, polishing, self-review, blind-fact-check, guard, and io-worker. Each chapter runs write → polish → guard → review → aggregate, then stops and asks whether to iterate again. It never loops on its own; you confirm every round.
The blind fact-checker is the part worth pointing at first. It gets a chapter with zero context, so it can't coast on what it already knows. It catches unsupported claims ("costs have fallen sharply" with no number attached) and structural tells (six paragraphs all the same length) that the other agents pass over.
The other thing that matters is the claim-evidence discipline. Every verb has to match the strength of the evidence behind it. A clean RCT can say "reduced"; an observational result should say "is associated with". That rule does more for academic credibility than any vocabulary swap.
Command line is the quickest:
/plugin marketplace add thefooliman/multi-agent-academic-paper-workflow
/plugin install paper-agent-workflow@multi-agent-academic-paper-workflow
/reload-plugins
Then run /paper-write to confirm it loaded.
Or use the marketplace UI:
/reload-plugins if prompted, then /paper-write to check it loaded.Manual clone, if you prefer:
git clone https://github.com/thefooliman/multi-agent-academic-paper-workflow
# copy the folder into your Claude Code plugins directory, then reload
Keep the whole tree intact: skills/, agents/, configs/, schemas/, utils/.
Run /paper-write. It asks for the topic, an outline (paste one or let it draft one), target length, intensity, and optionally a sample of your own previous writing for voice matching. That last one matters: hand it a paragraph you wrote, and the output matches your rhythm and word choices instead of a generic "human" style. Writing your own viewpoints and judgments into the prompt helps the same way — the more of your own thinking you feed in, the more it reads like you and not a clean summary.
Put the file in your working directory, run /paper-check, give the filename. It splits the manuscript into chapters and runs the same polish-review loop without generating anything new.
/paper-radical runs the full pattern catalog, the six-step loop, and the five-dimension score (pass line 42). Use it when standard polishing isn't enough and the paper has to clear an AI detector.
State is saved to state.json (plus workflow-log.json) in your working directory after every step. /paper-stop pauses cleanly. Run /paper-write again later and it offers to resume where you left off. Interrupting mid-iteration loses nothing.
The humanizer rules live in configs/, and they're meant to be edited:
humanizer-patterns-en.md: the pattern catalog, ~45 entries informed by Wikipedia's "Signs of AI writing", the academic-humanizer lineage, and de-aigc-skills. Each entry has a severity (🔴🟡🟢), watch-words, and a before/after. See NOTICE.md for sources and licenses.claim-evidence-discipline.md: verb↔evidence matching and hedging calibration.scoring-rubric.md: the five-dimension score used as the convergence measure.section-strategies.md: per-section rewrite intensity. Abstract and intro get the heaviest rewrite; methods and robustness get the lightest.polishing-rules-en.md: the operational rules that tie the rest together, plus the standard vs radical intensity tiers.skills/ entry skills (the slash commands)
agents/ worker sub-agents
configs/ rule sets and user config
schemas/ JSON contracts for agent output and state
utils/ state_manager + docx/tex parsing placeholders
examples/ sample chapter + humanizer before/after
.docx/.tex parsing is basic for now; complex macros won't split cleanly. Better parsing is planned for v1.1.The humanizer's goal is to return human-written and AI-assisted text to the language distribution of a real researcher, not to help fully AI-generated work evade disclosure. Academic integrity outranks detection scores. When a claim lacks evidence, the fix is to flag it, not to hide it. Comply with your institution's policies.