AgentsMeetRL — Awesome List of Agentic Reinforcement Learning
A curated, categorized collection of open-source RL-training projects for LLM agents, with technical details and an interactive dashboard.
- Source repo
- thinkwee/AgentsMeetRL
- Stars
- ★ 1.9k
- Last updated
- 5d ago
- Primary language
- HTML
- FA score
- 29/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platform
- Cost
- Free, no paid service needed
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Researchers wanting a comprehensive overview of current agentic-RL research can browse representative projects per category.
- Not a fit if
- Teams needing runnable training code or frameworks
- Users who need a fully offline, self-contained tool
- Developers not using Claude Code (the skill won't apply)
- Source review
- 29/100 · Major gaps 8 safety controls not found
What does this agent do, and when should you use it?
AgentsMeetRL is an awesome list that systematically curates open-source repositories for training LLM agents with reinforcement learning. It spans 16 categories, including Base Framework, Search & RAG, Web & GUI, Tool-Use, Code & SWE, Multi-Agent RL, VLM Agent, Environment, and more. Built on code analysis and manually reviewed, the list aims to be a reliable reference for researchers and practitioners. It provides an interactive dashboard for visual exploration and is updated regularly (snapshot date 2026-07-23). Additionally, the repository is packaged as a Claude Code skill, turning the corpus into an on-demand assistant for agentic-RL experiment design and troubleshooting.
AgentsMeetRL is an awesome list, not a runnable software. It organizes metadata for hundreds of open-source repositories, including links, star counts, dates, organizations, paper links, and the RL frameworks used. Users can browse category-organized tables (e.g., Base Framework, Search & RAG, Web & GUI) and expand rows to see technical details like RL algorithm, reward type, and tool usage. The repository also includes an interactive dashboard website for visual exploration. Moreover, it is packaged as a Claude Code skill (agents-meet-rl) that, once installed, answers natural-language queries about agentic-RL training, such as reward design, algorithm selection, and troubleshooting common issues like flat rewards or KL divergence spikes.
- Researchers wanting a comprehensive overview of current agentic-RL research can browse representative projects per category.
- Engineers comparing RL frameworks (e.g., veRL, OpenRLHF, trl) for a multi-turn tool-use agent can use the technical details tables.
- Practitioners facing training issues like reward stagnation or KL explosions can query the Claude Code skill for corpus-grounded advice.
- Students and hobbyists can find reproducible examples and papers to learn practical agentic-RL methods.
- Contributors can submit new agentic-RL projects for inclusion via Issues or PRs.
How do you install or deploy this agent?
The repository is a list and requires no installation. To use the Claude Code skill, install it via the plugin marketplace: run /plugin marketplace add thinkwee/claude-plugins then /plugin install agents-meet-rl@thinkwee in Claude Code. Alternatively, clone the repo manually with git clone https://github.com/thinkwee/AgentsMeetRL and copy AgentsMeetRL/skills/agents-meet-rl to ~/.claude/skills/.
How do you use this agent?
After installing the skill, ask Claude Code questions like "my GRPO search agent's reward is flat but eval keeps dropping" or "which RL framework should I pick for a multi-turn tool-use agent?" — the skill routes your symptom to fixes grounded in the list. For browsing, visit the interactive dashboard at https://thinkwee.top/amr/ or explore the category tables directly on GitHub.
What are this agent's strengths and limitations?
- Broad categorization across 16 distinct areas, making it easy to find relevant resources.
- Metadata is derived from code analysis and manually reviewed, adding reliability, and it is actively updated.
- Interactive dashboard and Claude Code skill enhance search and consultation efficiency.
- Each entry lists technical specifics (RL algorithm, reward type, tool usage) for easy comparison.
- As an awesome list, it provides no actual training code or runnable frameworks—only pointers.
- Based on code analysis, there may be inaccuracies or omissions despite manual review.
- Frequent updates may render some older entries outdated without further maintenance.
- The Claude Code skill only benefits Claude Code users; others must browse manually.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| AgentsMeetRL — Awesome List of Agentic Reinforcement Learning This agent | 29 · Major gaps | Agent plugin / skillFree | ★ 1.9k | 5d ago | HTML | — |
| MARTI: Multi-Agent Reinforced Training & Inference for LLMs | 36 · Major gaps | CLIFree | ★ 570 | 1mo ago | Python | — |
| Hands-On Modern RL | 30 · Major gaps | CLIFree | ★ 4.7k | 17d ago | Python | — |
| rLLM: Reinforcement Learning Framework for LLM Agents | 0 · Major gaps | CLIFree + model costs | ★ 5.9k | 10d ago | Python | OpenAI API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows this is a curated list with no executable code or permission requests. Therefore, all trust-related criteria score 0 because there is no evidence of permission management, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, or source attribution. For source attribution, although repository and paper links are listed, no verification or audit information is provided.
Self-consistency scores 2: the README's categorization and update logs are consistent, but no code or tests are provided to verify. Dependency availability scores 1: many external repositories are listed, but no dependency management or availability guarantees are provided. Failure messages score 0: no documentation of error handling or failure messages.
Audience and scenarios score 2: clearly targets researchers and developers with multiple usage scenarios. Capability boundaries score 1: although categories are defined, the specific capability boundaries of each entry are not clarified. Trigger precision scores 1: as a curated list, trigger conditions are unclear. Environment fit scores 1: no environment configuration or compatibility notes are provided.
Information architecture scores 2: clear categorization and well-structured tables. Install notes score 1: provides installation for Claude Code skill but no other installation or usage instructions. Naming stability scores 1: repository and category names are stable, but no version control. Examples and FAQ score 1: provides usage examples but no FAQ. Known limitations score 1: mentions possible inaccuracies but not detailed. License scores 0: no license mentioned. Versioning and changelog score 1: has update logs but no formal version numbers. Maintenance responsibility scores 1: welcomes contributions but no clear maintainers.
Output usability scores 2: provides structured tables and categories for easy use. Marginal value scores 2: as a curated list, provides valuable resource aggregation. Cost-benefit scores 1: no cost or benefit analysis provided.
Claim traceability scores 1: provides paper and repository links but no verification methods. Cross-source corroboration scores 1: some entries have multiple sources but not systematically verified. Fact-inference separation scores 1: some content is explicitly marked as inference but not fully separated.
- Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
- Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
- Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
- Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Not found in source: verifiable attributionInstall from the official repo or registry and check the publisher and URL to avoid look-alike packages.
- This repository is a curated list without executable code, so its security and reliability cannot be verified.
- License information is missing; confirm before use.
- Some entries may contain inaccuracies; cross-verification is recommended.