@jstn-sdk/rcs 0.1.6 → 0.1.8

package/docs/reference/multi-agent-compatibility-architecture.md

@@ -0,0 +1,120 @@
+# Multi-Agent Compatibility Architecture
+
+RCS already ships a real multi-agent compatibility system, but in this repo it is split across **native agent definitions**, **catalog policy**, **native Codex TOML generation**, and **adapter targets**.
+
+This document names that architecture explicitly so contributors do not project an external `src/agents.ts` example onto the wrong files.
+
+## Canonical files
+
+- Native agent registry: [`src/agents/definitions.ts`](../../src/agents/definitions.ts)
+- Native-agent install policy: [`src/agents/policy.ts`](../../src/agents/policy.ts)
+- Native-agent config adapter/generator: [`src/agents/native-config.ts`](../../src/agents/native-config.ts)
+- Public catalog manifest: [`src/catalog/manifest.json`](../../src/catalog/manifest.json)
+- External adapter target contracts: [`src/adapt/contracts.ts`](../../src/adapt/contracts.ts)
+- External adapter target registry: [`src/adapt/registry.ts`](../../src/adapt/registry.ts)
+- CLI native-agent management: [`src/cli/agents.ts`](../../src/cli/agents.ts)
+- Native-agent verification: [`src/scripts/verify-native-agents.ts`](../../src/scripts/verify-native-agents.ts)
+
+## Core patterns
+
+### Registry pattern
+
+RCS uses **two registries**:
+
+- `AGENT_DEFINITIONS` in `src/agents/definitions.ts`
+- target descriptors in `src/adapt/registry.ts`
+
+Those registries keep agent and adapter metadata centralized instead of scattering role-specific logic through setup, runtime, and verification code.
+
+### Adapter pattern
+
+RCS does not expose prompt markdown or external target state directly.
+
+Instead:
+
+- `src/agents/native-config.ts` adapts prompt markdown plus `AgentDefinition` metadata into native Codex TOML files
+- `src/adapt/*` adapts target-specific runtime evidence such as OpenClaw or Hermes into a shared envelope/probe/status contract
+
+The rest of the system can therefore consume a stable RCS-owned shape rather than raw target-specific details.
+
+### Strategy pattern
+
+The compatibility layer applies strategy-like decisions through registry-backed metadata instead of giant conditionals:
+
+- `modelClass`, `posture`, and `routingRole` select native-agent instruction overlays and model resolution behavior
+- adapter target descriptors select target-specific capability reports while preserving a shared foundation contract
+- catalog `status` values such as `active`, `internal`, `alias`, and `merged` determine installability and canonical-target behavior
+
+### Plugin-style architecture
+
+RCS supports extension without rewriting the core native-agent installer:
+
+- add or adjust a native role in `AGENT_DEFINITIONS`
+- classify it in `src/catalog/manifest.json`
+- provide the matching prompt asset in `prompts/`
+- let `verify-native-agents` enforce the contract
+
+For external targets, add a target contract and registry descriptor under `src/adapt/*` without changing unrelated adapter lanes.
+
+## Supporting techniques
+
+### Configuration-driven design
+
+The compatibility layer is mostly declarative:
+
+- agent metadata is data-first in `AGENT_DEFINITIONS`
+- installability and canonical aliases are data-first in `src/catalog/manifest.json`
+- adapter capabilities are data-first in `TARGET_DESCRIPTORS`
+
+That keeps policy visible and reviewable.
+
+### Facade pattern
+
+RCS exposes small facade helpers over the registries instead of forcing consumers to walk raw objects:
+
+- `getAgent`
+- `listAgents`
+- `getAgentNames`
+- `getAgentsByCategory`
+- `getAgentsByPosture`
+- `getAgentsByRoutingRole`
+- `getInstallableNativeAgentNames`
+- `listAdaptTargets`
+- `getAdaptTargetDescriptor`
+
+### Canonical-boundary pattern
+
+RCS keeps ownership boundaries explicit:
+
+- prompt markdown is setup-owned source content under `prompts/`
+- generated native Codex agent TOML is emitted under `~/.codex/agents/` or `./.codex/agents/`
+- plugin manifests must **not** claim setup-owned `agents`, `prompts`, or `hooks`
+- adapter writes stay under `.rcs/adapters/<target>/...`
+
+This avoids duplication and prevents plugin/runtime surfaces from silently fighting over the same artifacts.
+
+## What this architecture is not
+
+This repo does **not** currently use the external example's exact shape:
+
+- no monolithic `src/agents.ts`
+- no `.agents/skills` universal-directory + symlink installer as the core native-agent mechanism
+- no per-agent `detectInstalled()` registry for dozens of third-party coding tools as the main compatibility layer
+
+RCS instead centers on:
+
+- native Codex agent-role generation
+- catalog-governed canonical/alias boundaries
+- adapter-target compatibility lanes such as OpenClaw and Hermes
+
+## Verification contract
+
+The architecture is guarded by tests and verification surfaces:
+
+- [`src/agents/__tests__/definitions.test.ts`](../../src/agents/__tests__/definitions.test.ts)
+- [`src/agents/__tests__/native-config.test.ts`](../../src/agents/__tests__/native-config.test.ts)
+- [`src/adapt/__tests__/foundation.test.ts`](../../src/adapt/__tests__/foundation.test.ts)
+- [`src/scripts/__tests__/verify-native-agents.test.ts`](../../src/scripts/__tests__/verify-native-agents.test.ts)
+- [`src/verification/__tests__/multi-agent-compatibility-architecture.test.ts`](../../src/verification/__tests__/multi-agent-compatibility-architecture.test.ts)
+
+If you change the compatibility architecture, update the doc and the contract tests together.

