Skip to content

feat: add Cocos Creator 3.8 engine support (4 agents + engine reference docs)#95

Open
realUC wants to merge 2 commits into
Donchitos:mainfrom
realUC:main
Open

feat: add Cocos Creator 3.8 engine support (4 agents + engine reference docs)#95
realUC wants to merge 2 commits into
Donchitos:mainfrom
realUC:main

Conversation

@realUC

@realUC realUC commented Jun 27, 2026

Copy link
Copy Markdown

Summary

Adds Cocos Creator 3.8 as a fully-supported engine alongside Godot 4, Unity, and Unreal Engine 5. The new engine set follows the existing tiered specialist pattern (1 lead + 3 sub-specialists) and is wired into the same routing, setup, and testing infrastructure.

This brings the engine specialist table from three to four major engines, with Cocos Creator's strengths (mini-game platforms, TypeScript-first component model, lightweight web/mobile build) available to projects that need them.

Motivation

Cocos Creator is widely used for mini-games (WeChat / ByteDance / Alipay / Baidu / Honor) and lightweight cross-platform titles — a segment not well served by the existing Godot/Unity/Unreal sets. Because the LLM knowledge cutoff predates Cocos Creator 3.8.x (Spine 4.2 support, Box2D JSB native binding, restored node.x/node.y accessors), version-pinned reference docs are included so agents give correct, non-deprecated guidance rather than hallucinating pre-3.0 APIs (e.g. cc.loader, cc.Class).

What's Added

1. Engine Specialist Agents (4 new)

All four include the required Collaboration Protocol section (ask → options → decision → draft → approval) and documented domain boundaries.

Agent Role
cocos-specialist Lead — engine routing, project structure, multi-platform build config
cocos-ts-specialist TypeScript components, @ccclass/@property decorators, assetManager async loading, event bus
cocos-shader-specialist Effect files (CCEffect / CCProgram), built-in uniforms, Custom Render Pipeline
cocos-ui-specialist UITransform, Layout, Widget, Canvas, multi-resolution adaptation, mini-game UI constraints

2. Engine Reference Documentation (11 new)

Pinned to Cocos Creator 3.8.6, under docs/engine-reference/cocos/:

  • VERSION.md — version pin + knowledge-gap baseline
  • breaking-changes.md — 3.x migration notes
  • deprecated-apis.md — deprecated APIs + replacements (cc.loaderassetManager, cc.Class → ES classes, etc.)
  • current-best-practices.md — modern project structure, component pattern, async loading, object pooling, multi-platform config
  • modules/animation, audio, input, navigation, networking, physics, rendering, ui

The rendering.md doc codifies the rule that built-in uniforms (cc_matViewProj, cc_matWorld, cc_matWorldIT, cc_cameraPos, cc_time, cc_screenSize, …) are engine-injected and must NOT be redeclared in Effect files — redeclaration causes black screen / compile errors. The shader specialist agent and its test spec are aligned to enforce this.

3. Testing Specs (4 new)

Under CCGS Skill Testing Framework/agents/engine/cocos/, mirroring the existing engine test structure. Each has 5 behavioral cases covering correctness, anti-patterns, and critical errors. Registered in catalog.yaml with last_updated bumped.

4. Framework Wiring (modified)

  • .claude/skills/setup-engine/SKILL.md — Cocos Creator added to engine options, with honest tradeoffs, Technology Stack template, naming conventions, file-extension routing, and knowledge-gap baseline block.
  • CLAUDE.md — Cocos Creator 3.8 + TypeScript added to Technology Stack choices.
  • README.md — Cocos Creator row added to the Engine Specialists table; "all three major engines" → "all four major engines".
  • CCGS Skill Testing Framework/README.mdcocos/ directory added to folder structure.
  • CCGS Skill Testing Framework/skills/utility/setup-engine.md — Case 3b added for Cocos mini-game platform config.

Files Changed

26 files, +2826 / −14.

