agentleFS
TabooHarmonyTabooHarmony/roblox-brain/AGENTS.md

roblox-brain

AGENTS.md150 starsChanged 51 days agoMITView on GitHub

roblox-brain is a skills-only repository for AI coding agents working with Roblox Studio. 29 curated skills in four libraries (core, gameplay, design, tools), covering Luau, Roblox engine APIs, architecture, game design, security, monetization, localization, and workflow. Distributed via npx skills add TabooHarmony/roblox-brain; install a single library with npx skills add TabooHarmony/roblox-brain/skills/core (or another library path). No plugin code. No build system. No npm package. Just SKILL.md files. Each skill follows progressive disclosure. Compatible hosts discover skill names and descriptions, then…

The file

Show its rules and text

We split this AGENTS.md into the rules it gives an agent

24 rules. About 0 are common, 7 are distinctive, and 17 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 17

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    No plugin code. No build system.

    line 7
  2. Very specific:

    Every SKILL.md must have:

    line 31
  3. Very specific:

    Create skills/<core|gameplay|design|tools>/roblox-<name>/SKILL.md following the structure above

    line 62
  4. Very specific:

    Create that skill's references/full.md with complete reference (under 50,000 chars)

    line 63
  5. Distinctive:

    Add a row to README.md in the appropriate skills table and update the skill count badge

    line 64
  6. Very specific:

    Run python3 scripts/validate_skills.py: must pass

    line 65
  7. Very specific:

    If the skill makes API claims, add focused entries to scripts/api_drift_registry.yaml

    line 66
  8. Very specific:

    Run python3 scripts/validate_skills.py before committing.

    line 70
  9. Distinctive:

    All skill content must trace to a real source: official Roblox creator-docs, repositories with a compatible license or explicit reuse …

    line 94
  10. Very specific:

    Use sources: [original] for synthesis/opinion/design heuristics

    line 97
  11. Distinctive:

    When reusing material from external repos, preserve attribution and license terms alongside the material.

    line 98
  12. Very specific:

    Use raw.githubusercontent.com URLs (not github.com/blob/) for source links

    line 99
  13. Very specific:

    Keep SKILL.md lean; if it hits 2,500+ chars, move detail to references/full.md

    line 110
  14. Very specific:

    No orphaned reference files.

    line 112
  15. Distinctive:

    Don't add plugin code, build systems, or npm packaging (this is skills-only)

    line 131
  16. Distinctive:

    Don't write skills from training data without verifying against current Roblox docs

    line 132
  17. Distinctive:

    Don't break cross-references when renaming or splitting skills

    line 133
  18. Very specific:

    Don't commit without running python3 scripts/validate_skills.py

    line 134
  19. Very specific:

    Don't leave sources: []: use [original] for synthesis

    line 135
  20. Very specific:

    Don't use `lua; use `luau

    line 136
  21. Very specific:

    Don't add ## Overview sections.

    line 137
  22. Very specific:

    Don't create standalone reference files.

    line 138
  23. Distinctive:

    Don't exceed 50,000 chars in reference files

    line 139
  24. Very specific:

    Don't use github.com/blob/ URLs.

    line 140

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

AGENTS.md: Guide for AI Agents and Contributors

What This Repo Is

roblox-brain is a skills-only repository for AI coding agents working with Roblox Studio. 29 curated skills in four libraries (core, gameplay, design, tools), covering Luau, Roblox engine APIs, architecture, game design, security, monetization, localization, and workflow. Distributed via npx skills add TabooHarmony/roblox-brain; install a single library with npx skills add TabooHarmony/roblox-brain/skills/core (or another library path).

No plugin code. No build system. No npm package. Just SKILL.md files.

Architecture: Progressive Disclosure

Each skill follows progressive disclosure. Compatible hosts discover skill names and descriptions, then load the selected entry point and references as needed.

  1. Skill metadata: the host discovers names and descriptions from each SKILL.md
  2. SKILL.md (~600 tokens, max 3,000 chars): quick reference, enough for most tasks
  3. references/full.md (max 50,000 chars): detailed guidance and examples for full skills

Thin routing skills declare kind: router and may omit references/full.md when their entire job is selecting the right domain skills.

The README is the human-facing catalogue. Skill hosts discover the individual SKILL.md files directly.

skills/gameplay/roblox-gui/
├── SKILL.md              ← Quick Reference (under 3,000 chars)
└── references/
    └── full.md           ← Full Reference (under 50,000 chars)

Skill File Structure

Every SKILL.md must have:

---
name: roblox-<name>
description: >
  One-line description, under 150 chars.
when_to_use: "Optional trigger phrases: the words users type when they need this skill (fix, change, review, audit, symptoms), under 300 chars."
last_reviewed: YYYY-MM-DD
sources:
  - https://create.roblox.com/docs/...    # or [original] for synthesis
---

# Skill Title

## When to Load

1-3 sentences on when an agent should load this skill. Include cross-refs to related skills in backticks.