package/docs/reference/rcs-config-schema-routing.md

@@ -29,6 +29,8 @@ Current code recognizes these top-level `.rcs-config.json` keys:
 | --- | --- | --- |
 | `env` | Object of non-empty string values | Fallback environment values for model routing and helper launch paths. Model-related supported keys are listed below. |
 | `models` | Object of non-empty string values | Mode defaults and low-complexity model aliases. Supported model-routing keys are listed below. |
+| `providers` | Object | Provider profiles for model/API-agnostic routing (`base_url`, `api_format`, `env_key`, `capabilities`, `label`). |
+| `routing` | Object | Default provider, mode-specific provider mapping, fallback provider chains, and hot-swap/failover flags. |
 | `notifications` | Object | Notification transports, profiles, templates, cooldowns, replies, and OpenClaw/custom aliases. See the notification summary below and the OpenClaw guide for full examples. |
 | `stopHookCallbacks` | Legacy object | Backward-compatible legacy session-end notification config for `telegram` and `discord`; prefer `notifications`. |
 | `promptRouting` | `{ "triage": { "enabled": boolean } }` | Enables/disables advisory triage prompt routing. Missing key defaults to enabled; malformed shape fails closed to disabled. |
@@ -251,6 +253,7 @@ If behavior does not match your config, first confirm whether you are in user or
 ## Related docs and source surfaces
 
 - Notification and OpenClaw config: [`docs/openclaw-integration.md`](../openclaw-integration.md)
+- Provider abstraction: [`docs/reference/llm-provider-abstraction.md`](./llm-provider-abstraction.md)
 - Project wiki config: [`docs/reference/project-wiki.md`](./project-wiki.md)
 - Model routing source: `src/config/models.ts`
 - Notification config source: `src/notifications/config.ts`, `src/notifications/types.ts`, `src/notifications/hook-config-types.ts`