Area Files
Agents (new) cocos-specialist, cocos-ts-specialist, cocos-shader-specialist, cocos-ui-specialist
Engine reference (new) 11 docs under docs/engine-reference/cocos/
Test specs (new) 4 under CCGS Skill Testing Framework/agents/engine/cocos/
Framework (modified) setup-engine/SKILL.md, CLAUDE.md, README.md, catalog.yaml, testing README, testing setup-engine.md

Testing

Verified end-to-end in a Claude Code session:

  1. Agent structure — all 4 agent files contain the required Collaboration Protocol section and respect domain boundaries (no cross-domain file modification without delegation).
  2. Routing/setup-engine correctly presents Cocos Creator as a fourth option and routes .ts / .effect / .scene files to the right sub-specialist.
  3. Reference docscatalog.yaml validates; all 4 Cocos test specs resolve to existing files.
  4. Bug fix verification — the shader specialist's example Effect file and its test spec were checked for the built-in uniform redeclaration issue. Example code uses cc_matViewProj / cc_matWorld directly without redeclaring them, and the test spec treats redeclaration as a critical error (Case 3). The two are consistent.

I did not run the hook suite against Cocos-specific paths since the change adds agent/reference files (no new hooks); existing hooks continue to pass on the modified framework files.

CONTRIBUTING.md Compliance

  • Agents include a Collaboration Protocol section
  • Agents don't modify files outside their documented domain without explicit delegation
  • Reference index updated (catalog.yaml last_updated bumped; README Engine Specialists table updated; CLAUDE.md Technology Stack updated)
  • No new skills/skills-flat-md; existing setup-engine skill extended in place
  • Conventional Commit message (feat:)
  • Platform-neutral — docs/agents only, no shell/grep changes

Notes / Caveats

  • Cocos Creator is optional — projects not using it incur no overhead; the agents are only loaded when routed to.
  • The reference docs are intentionally version-pinned to 3.8.6. If the maintainer prefers a different pin (e.g. 3.8.x latest), the VERSION.md baseline is the single place to adjust.
  • I kept the change additive and avoided touching unrelated engine sets or the studio hierarchy. If a different naming/layout convention is preferred (e.g. cocoscreator- prefix vs cocos-), happy to rename.

Thanks for considering this — happy to address any review feedback.

Donchitos and others added 2 commits June 27, 2026 17:05
…tructure, polish (Donchitos#50)

* Add /vertical-slice skill, prototype overhaul, and workflow integration

- Add /vertical-slice skill for pre-production validation (Phase 4 gate)
- Overhaul /prototype skill with two-mode design: concept prototype (Phase 1)
  vs vertical slice (Phase 4), with clearer differentiation and higher standards for VS
- Update prototyper agent to own both prototype and vertical-slice workflows
- Add prototype-report.md and vertical-slice-report.md output templates
- Update WORKFLOW-GUIDE, quick-start, skills-reference, agent-coordination-map,
  and skill-flow-diagrams to fully integrate both skills into the 7-phase pipeline
- Remove orphaned empty quick-prototype/ directory

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* sync v1 counts + polish

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Add entity inventory flow, relax vertical-slice gate, improve UX authoring prompts

- /asset-spec: new Phase 0b entity & screen inventory when no argument and no
  existing inventory — reads GDDs/art-bible, proposes categorized list, writes
  design/assets/entity-inventory.md collaboratively
- /asset-spec: entity/character target falls back to inline user description
  when no source doc exists, rather than failing
- /gate-check: vertical slice changed from blocking to CONCERNS-only when
  absent; built-but-broken slice still fails; adds entity inventory as gate artifact
- /ux-design: convert inline approval prompts to AskUserQuestion for structured
  option capture at key authoring decision points
- workflow-catalog.yaml: entity-inventory step added to pre-production; UX spec
  min_count raised to 3; vertical-slice and prototype marked required: false with
  updated descriptions
- .gitignore: exclude marrow/ eval tooling directory

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Add missing AskUserQuestion widgets to 7 skills

Audit found 11 decision points across 7 skills where structured option
prompts were missing — using plain text, auto-selection, or no gate at all.

