Research Writing Skill

Research Writing is an Agent Skill for structuring, drafting, reviewing, and validating academic research writing workflows.

It helps agents support paper work without inventing research content, citations, experiments, or venue requirements. The skill focuses on research positioning, writing-quality checks, venue adaptation, simulated review, manuscript assembly, reviewer responses, and pre-submission risk checks.

Core claim:

Research writing quality depends on the fit between claims, evidence, audience, and submission constraints, not on fluent academic prose alone.

What Is Included

• skills/research-writing/ Canonical Agent Skills compatible skill folder.
• plugins/research-writing/ Codex plugin wrapper around the same skill.
• .agents/plugins/marketplace.json Codex marketplace file for repo-based installation.
• docs/claude/README.md Claude Code installation notes.

Install In Codex

After this repository is on GitHub, install it as a Codex marketplace:

codex plugin marketplace add YOUR_GITHUB_USERNAME/research-writing-skill
codex plugin marketplace upgrade research-writing-skill

Then open Codex, go to Plugins, choose the Research Writing marketplace, and install the Research Writing plugin.

For local testing before publishing:

codex plugin marketplace add /absolute/path/to/research-writing-skill

Install In Claude Code

Copy the canonical skill folder into your Claude skills directory:

mkdir -p ~/.claude/skills
cp -R skills/research-writing ~/.claude/skills/research-writing

See docs/claude/README.md for project-scoped installation options.

Use

Ask your agent to use the skill for tasks such as:

Use research-writing to run a writing-quality audit on this draft paper.

Use research-writing to compare this draft against ACL-style expectations.

Use research-writing to prepare a response letter from these reviewer comments.

Method Shape

The skill routes work through focused modes:

• research positioning
• writing assistance
• writing-quality audit
• venue positioning
• simulated review
• manuscript assembly
• reviewer response
• pre-submission check
• asynchronous full pass

The public version is English-first. It keeps the original skill's core engineering shape while removing platform-only metadata, private development records, non-English interface text, and non-public test artifacts.

Author

Created and maintained by Odinary-AI.

License

MIT.