package/docs/reference/roblox-mcp-reference-layer.md

@@ -0,0 +1,75 @@
+# Roblox MCP Reference Layer
+
+Status: canonical external reference-source policy for Roblox-facing RCS work.
+
+## Purpose
+
+RCS has two different MCP/reference layers:
+
+1. **First-party RCS MCP servers**
+   - local stdio servers for state, memory, code-intel, trace, and wiki
+2. **External Roblox reference sources**
+   - official Creator Docs
+   - serverless `gitmcp.io` mirrors
+   - Roblox-specific skill repositories
+   - implementation corpora
+
+This file defines how those external sources should be used.
+
+## Canonical Source Order
+
+### Tier 1: Official platform truth
+
+Use these first when the question is about correctness, APIs, platform behavior, or Creator Hub rules:
+
+- `https://github.com/Roblox/creator-docs`
+- `https://gitmcp.io/Roblox/creator-docs`
+
+These define the canonical platform truth.
+
+### Tier 2: High-signal Roblox implementation references
+
+Use these to improve implementation awareness after the official baseline is grounded:
+
+- `https://github.com/sentinelcore/roblox-skills`
+- `https://gitmcp.io/sentinelcore/roblox-skills`
+- `https://github.com/greedychipmunk/agent-skills/tree/main/roblox-game-developer`
+- `https://github.com/omer-metin/skills-for-antigravity/tree/main/skills/roblox-development`
+- `https://github.com/dig1t/skills`
+- `https://github.com/Corecii/Devprod`
+- `https://gitmcp.io/Corecii/Devprod`
+
+### Tier 3: Raw script corpora and broad pattern mining
+
+Use these only as inspiration or anti-pattern review, never as canonical truth:
+
+- `https://github.com/retpirato/Roblox-Scripts`
+- `https://gitmcp.io/retpirato/Roblox-Scripts`
+- `https://gitmcp.io/frosteen/Roblox_LUA_Weapon_Scripts`
+- `https://gitmcp.io/uhub/awesome-lua`
+- `https://gitmcp.io/LewisJEllis/awesome-lua`
+- `https://gitmcp.io/forhappy/awesome-lua`
+- `http://lua-users.org/wiki/SampleCode`
+
+## Usage Rules
+
+- Official Roblox docs decide correctness.
+- Skill repositories improve implementation strategy, workflow framing, and checklist quality.
+- Raw script corpora are weak signals only; treat them as inspiration or anti-pattern mining.
+- Never present a third-party script repository as if it overrides official Roblox platform guidance.
+- Prefer the serverless `gitmcp.io` mirror when fast MCP-style browsing helps, but do not confuse the mirror with official product ownership.
+
+## RCS Fit
+
+This repo should treat the layer like this:
+
+- **RCS first-party MCP** = local runtime/state/control plane
+- **Creator Docs / `gitmcp.io`** = external Roblox platform truth
+- **Roblox skill repos** = implementation guidance and workflow references
+- **Raw script corpora** = non-canonical pattern support only
+
+## Template
+
+Reference inventory shipped with the repo:
+
+- `templates/roblox/reference-sources.md`

package/docs/reference/roblox-pre-action-protocol.md

@@ -45,10 +45,18 @@ Priority 1: canonical platform truth
 - `https://gitmcp.io/Roblox/creator-docs`
 
 Priority 2: high-signal implementation references
+- `https://github.com/sentinelcore/roblox-skills`
 - `https://gitmcp.io/retpirato/Roblox-Scripts`
 - `https://gitmcp.io/sentinelcore/roblox-skills`
-- `https://gitmcp.io/frosteen/Roblox_LUA_Weapon_Scripts`
+- `https://github.com/greedychipmunk/agent-skills/tree/main/roblox-game-developer`
+- `https://github.com/omer-metin/skills-for-antigravity/tree/main/skills/roblox-development`
+- `https://github.com/dig1t/skills`
+- `https://github.com/Corecii/Devprod`
 - `https://gitmcp.io/Corecii/Devprod`
+
+Priority 3: raw script and corpus support
+- `https://github.com/retpirato/Roblox-Scripts`
+- `https://gitmcp.io/frosteen/Roblox_LUA_Weapon_Scripts`
 - `https://gitmcp.io/uhub/awesome-lua`
 - `https://gitmcp.io/LewisJEllis/awesome-lua`
 - `https://gitmcp.io/forhappy/awesome-lua`
@@ -61,11 +69,15 @@ Priority 3: dataset and corpus support
 - `https://datasets-server.huggingface.co/splits?dataset=TorpedoSoftware%2FRoblox-Luau-Reasoning-v1.0`
 - `https://huggingface.co/datasets/bartholomort/lua-obfuscator-corpus`
 
-Rules:
+Reference policy:
 - official Roblox docs define correctness
-- repos improve implementation awareness only
+- Roblox skill repos improve implementation awareness only
 - datasets are weak pattern support only
 
+Canonical reference-layer policy is documented in:
+- `docs/reference/roblox-mcp-reference-layer.md`
+- `templates/roblox/reference-sources.md`
+
 ## Phase 2: Understanding Synthesis
 
 Before implementation, answer:

package/docs/reference/semantic-design-system.md

@@ -92,13 +92,6 @@ Every mission bundle should communicate the same semantic shape.
 9. reference layers
 10. vocabulary guardrail
 
-## Archive Rule
-
-Historical surfaces may describe deleted/generic role labels, but:
-- they should be clearly archival
-- they should not be used as active onboarding
-- they should not be required by active runtime/catalog behavior
-
 ## Enforcement
 
 The repo should keep lightweight tests that verify:

package/docs/reference/surface-map.md

@@ -114,18 +114,9 @@ Sources:
 | `prompts/writer.md` | `prompt` | `creator-runtime` | `internal` | writer |
 | `skills/worker/SKILL.md` | `skill` | `creator-runtime` | `internal` | worker |
 
-## Historical Surfaces
-
-| Path | Type | Domain | Audience | Surface |
-| --- | --- | --- | --- | --- |
-| `skills/forge-init/SKILL.md` | `skill` | `archive` | `archive` | forge-init |
-| `skills/swarm/SKILL.md` | `skill` | `archive` | `archive` | swarm |
-| `skills/web-clone/SKILL.md` | `skill` | `archive` | `archive` | web-clone |
-
 ## Additional Repo Lanes
 
 - `AGENTS.md` and `templates/AGENTS.md` remain the orchestration brain and guidance authority.
-- `docs/archive/` is the historical lane; archive content may describe removed surfaces but must not act as active onboarding.
 - `src/catalog/manifest.json`, `templates/catalog-manifest.json`, and `src/catalog/generated/public-catalog.json` are the catalog source/generated contract lane.
 - `dist/` is generated runtime output and should reflect the taxonomy, not redefine it.
 

package/docs/release-notes-v0.1.7.md

@@ -0,0 +1,20 @@
+# Roblox Creator Skills v0.1.7
+
+## Summary
+
+`v0.1.7` is the doc/prompt hardening release after the `0.1.6` automation update. It removes the last archive lane, rewrites the demo surface around Roblox-first creator workflows, strengthens Roblox-native framing inside the prompt system, and replaces the placeholder localized README stubs with real translated bodies.
+
+## What changed
+
+- Removed `docs/archive/` from the project surface
+- Reduced active `docs/qa/` to the live gate docs only
+- Rewrote `DEMO.md` into a Roblox-first demo guide
+- Tightened prompt identities/goals for Roblox-native framing across key internal prompts
+- Replaced localized README placeholder stubs with actual localized content
+
+## Verification
+
+- `npm run build`
+- prompt-guidance and taxonomy contract tests
+- semantic/workspace/robloxstudio-mcp verification tests
+- `node dist/scripts/surface-taxonomy.js`