## Quick Reference

Dense table or list format. The most useful info inline. Code examples in ```luau blocks.

> Full code examples and API tables: [references/full.md](references/full.md)

Required sections: ## When to Load, ## Quick Reference Banned sections: ## Overview, ## 1. Overview, ## Full Reference (use references/full.md)

Adding a New Skill

  1. Create skills/<core|gameplay|design|tools>/roblox-<name>/SKILL.md following the structure above
  2. Create that skill's references/full.md with complete reference (under 50,000 chars)
  3. Add a row to README.md in the appropriate skills table and update the skill count badge
  4. Run python3 scripts/validate_skills.py: must pass
  5. If the skill makes API claims, add focused entries to scripts/api_drift_registry.yaml

Validation

Run python3 scripts/validate_skills.py before committing. Checks:

  • SKILL.md under 3,000 chars (the when_to_use line is not counted)
  • when_to_use, when present, is a non-empty string under 300 chars
  • references/full.md under 50,000 chars
  • Description under 150 chars
  • Required frontmatter: name, description, last_reviewed, sources
  • sources field is not empty (use [original] for synthesis)
  • ## When to Load section present
  • ## Quick Reference section present
  • No ## Overview or ## 1. Overview in SKILL.md
  • No ## Full Reference in SKILL.md
  • No lua code blocks (use luau)
  • references/full.md exists unless kind: router
  • Cross-references (`roblox-X` in backticks) point to existing skills

Additional CI checks:

  • python3 scripts/verify_api_drift.py: API claims against live Roblox creator-docs
  • python3 scripts/verify_source_urls.py: source URLs are reachable
  • python3 scripts/verify_version_pins.py: tool version pins (non-blocking)

CI runs on all branches (.github/workflows/ci.yml).

Sourcing Rules

  • All skill content must trace to a real source: official Roblox creator-docs, repositories with a compatible license or explicit reuse permission, or original work
  • Training-data-only content is unacceptable. Verify API references against current docs
  • Track sources in the sources: frontmatter field
  • Use sources: [original] for synthesis/opinion/design heuristics
  • When reusing material from external repos, preserve attribution and license terms alongside the material; never copy without permission
  • Use raw.githubusercontent.com URLs (not github.com/blob/) for source links
  • Mark temporal claims with event dates: <!-- temporal: 2026-06 -->

Conventions

  • Descriptions must trigger on changing, fixing, debugging, reviewing and auditing, not only on building. Hosts match the user's words against description + when_to_use, and bug reports rarely say "implement". Put the main use case first, and put the symptoms users describe ("overlaps on mobile", "data not saving") in when_to_use.
  • Luau code blocks use luau (not lua)
  • API names in backticks: UserInputService, RemoteEvent
  • Cross-skill references in backticks: `roblox-networking`, `roblox-data`
  • Last reviewed date: update when you verify content against current docs
  • Keep SKILL.md lean; if it hits 2,500+ chars, move detail to references/full.md
  • Mark illustrative reference code as illustrative and require project-specific verification
  • No orphaned reference files; all content goes in references/full.md

Key Files

FilePurpose
skills/{core,gameplay,design,tools}/*/SKILL.mdQuick reference for each skill
skills/{core,gameplay,design,tools}/*/references/full.mdFull reference for each non-router skill
scripts/validate_skills.pyValidation script for skill structure
scripts/verify_api_drift.pyAPI drift checker against live creator-docs
scripts/verify_source_urls.pySource URL reachability checker
scripts/verify_version_pins.pyTool version pin monitor (non-blocking)
scripts/mirror_creator_docs.pyRefresh the local docs cache for API checks
scripts/api_drift_registry.yamlRegistry of API claims to verify
requirements.txtPython dependencies (pyyaml)
.github/workflows/ci.ymlCI: validation + drift + source URLs + install test

What NOT to Do

  • Don't add plugin code, build systems, or npm packaging (this is skills-only)
  • Don't write skills from training data without verifying against current Roblox docs
  • Don't break cross-references when renaming or splitting skills
  • Don't commit without running python3 scripts/validate_skills.py
  • Don't leave sources: []: use [original] for synthesis
  • Don't use lua; use luau
  • Don't add ## Overview sections; use ## When to Load → ## Quick Reference directly
  • Don't create standalone reference files; all content goes in references/full.md
  • Don't exceed 50,000 chars in reference files
  • Don't use github.com/blob/ URLs; use raw.githubusercontent.com

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.