一键导入
omg-kit
Maintain Oh My Game Kit repos as a kit maintainer. Use for releasing kits, scaffolding new kits, cross-kit validation, schema migration, and E2E test runs.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Maintain Oh My Game Kit repos as a kit maintainer. Use for releasing kits, scaffolding new kits, cross-kit validation, schema migration, and E2E test runs.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | omg-kit |
| description | Maintain Oh My Game Kit repos as a kit maintainer. Use for releasing kits, scaffolding new kits, cross-kit validation, schema migration, and E2E test runs. |
This skill was ported from upstream reference material. Interpret command names, paths, and agent-routing guidance as Codex/Oh My Game Kit equivalents. Prefer active Codex tools and local project instructions over Codex-specific mechanics when they conflict.
Unified kit maintenance skill. All subcommands are for kit maintainers only, not end-users.
| Subcommand | Usage | Purpose |
|---|---|---|
validate | omg-kit validate [--cross-kit] | Schema, activation paths, keyword conflicts, preset refs, universal omg- prefix (skills + agents — see rules/naming-convention.md) |
release | omg-kit release [--skip-validate] | Guided release: validate → CI gate → GitHub release → Discord |
sync | omg-kit sync [--pull] [--status-only] | Pull all kits, check version compat, release-action alignment |
scaffold | omg-kit scaffold <name> [--org] | Create new kit repo from template with full boilerplate |
audit | omg-kit audit [--all] | Health audit: file counts, modules, keywords, CI, freshness |
migrate | omg-kit migrate [--dry-run] | Schema migration (e.g., registryVersion 1→2) |
test | omg-kit test [--all] [--kit <name>] | Comprehensive 15-phase E2E testing: install, modules, CLI, release infra |
omg-kit validate # Validate current kit repo
omg-kit validate --cross-kit # Validate across all installed kits
omg-kit release # Guided release workflow
omg-kit sync --pull # Pull and check all kits
omg-kit scaffold oh-my-game-kit-x # Scaffold new kit
omg-kit audit --all # Audit all kit repos
omg-kit migrate --dry-run # Preview schema migration
omg-kit test --all # E2E test all kits
omg-kit test --kit unity # E2E test specific kit
Read the reference file for full details on each subcommand:
references/validate.md — per-kit and cross-kit validation checksreferences/release.md — release workflow steps and gatesreferences/sync.md — cross-kit synchronization protocolreferences/scaffold.md — new kit scaffolding templatereferences/audit.md — health audit checks and scoringreferences/migrate.md — registry schema migration steps (v1→v2)references/test.md — 12-phase E2E test planreferences/cli-commands.md — omg new, omg modules create, omg migrate, omg migrate metadata, omg rollback (cli@v4.14.0; flags + gotchas grounded in source)inject-origin-metadata.cjs runs post-merge — it strips and re-injects canonical origin / repository / module / protected (and version for SKILL.md) from env vars + module-resolution. So files already on main MUST NOT be hand-edited for these fields. BUT new files in PR branches MUST hand-stamp at least origin: (validator validate-origin-injection-coverage.cjs checks marker presence pre-merge — values are best-effort, CI overwrites them). Format per file type lives in memory feedback_origin_marker_handauthor.md and the omg-skill-creator "Required Frontmatter" section.release-modules.cjs (custom release script), NOT semantic-release directly. Flat kits (e.g., oh-my-game-kit-rn) still use semantic-release. release-modules.cjs produces per-module ZIPs + manifest.json index.omg-modules.json is GENERATED for the modules section — do not hand-author it. Source of truth is each module.json. Only presets in omg-modules.json are hand-authored..releaserc.json MUST include package.json and .agents/metadata.json in @semantic-release/git assets — without this, package.json stays at 0.0.0 forever. Every kit repo needs: "assets": ["package.json", "CHANGELOG.md", ".agents/metadata.json"]version or buildDate in source metadata.json — CI generates these from package.json. Source should only contain manually-maintained fields (name, repository, deletions)."full": "*" in omg-modules.json presets. The interactive selector renders a synthetic "Full — all N modules" choice from any "*"-valued preset (it keys off the value, not the name), so kits with no presets or a differently-named "*" preset look fine in the UI — but scripted invocations like omg new --preset full fail when the preset is named anything other than literally "full". Apr 2026 audit caught: web (everything: "*"), marketing + nakama (no presets section at all), designer (array-shorthand drift). Always name the "*" preset literally "full" even when the kit has only one required module (future-proofs scripted use). Other named presets (rpg, fullstack, etc.) MUST use v3 object form { "modules": [...], "description"?: "...", "crossKitModules"?: [...] } — array shorthand ["mod-a", "mod-b"] resolves at runtime but breaks cross-kit format consistency.module new and kit new are NOT real CLI subcommands. The shipped commands are omg new (project bootstrap) and omg modules create (module scaffold). Older audits and design docs use the legacy names — when documenting maintainer flows, reference the real CLI surface (references/cli-commands.md).omg rollback requires a snapshot that nothing currently creates. The H7 rollback read-side shipped in v4.14.0, but no production install/update path calls createSnapshot() yet. Until the write-side lands, omg rollback --kit X --to-snapshot pre-Y will reliably fail with snapshot '...' does not exist. Direct users to omg install --reset for now if they're truly stuck — and NEVER recommend rm -rf ~/.agents/. omg rollback --yes is a documented no-op as of v4.14.0 — declared in the CLI registry, never read by the handler. Do not promise confirmation-bypass behavior in maintainer guidance.--to-snapshot MUST literally start with pre-. Bare-version arguments (--to-snapshot 2.4.1) are rejected. Always render the pre- prefix in examples.detect: stub — when scaffold or omg-modules create writes a new module.json for a non-base module, emit a _disabled: true detect stub so P6d CI and doctor #41 stay green while the author tightens patterns:
"detect": {
"_disabled": true,
"_note": "AUTO-GENERATED STUB. Tighten files+anyOf then remove _disabled to activate.",
"files": ["**/*.md"],
"anyOf": [{ "pattern": "TightenThisPattern_<moduleName>", "minHits": 2 }]
}
The stub is schema-valid (passes P6d), scanner silently skips modules with _disabled: true, and doctor #41 surfaces the stub as "needs activation" so the author can't forget.omg- naming prefix (MANDATORY) — every skill directory + agent file + frontmatter name: MUST start with omg-. Tiers: core omg-{slug}, kit-wide omg-{kit}-{slug}, module-scoped omg-{kit}-{module}-{slug}. Slug MUST NOT redundantly start with the kit-short or any module-segment-token (algorithm v2 strips them). Three release-action gates are strict-by-default: validate-skill-prefix.cjs, validate-agent-prefix.cjs, validate-new-name-conformance.cjs. SSOT lives at rules/naming-convention.md + .agents/skills/omg-skill-creator/references/architecture-rules.md §0 + .agents/skills/omg-agent-creator/references/architecture-rules.md §0. When scaffold-ing a new kit/module or auditing an existing one, run node /path/to/release-action/scripts/validate-skill-prefix.cjs and validate-agent-prefix.cjs locally — fail-fast catches missed renames before push.omg-activation-*.json sessionBaseline[] and mappings[].skills[] arrays canonically use the bare slug form (e.g., nakama-rpc for the dir omg-nakama-rpc). The CI prefixer's auto-prefix-skills.cjs::buildSelfHealMap() self-heals legacy refs to canonical dir names at release time but the SSOT in fragment files stays bare so authors can write naturally. The release-action gate validate-activation-skill-resolution.cjs (per-PR safety net, added 2026-05-11) accepts BOTH bare and full-prefixed forms — anything else fails. When authoring or reviewing activation fragments, use the local bare slug ({slug} after kit + module strip) — NOT {kit}-{slug} shorthand that skips the module segment, which is the legacy form the self-heal does not accept and the gate flags. See omg-doctor check #47 for the doctor-time mirror.OMG vision: the kit should self-teach and improve using aggregated consumer data. Planned pipeline:
Status: Not yet implemented. See AGENTS.md "Self-Improving AI Vision" for context.
Triggers on: kit validate, kit release, kit sync, kit scaffold, kit audit, kit migrate, kit test, kit health, kit status, cross-kit, maintainer, test kit, test install, test modules, e2e kit, validate installation, self-improving, auto-gotcha, gotcha generation, ai aggregation, learn from users, omg new, omg rollback, omg migrate, omg modules create, module new, kit new, module migrate, migrate metadata, rollback snapshot, pre-snapshot
Unity engine adapter: .asmdef graph, Roslyn class diagrams with Unity stereotypes (MonoBehaviour, ScriptableObject, DOTS), UPM package deps, and C4 architecture views
Cocos Creator 3.8.7 playable-ad parameter discovery + implementation. Scans project config files (constant.ts, *Config.ts) AND all Canvas UI nodes via MCP, reads scene values for defaults, wires bindings, auto-assigns. Modes: --quick, --standard, --deep, --exhaustive, --auto.
Compare CK updates against GameKit baseline, suggest which changes to incorporate.
Answer technical questions with context-aware skill activation. Use for 'how does X work', 'what is the best way to', 'explain this pattern' queries.
Check context usage limits, monitor time remaining, optimize token consumption, debug context failures. Use when asking about context window, token budget, or agent context sizing.
Implement features end-to-end: plan, code, test, review via registry agents. Use for 'implement X', 'build Y feature', 'add Z functionality'. Handles full workflow.