Skills patched:
- create-epics: per-epic approval + producer CONCERNS verdict
- sprint-plan: producer CONCERNS verdict with scope/timeline options
- milestone-review: AT RISK / OFF TRACK producer verdicts require acknowledgement
- retrospective: existing-retro handling converted from plain text [A]/[B]
- quick-design: classification confirmation + draft approve/revise/redirect
- tech-debt add mode: category (6 options) + effort (S/M/L/XL) structured capture
- regression-suite: no-arg mode selection instead of silent auto-detect
- hotfix: severity confirmation gate before workflow begins

Also added AskUserQuestion to allowed-tools headers for retrospective,
quick-design, tech-debt, regression-suite, and hotfix.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Prep v1 stable: fix WORKFLOW-GUIDE counts, stale agent names, and skill model fields

- WORKFLOW-GUIDE.md: correct agent count (48→49), skill count (66/68→73),
  add 6 missing skills to Appendix B, fix Creative category count (2→4),
  replace 3 non-existent agent names with correct ue-*/unity-* specialists,
  add missing godot-csharp/gdextension specialists to hierarchy,
  fix production/stories/ paths → production/epics/
- coordination-rules.md: replace "not yet used" with opt-in env var note
- quick-start.md: rename duplicate "Validate the concept" label → "Prototype the mechanic"
- skill-flow-diagrams.md: remove duplicate legacy UX pipeline section
- All 62 skills missing model: field now have explicit model: sonnet

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix: comprehensive skill audit — consistency, UX, and flow gaps

Two-pass audit fixing ~35 bugs across 41 files.

Pre-production flow:
- Brainstorm next-steps split into Path A (design-first) and Path B
  (prototype-first) — eliminates "prototype after architecture" confusion
- /architecture-review added to pre-production flow in brainstorm and
  create-architecture handoffs
- gate-check traceability check corrected to requirements-traceability.md
- dev-story TR registry error now points to /architecture-review (not /create-epics)
- start now writes production/stage.txt on first onboarding

AskUserQuestion gaps filled:
- balance-check, code-review, hotfix, day-one-patch, consistency-check
  all gain closing widgets and/or missing allowed-tools declarations
- hotfix git branch creation now requires user confirmation
- sprint-plan review-mode setup moved to Phase 0 (before gates run)
- team-combat gains architecture→implementation approval gate
- design-review APPROVED path consolidated from 3 widgets to 1 multiSelect

All 9 team-* skills:
- Phase 0 review-mode resolution added (solo/lean/full now respected)
- team-audio output path fixed (design/gdd/ → design/audio/)
- team-level final doc compilation delegated to level-designer subagent
- team-narrative localization-lead added to composition list
- team-qa sprint path fixed (flat files, not directories)
- team-release NO-GO override captures written justification
- team-live-ops Cancel verdict now explicitly BLOCKED

Other fixes:
- Art bible path standardized to design/art/art-bible.md (3 wrong refs)
- AD-PHASE-GATE added to lean-mode skip list in director-gates.md
- design-system duplicate 5d heading fixed; skeleton decline path added;
  mandatory agent spawns now respect review mode
- story-readiness acceptance criteria thresholds now type-aware
- create-stories gains multi-ADR and no-ADR handling guidance
- consistency-check creates docs/consistency-failures.md on first run
- retrospective frontmatter bash injection replaced with explicit Bash call
- smoke-check ls -t gains PowerShell fallback
- Conventional Commits format documented in coding-standards.md
- gate-check: ADR acceptance gate, QA plan check, chain-of-verification
  tool-action requirement all added

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix: expose --review flag in argument-hints for all team-* skills

All 9 team-* skills already implement Phase 0 review-mode resolution
internally (full/lean/solo), but none advertised [--review full|lean|solo]
in their argument-hint. Users had no way to discover the per-run override.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* docs: add SECURITY.md with coordinated disclosure policy

Defines scope, reporting process (GitHub private vulnerability reporting),
contributor security guidelines for hooks/skills/agents, and 90-day
coordinated disclosure timeline.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* docs: add CONTRIBUTING.md with framework contribution guidelines

Covers what PRs are welcome, skill/hook/agent technical requirements,
the collaborative principle, testing expectations, commit format,
and platform compatibility requirements.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* docs: add v1.0.0-beta → v1.0 upgrade section to UPGRADING.md

Documents the 17 commits since the beta tag: new /vertical-slice gate,
entity inventory flow in /map-systems, AskUserQuestion widgets across
7 skills, --review flag exposure on team-* skills, bug fixes
(Donchitos#21, Donchitos#36, Donchitos#42, Donchitos#43, Donchitos#45), and the new CONTRIBUTING.md and SECURITY.md.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
@realUC realUC requested a review from Donchitos as a code owner June 27, 2026 09:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants