Agent Skills
› udecode/plate
› slate-ar-recipe
slate-ar-recipe
GitHub用于选择 Codex Autoresearch 配方，生成只读设置计划并映射非性能循环指标。在用户询问运行何种循环或任务测量不明确时触发，提供配方推荐、列表及详细配置。
Trigger Scenarios
询问应运行何种 Autoresearch 循环
需要分析测试运行时、类型检查、包体积等指标
Install
npx skills add udecode/plate --skill slate-ar-recipe -g -y
SKILL.md
Frontmatter
{
"name": "slate-ar-recipe",
"metadata": {
"skiller": {
"source": ".agents\/rules\/slate-ar-recipe.mdc"
}
},
"description": "Slate v2 Autoresearch recipe picker. Lists\/recommends Codex Autoresearch recipes, produces read-only setup plans, and maps non-perf loops such as test runtime, typecheck, bundle size, memory, command latency, and quality-gap.",
"argument-hint": "[recommend | list | show <recipe> | setup <recipe> | prompt]",
"disable-model-invocation": true
}
Slate AR Recipe
Handle $ARGUMENTS by loading slate-ar and using Codex Autoresearch recipes.
Use this when the user asks what Autoresearch loop to run, or when a task is measured but not obviously performance-specific.
Contract
- Default to read-only
recipes recommend,recipes list,recipes show, andsetup-plan. - Target cwd is
.tmp/slate-v2. - Do not start packets unless the user explicitly asks to run the loop.
- Do not trust external recipe catalogs without inspecting them first.
- In Slate v2, prefer explicit Bun/package commands over a generic npm recipe
when the recommender picks
node-test-runtime.
Built-In Recipe Uses
node-test-runtime/vitest-runtime: test gate repeatability. Often route toslate-ar-gate.typescript-compile-time: typecheck/runtime graph pressure.bundle-size: package or site bundle footprint.memory-usage: workload memory smoke, after replacing the placeholder with the real workload.command-latency: custom command duration, after replacing the placeholder.quality-gap: broad research checklist. Usually route toslate-ar-quality.custom: last resort when no built-in recipe fits.
Commands
Use direct CLI until a root shortcut exists:
node ../codex-autoresearch/plugins/codex-autoresearch/scripts/autoresearch.mjs recipes list --cwd .tmp/slate-v2
node ../codex-autoresearch/plugins/codex-autoresearch/scripts/autoresearch.mjs recipes recommend --cwd .tmp/slate-v2
node ../codex-autoresearch/plugins/codex-autoresearch/scripts/autoresearch.mjs recipes show <recipe-id> --cwd .tmp/slate-v2
node ../codex-autoresearch/plugins/codex-autoresearch/scripts/autoresearch.mjs setup-plan --cwd .tmp/slate-v2 --recipe <recipe-id>
node ../codex-autoresearch/plugins/codex-autoresearch/scripts/autoresearch.mjs doctor --cwd .tmp/slate-v2 --check-benchmark --explain
If the recipe output is customized, run benchmark-lint before any packet.
Handoff
Report the selected recipe, metric, direction, benchmark command, checks
command, caveats, and recommended owner: slate-ar, slate-ar-quality,
slate-ar-gate, slate-ar-perf, slate-patch, or slate-plan.
Version History
- af5e430 Current 2026-08-20 19:28