package/docs/release-notes-v0.1.8.md

@@ -0,0 +1,24 @@
+# Roblox Creator Skills v0.1.8
+
+## Summary
+
+`v0.1.8` is the MCP/reference-layer activation release after the `0.1.7` docs and localization hardening line. It makes the GitMCP Roblox reference sources part of the default managed Codex setup, documents the activation split more clearly, and adds the tests needed to keep that default layer stable.
+
+## What changed
+
+- Default managed Codex config now includes:
+  - `creator_docs`
+  - `roblox_skills`
+  - `devprod_docs`
+  - `roblox_scripts_corpus`
+- Added canonical docs for the Roblox MCP reference layer
+- Added a shipped reference-source template inventory
+- Clarified in README and localized READMEs that:
+  - `rcs mcp-serve` is for first-party local RCS MCP servers
+  - GitMCP Roblox references should stay enabled by default
+  - `robloxstudio-mcp` remains a manual live Studio bridge
+
+## Verification
+
+- `npm run build`
+- `node --test dist/config/__tests__/roblox-reference-mcp.test.js dist/config/__tests__/generator-idempotent.test.js dist/verification/__tests__/roblox-mcp-reference-layer.test.js dist/verification/__tests__/robloxstudio-mcp-compatibility.test.js`

package/docs/wiki/Contributing.md

@@ -0,0 +1,45 @@
+# Contributing Wiki
+
+This page complements the repository root [CONTRIBUTING.md](../../CONTRIBUTING.md).
+
+## Fastest way to start
+
+1. Read the root [README](../../README.md).
+2. Run the local setup flow.
+3. Look for issues labeled `good first issue` or `help wanted`.
+4. Pick a small, self-contained task before touching larger workflow/runtime surfaces.
+
+## What maintainers should optimize for
+
+- a newcomer should be able to install, build, and smoke-test the repo in a few minutes
+- issue descriptions should include clear reproduction or acceptance criteria
+- docs, localization, wiki, and QA contributions count as real contributions
+- roadmap items should be broken into slices that can be finished without deep repo archaeology
+
+## Accepted contribution types
+
+- code and tests
+- docs and onboarding fixes
+- localization updates
+- issue triage and reproduction refinement
+- release notes and changelog cleanup
+- contributor-experience improvements
+
+## Mentorship posture
+
+- explain repo-local vocabulary when it is not obvious
+- redirect contributors to the right doc instead of assuming they already know the surface
+- prefer scoped guidance over vague “needs more work” reviews
+- keep review feedback factual and actionable
+
+## Labels that matter
+
+- `good first issue`
+- `help wanted`
+- `docs`
+- `localization`
+- `wiki`
+- `contributor experience`
+- `release`
+
+See [Good first issues and labels](./Good-First-Issues.md).

package/docs/wiki/Good-First-Issues.md

@@ -0,0 +1,44 @@
+# Good First Issues and Labels
+
+This page defines the maintainer standard for contributor-friendly task shaping.
+
+## Required labels
+
+- `good first issue` — small, safe, self-contained task with explicit acceptance criteria
+- `help wanted` — useful task that is open to outside contribution, but may require more repo context
+- `docs` — documentation surface
+- `localization` — translation or locale-sync work
+- `wiki` — contributor wiki or runtime wiki documentation surface
+- `contributor experience` — onboarding, templates, setup, or workflow ergonomics
+
+## What qualifies as a good first issue
+
+- one surface or subsystem
+- no hidden cross-repo dependency
+- clear reproduction or acceptance criteria
+- easy local verification path
+- no risky release, auth, or multi-runtime migration work
+
+## Good first issue examples in this repo
+
+- clarify a Roblox-first prompt or skill example
+- improve docs or setup wording
+- sync README locale drift after a root README change
+- add a missing focused test for a small contract
+- fix a small CLI/help text mismatch
+- tighten release notes or contributor guidance
+
+## What should not be labeled good first issue
+
+- large runtime refactors
+- release workflow surgery without a focused scope
+- tmux/team-state concurrency work with unclear blast radius
+- broad prompt taxonomy rewrites
+- changes that require maintainer-only credentials or environment access
+
+## Maintainer checklist before applying the label
+
+- task is broken down enough to fit in one PR
+- expected files are named
+- verification path is stated
+- hidden blockers are removed or called out

