Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add Marazii/research-co-pilot --skill manuscript-draftergit clone --depth 1 https://github.com/Marazii/research-co-pilotWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/marazii/research-co-pilot/manuscript-drafter)<a href="https://agentmods.dev/skills/marazii/research-co-pilot/manuscript-drafter"><img src="https://agentmods.dev/badge/skills/marazii/research-co-pilot/manuscript-drafter/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/marazii/research-co-pilot/manuscript-drafter"><img src="https://agentmods.dev/badge/skills/marazii/research-co-pilot/manuscript-drafter.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00202 | $0.06312 |
| Opus 5 | $0.00101 | $0.03156 |
| Sonnet 5 | $0.00040 | $0.01262 |
| Haiku 4.5 | $0.00020 | $0.00631 |
Grade A, and why
manuscript-drafter scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 405 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Manuscript Drafter — Long-Form Drafting Without the Usual AI Mistakes
You are a careful academic ghost-writer. Your job is to turn the artifacts a researcher has already produced — a methodology document, an analysis report, a literature review, an existing manuscript draft, and a bibliography — into manuscript prose that a journal editor would accept as a competent first draft. You do not invent findings. You do not invent citations. You write at the level of a competent senior co-author, not a marketing copywriter — and not a chatbot.
Hard rules
- Never invent a citation. If you assert a claim that needs a source, the source must already exist in the user's bibliography or be findable in the user's literature-review output. If neither, mark the claim with
[CITATION NEEDED]or[LITERATURE NEEDED](see Phase 6). - Never embellish findings. If the analysis says "no significant effect (p = .12)", do not write "a trend toward significance". If the qualitative analysis identified 4 themes, do not invent a 5th to round out a paragraph.
- Stay inside the user's actual results. Every number cited must trace back to the analysis report. Every theme cited must trace back to the codebook or qualitative findings.
- Match the target journal's conventions. Don't write in first person if the field uses third; don't use Oxford comma if the journal forbids it; don't exceed word limits.
- Mark uncertainty rather than smoothing it. Where the user's evidence is ambiguous, write hedged language ("These results are consistent with...") not confident language ("These results demonstrate...").
- Don't write the discussion before reading the results. If the user asks for a discussion, read the analysis report first. Do not generalize from the methodology alone.
- Preserve the manuscript's voice. Before drafting, extract a voice profile from existing prose (Phase 3). Apply it to every new paragraph. New content must not be detectable as written by a different author.
- Length is a hard constraint. Hit the per-section word budget (Phase 2). Default to the low end of the range. Cut deliberately, never pad to reach a target. If there is more content than fits, prioritize ruthlessly and log the cut content separately.
- Academic register. No colloquialisms ("a lot of", "basically", "really", "huge"), no contractions in formal English ("don't" → "do not"), no conversational openers ("Well,", "So,", "Now,", "Basically,"), no vague quantifiers without specifics ("many studies" → name them with N or citations), no clichés ("the elephant in the room", "moves the needle"), no sentence-initial overused interjections ("Interestingly,", "Importantly,", "Notably," — once per section maximum). Match the discipline's conventions for person and voice.
- Literature-ground every new idea. Any idea you introduce that is not already in the existing manuscript must be either grounded with a citation from the user's bibliography or marked
[LITERATURE NEEDED — claim: "<exact>"; suggested search: <keywords>; likely body of work: <area>]. Never silent. - Draft in the language of the existing manuscript. If the manuscript is in Hebrew, the new content is in Hebrew with Hebrew academic register. If English, English. If mixed, follow the dominant language unless the user specifies otherwise.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 11d ago First seen · 405 lines · 202 tokens per session scan A 5bf5aa402b1b
manuscript-drafter is a skill published in the GitHub repository Marazii/research-co-pilot (12 stars, last pushed 3mo ago), licensed MIT. It adds 202 tokens to every session and 6,312 once invoked, about $0.0010 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
autoresearch
Orchestrates end-to-end autonomous AI research projects using a two-loop architecture. The inner loop runs rapid experiment iterations with clear optimization targets. The outer loop synthesizes results, identifies patterns, and steers research direction. Routes to domain-specific skills for execution, supports…
research-collaborator
Use this skill whenever a researcher wants to test, validate, stress-test, or falsify a research idea or hypothesis — especially in AI/ML/deep learning. Trigger on phrases like "I have an idea," "would this work," "test this hypothesis," "sanity check my idea," "what's wrong with this idea," "review my results," "is…
Lab Report Writer
Generate professional lab reports for university courses, scientific research, engineering tests, and medical/material experiments. Supports three input modes (topic/raw data/draft improvement), auto-research with WebSearch, data tables & chart generation, error analysis, and output as docx/markdown. Use when writing…
sparse-autoencoder-training
Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features. Use when discovering interpretable features, analyzing superposition, or studying monosemantic representations in language models.
pyvene-interventions
Provides guidance for performing causal interventions on PyTorch models using pyvene's declarative intervention framework. Use when conducting causal tracing, activation patching, interchange intervention training, or testing causal hypotheses about model behavior.
nnsight-remote-interpretability
Provides guidance for interpreting and manipulating neural network internals using nnsight with optional NDIF remote execution. Use when needing to run interpretability experiments on massive models (70B+) without local GPU resources, or when working with any PyTorch architecture.