rlm
Skill1 starsChanged 3 months agoMITView on GitHub
Use when developing or validating reinforcement-learning-machine/coding-agent behavior end to end. Especially useful for agent workflow design, tool/action contracts, qualitative coding-agent generalization, project-building gates, promotion loops, and coordinating SFT, RL, eval, harness, and dataset work into one production path.
12 other projects use a version of this.
Adopted by
Most recently taken up by
- osick/ctx.skillstheir own version, first seen Oct 13, 2026
- Trampoline-AI/predict-rlmtheir own version, first seen Oct 8, 2026
- brainqub3/claude_code_RLMtheir own version, first seen Oct 8, 2026
- MiracleWeb3/rlmtheir own version, first seen Oct 8, 2026
- brainqub3/RLMtheir own version, first seen Jun 22, 2026
- Jinshi1945/claude_code_RLMtheir own version, first seen Feb 22, 2026
Other versions12 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 12
- Trampoline-AI/predict-rlm★ 436same name
- brainqub3/claude_code_RLM★ 393same name
- brainqub3/RLM★ 393same name
- Tenobrus/claude-rlm★ 94same name
- richardwhiteii/rlm★ 49same name
- andremoreira73/Useful-Skills-from-lyfX★ 8same name
- endjin/RLM★ 6same name
- Jinshi1945/claude_code_RLM★ 1same name
- rgoldr88/claude-rlm★ 1same name
- arkaigrowth/rlm★ 0same name
2 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository32 other files it gives its agents.
- advertising-imagery-agentSkill
- agent-watchSkill
- aria-intelligenceSkill
- bambu-labsSkill
- cadSkill
- cad-viewerSkill
- dataset-engineeringSkill
- design-business-agentSkill
- design-loopSkill
- dxfSkill
- elite-engineeringSkill
- emil-deign-skillSkill
- env-harnessSkill
- gcodeSkill
- human-writingSkill
- implicit-cadSkill
- inference-engineeringSkill
- kernel-engineeringSkill
- lamp-method-first-principles-thinkingSkill
- marketing-and-advertising-agentSkill
- model-evalSkill
- new-project-architectureSkill
- opportunity-hunterSkill
- rlSkill
- sdfSkill
- sendcutsendSkill
- sftSkill
- srdfSkill
- step-partsSkill
- synthesys-bug-triageSkill
- taste-maxSkill
- urdfSkill
The file
We split this skill into the rules it gives an agent
6 rules. 0 are common, 6 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 10 edited versions we read keep these.
6
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Use this skill when the work concerns the full coding-agent model system, not just one training or eval component.
- Distinctive: line 33
Keep action chains and tool calls distinct.
- Distinctive: line 40
test with real signal
- Distinctive: line 52
Do not let generic prose satisfy either contract.
- Distinctive: line 83
Run qualitative live or project-shaped tests after numeric gates improve.
- Distinctive: line 84
Document artifact paths, commands, metrics, and the next bottleneck.
“Common” is measured: kept by at least 6 in 10 of the 10 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
RLM
Use this skill when the work concerns the full coding-agent model system, not just one training or eval component.
Principle
RLM quality is the ability to turn a messy production request into correct bounded action: inspect, reason, edit, test, verify, document, and stop safely.
The model should learn the operating style, not just fixture strings.
Agent Behavior Targets
Train and evaluate for:
- reading the repo before editing
- identifying the real invariant
- choosing a bounded implementation slice
- using tools with exact arguments
- preserving workspace and host boundaries
- running meaningful tests
- inspecting failures instead of guessing
- updating docs after meaningful passes
- promoting only with evidence
- stopping at a clean checkpoint
Contract Surfaces
Keep action chains and tool calls distinct.
Action-chain rows test sequencing and judgment:
- inspect first
- isolate bottleneck
- patch narrowest useful path
- test with real signal
- verify no regression
- document decision
Tool-call rows test executable precision:
- exact tool name
- exact argument object
- strict JSON boundary
- schema validity
- safe side effects
Do not let generic prose satisfy either contract.
Project-Building Quality
A useful RLM project task should force the model to produce a working path, not an essay.
Reward:
- concrete files or modules
- interface/contract definition
- minimal but real implementation
- deterministic tests
- endpoint, CLI, or probe surface
- user-perspective smoke
- numeric evidence where possible
- limitations and rollback notes
Reject:
- checklist-only plans
- broad claims without verification
- unsafe host access
- overfit wording
- toy examples that do not expose coding-agent quality
End-To-End Loop
- Identify the current bottleneck from eval and live behavior.
- Decide whether it is data, SFT, RL, harness, runtime, or model serving.
- Build the smallest focused round that can expose movement.
- Preserve older gates before promotion.
- Run qualitative live or project-shaped tests after numeric gates improve.
- Document artifact paths, commands, metrics, and the next bottleneck.
Quality Rubric
Score the model on:
- instruction retention
- tool exactness
- argument exactness
- workflow coherence
- repo awareness
- test selection
- debugging discipline
- safety boundary handling
- project usefulness
- concise stopping behavior
The final goal is not just exact match. The goal is a small model that can reliably help build real systems under the user’s engineering paradigms.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.