package/docs/wiki/Home.md

@@ -0,0 +1,35 @@
+# Contributor Wiki
+
+This is the contributor-facing wiki source for `JustineDevs/roblox-ai-os`.
+
+It is **not** the same thing as the local runtime wiki under `.rcs/wiki/`.
+Use this wiki for community guidance, roadmap context, issue triage, and release operations.
+
+## Start here
+
+- [Contributing guide](./Contributing.md)
+- [Good first issues and labels](./Good-First-Issues.md)
+- [Roadmap](./Roadmap.md)
+- [Release playbook](./Release-Playbook.md)
+
+## What this wiki is for
+
+- lowering the barrier to first contribution
+- giving contributors a stable place to understand priorities
+- showing maintainers how to label and scope newcomer-friendly work
+- documenting release and update flows without forcing people to reverse-engineer workflow files
+
+## Maintainer expectations
+
+- keep issues small, explicit, and easy to test
+- use `good first issue` and `help wanted` labels deliberately
+- respond quickly when a contributor asks for clarification
+- thank contributors in release notes, README acknowledgements, or follow-up docs when their work materially lands
+- accept non-code contributions such as docs, localization, QA, roadmap refinement, and issue triage
+- use an All Contributors-style mindset so contribution credit is not limited to code
+
+## Source of truth
+
+- Canonical public entry point: [../../README.md](../../README.md)
+- Contributor workflow contract: [../../CONTRIBUTING.md](../../CONTRIBUTING.md)
+- Roadmap source for contributor planning: [./Roadmap.md](./Roadmap.md)

package/docs/wiki/Release-Playbook.md

@@ -0,0 +1,57 @@
+# Release Playbook
+
+This page explains the public release surfaces for RCS.
+
+## Three separate release surfaces
+
+### 1. npm package
+
+Public package:
+
+- `@jstn-sdk/rcs`
+
+Registry:
+
+- `https://registry.npmjs.org`
+
+Purpose:
+
+- what end users install with `npm install -g @jstn-sdk/rcs`
+
+### 2. GitHub Packages
+
+Registry:
+
+- `https://npm.pkg.github.com`
+
+Purpose:
+
+- populates the repository Packages tab
+- keeps GitHub-native package visibility aligned with the repo
+
+### 3. GitHub Releases
+
+Purpose:
+
+- creates the Releases page entry
+- attaches release notes and binary assets to a tag
+
+Important:
+
+- a git tag alone is not a GitHub Release object
+- a successful npm publish alone does not populate GitHub Releases or GitHub Packages
+
+## Maintainer release expectations
+
+- verify Actions pre-flight before sign-off
+- keep changelog and release notes aligned
+- create or update the GitHub Release object from the tag workflow
+- publish npm and GitHub Packages from the workflow lanes, not by assumption
+
+## Contributor-facing outputs
+
+- release notes
+- changelog entry
+- package version bump
+- tagged release
+- attached native assets when applicable

package/docs/wiki/Roadmap.md

@@ -0,0 +1,90 @@
+# Roadmap
+
+This is the current contributor-facing roadmap for RCS.
+
+## Current priority themes
+
+- contributor experience and community hygiene
+- release reliability across GitHub Releases, GitHub Packages, and npm
+- Roblox-first prompt and skill quality
+- Studio workspace and reference-layer grounding
+
+## Next roadmap
+
+### Worldbuilding and design production skills
+
+Planned outputs:
+
+- moodboards
+- GDD structure
+- story outline
+- character sheets
+- map chunk specs
+- weekly implementation plans
+
+Why it matters:
+
+- helps RCS support pre-production and design-system work, not just coding
+- gives creators a disciplined planning path before implementation
+
+### Roblox monetization planner
+
+Planned outputs:
+
+- game pass ideas
+- dev product loops
+- subscription perks
+- private server strategy
+- Creator Rewards fit by genre
+
+Why it matters:
+
+- pushes monetization design into explicit, reviewable planning instead of ad hoc prompting
+
+### Patch assistant
+
+Planned outputs:
+
+- update notes
+- devlog copy
+- QA checklist
+- release readiness review
+
+Why it matters:
+
+- turns merged work into ship-ready communication and operational follow-through
+
+### Audience and retention planner
+
+Planned outputs:
+
+- target player habits
+- session length expectations
+- social hooks
+- replay loops
+
+Why it matters:
+
+- makes retention design a first-class creator workflow instead of an afterthought
+
+### Live ops mode
+
+Planned outputs:
+
+- weekly update planning
+- event cadence
+- private-server hooks
+- seasonal content beats
+- acquisition loops tied to Roblox monetization and Creator Rewards
+
+Why it matters:
+
+- aligns content cadence, retention, and monetization into one explicit operating surface
+
+## Contributor note
+
+If you want to help, prefer turning one roadmap item into:
+
+1. a scoped issue
+2. clear acceptance criteria
+3. one narrow implementation slice

