SkillFoxx

ERL Research Skills

28 skills and three behavior contracts for Robotics and AI research: literature, experiments, papers, reviewer replies.

  • 51 ★ on the repository
  • Last commit
  • MIT
Skill

Extended access

Why this level

  • The skills read paper and repository files and prepare manuscript edits, while contracts change agent behavior across all project tasks.
All reasons and checks

ucl-erl/skillsAuthor:UCL-ERL

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add skills/erl-research-skills

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

  • The skills read paper and repository files and prepare manuscript edits, while contracts change agent behavior across all project tasks.

This is third-party code. Review the repository files before installing.

What it does

ERL Research Skills is a public set from the ERL lab at UCL for agent research tasks. The repository splits capabilities into skills that define a workflow and behavior contracts that define how the agent behaves throughout the work. Contracts are persistent instructions in `AGENTS.md` and `CODEX.md`, chosen with the guide and copied into a project. `catalog.json` describes 28 skills. The main flows are literature to research direction, experiments to paper, and paper to submission. They include reading cards, an evidence matrix, citation and manuscript structure audits, red-team review, reviewer responses and a LaTeX submission check. Paper review has three modes: `quick`, `standard` and `deep`. Examples and curation rules accompany the skills.

Who it is for. Graduate students and researchers in Robotics and AI who write papers and prepare experiments with an agent.

Good fit when

  • You need to enter a new area and build a literature review with an evidence matrix.
  • Remote experiment results must be packed into a dossier and checked against the paper code.
  • Before submission you need an audit of citations, manuscript structure and LaTeX.
  • A vague research plan needs stress-testing by interrogation.

Not a fit when

  • You need a paper outside Robotics and AI without adaptation.
  • You want an autonomous paper generator with no author review.
  • The agent has no access to the paper and repository files.

Example request

Use rai-paper-flow in deep mode on my paper and prepare an audit of citations and structure plus a list of risks before submission.

Limitations

The repository has no single install command: skills are connected the way your agent client supports. Behavior contracts must be chosen and merged into a project by hand and are not invoked as skills. Results rest on the sources you give the agent, and audit conclusions need rechecking, since the skills do not replace a supervisor. A science lab develops the set and part of the material is oriented to Codex.

How to disable. Delete the copied skill folders and the `AGENTS.md` or `CODEX.md` contracts from the project. The cloned repository can be removed entirely.

FAQ

How do I get started?

Clone the repository, connect the skill folders the way your client supports and invoke a skill by name, for example `rai-research-flow`.

How does a skill differ from a behavior contract?

A skill defines a workflow while a contract permanently defines how the agent behaves. Contracts live in `agent-instructions` and are copied into a project.

Which modes does rai-paper-flow have?

`quick` for daily edits, `standard` for normal multi-section revision and `deep` for submission, rebuttal and full review.

Where do I start research?

With `rai-research-flow`: it leads from paper search and reading cards to an evidence matrix and a direction.

How do I validate the repository?

Run `python scripts/validate_catalog.py` and `python scripts/validate_forward_tests.py` after cloning.

Discussion

No questions or reviews yet.

Skills

A library of 280+ research skills: literature search, experiment design, analysis, reference checks and paper preparation

SkillExtended access124

A skill for master and PhD students that runs research through novelty and reality gates, from review to claim checks

SkillMinimal access71

Research Skills gives Claude Code, Codex and Copilot CLI plugins for research: literature, grants, papers, figures, slides.

PluginExtended access55

A marketplace of five Claude Code plugins for researchers: literature search, research design, project memory and manuscript work

PluginBroad access655

A skill that checks a manuscript against practices of 883 ICLR, ICML and NeurIPS Oral papers and suggests edits

SkillMinimal access580

Claude Code skills for the academic pipeline: research, paper drafting, peer review and citation checks

SkillExtended access56.7K

Comparison with similar tools

Cards that share the main search query, and how they differ.

ToolOfficialRepository starsMain difference
ERL Research Skills (this card)no51
Nature-Paper-Skillsno67127 skills following the paper logic from hypothesis to submission
Claude Scientific Writerno2,571a plugin that writes papers and grants from found literature

README in short

The repository separates skills from behavior contracts and describes three main work flows. The catalog lives in `catalog.json` with curation rules, examples and validation schemas.

README badge

Original SKILL.md

---
name: robotics-ai-coding-flow
description: Applies Robotics & AI code quality checks for Python, C++, ROS, simulation, ML experiments, and reproducibility. Use when implementing, debugging, testing, or reviewing Robotics and AI research code.
---

# Robotics AI Coding Flow

Use this as a Robotics & AI layer on top of normal coding skills.

## Inputs

- User request, codebase, failing logs/tests, issue description, or desired feature.
- Runtime constraints: Python/C++, ROS version if relevant, simulator/hardware, dataset, compute, and safety boundary.
- Expected artifact: patch, debug diagnosis, test, review findings, reproduction instructions, or experiment-quality check.

If hardware or simulator access is unavailable, use the smallest offline verification loop and state what remains unvalidated.

## Workflow

1. Read the repo's existing setup, tests, configs, and domain terminology.
2. Identify the code layer:
   - algorithm;
   - ML training/evaluation;
   - simulator;
   - ROS node/package;
   - hardware interface;
   - data processing;
   - plotting/reporting.
3. Choose the smallest verification loop that can fail for the current change.
4. Implement or debug with tests where practical.
5. Run Robotics-specific checks from `references/robotics-code-checks.md`.
6. Report what was verified and what still needs simulation or hardware validation.

## Verification Matrix

```markdown
| Layer | Likely checks | Hardware/sim caveat |
| --- | --- | --- |
| Algorithm | unit tests, property checks, toy cases | May miss timing and integration failures. |
| ML training/evaluation | smoke run, config diff, metric recomputation | May miss full-scale variance and compute limits. |
| Simulator | deterministic scenario, asset/config check | May not transfer to real hardware. |
| ROS node/package | build/test, launch/config check, topic contract | May miss live timing and transport issues. |
| Hardware interface | dry-run guards, bounds checks, log replay | Real hardware requires explicit user approval and safety context. |
| Plotting/reporting | artifact trace, metric source check | Plot correctness depends on upstream logs. |
```

Report a problem