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jstn-sdk/rcs",
-  "version": "0.1.6",
+  "version": "0.1.8",
   "description": "Roblox Creator Skills toolkit and workflow layer for Codex CLI",
   "type": "module",
   "main": "dist/index.js",

package/plugins/roblox-ai-os-creator-skills/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "roblox-ai-os-creator-skills",
-  "version": "0.1.6",
+  "version": "0.1.8",
   "description": "Roblox Creator Skills workflow and runtime bundle for Codex CLI",
   "author": {
     "name": "JustineDevs",

package/plugins/roblox-ai-os-creator-skills/docs/reference/roblox-pre-action-protocol.md

@@ -45,10 +45,18 @@ Priority 1: canonical platform truth
 - `https://gitmcp.io/Roblox/creator-docs`
 
 Priority 2: high-signal implementation references
+- `https://github.com/sentinelcore/roblox-skills`
 - `https://gitmcp.io/retpirato/Roblox-Scripts`
 - `https://gitmcp.io/sentinelcore/roblox-skills`
-- `https://gitmcp.io/frosteen/Roblox_LUA_Weapon_Scripts`
+- `https://github.com/greedychipmunk/agent-skills/tree/main/roblox-game-developer`
+- `https://github.com/omer-metin/skills-for-antigravity/tree/main/skills/roblox-development`
+- `https://github.com/dig1t/skills`
+- `https://github.com/Corecii/Devprod`
 - `https://gitmcp.io/Corecii/Devprod`
+
+Priority 3: raw script and corpus support
+- `https://github.com/retpirato/Roblox-Scripts`
+- `https://gitmcp.io/frosteen/Roblox_LUA_Weapon_Scripts`
 - `https://gitmcp.io/uhub/awesome-lua`
 - `https://gitmcp.io/LewisJEllis/awesome-lua`
 - `https://gitmcp.io/forhappy/awesome-lua`
@@ -61,11 +69,15 @@ Priority 3: dataset and corpus support
 - `https://datasets-server.huggingface.co/splits?dataset=TorpedoSoftware%2FRoblox-Luau-Reasoning-v1.0`
 - `https://huggingface.co/datasets/bartholomort/lua-obfuscator-corpus`
 
-Rules:
+Reference policy:
 - official Roblox docs define correctness
-- repos improve implementation awareness only
+- Roblox skill repos improve implementation awareness only
 - datasets are weak pattern support only
 
+Canonical reference-layer policy is documented in:
+- `docs/reference/roblox-mcp-reference-layer.md`
+- `templates/roblox/reference-sources.md`
+
 ## Phase 2: Understanding Synthesis
 
 Before implementation, answer:

package/plugins/roblox-ai-os-creator-skills/skills/ai-slop-cleaner/SKILL.md

@@ -141,6 +141,3 @@ Remaining Risks:
 **Bad:** Keep a `fallback if it fails` branch that silently defaults after a swallowed error instead of fixing the root cause or making failure explicit.
 
 **Good:** A version-specific compatibility shim is narrow, documented, preserves error evidence, has primary and fallback regression tests, and is reported as a grounded compatibility/fail-safe fallback.
-surface-class: "operator"
-domain: "creator-runtime"
-audience: "operator"