launch-prep — community launch-prep, benchflow, community, ide skills, Claude Code, Cursor, Windsurf

v1.0.0

关于此技能

Framework for creating high fidelity and complex RL environments and evaluation tasks

benchflow-ai benchflow-ai
[203]
[16]
更新于: 4/16/2026

Killer-Skills Review

Decision support comes first. Repository text comes second.

Reference-Only Page Review Score: 1/11

This page remains useful for operators, but Killer-Skills treats it as reference material instead of a primary organic landing page.

Review Score
1/11
Quality Score
46
Canonical Locale
en
Detected Body Locale
en

Framework for creating high fidelity and complex RL environments and evaluation tasks

核心价值

Framework for creating high fidelity and complex RL environments and evaluation tasks

适用 Agent 类型

Suitable for operator workflows that need explicit guardrails before installation and execution.

赋予的主要能力 · launch-prep

! 使用限制与门槛

Why this page is reference-only

  • - Current locale does not satisfy the locale-governance contract.
  • - The page lacks a strong recommendation layer.
  • - The page lacks concrete use-case guidance.
  • - The page lacks explicit limitations or caution signals.
  • - The underlying skill quality score is below the review floor.

Source Boundary

The section below is supporting source material from the upstream repository. Use the Killer-Skills review above as the primary decision layer.

实验室 Demo

Browser Sandbox Environment

⚡️ Ready to unleash?

Experience this Agent in a zero-setup browser environment powered by WebContainers. No installation required.

Boot Container Sandbox

常见问题与安装步骤

以下问题与步骤与页面结构化数据保持一致,便于搜索引擎理解页面内容。

? FAQ

launch-prep 是什么?

Framework for creating high fidelity and complex RL environments and evaluation tasks

如何安装 launch-prep?

运行命令:npx killer-skills add benchflow-ai/benchflow/launch-prep。支持 Cursor、Windsurf、VS Code、Claude Code 等 19+ IDE/Agent。

launch-prep 支持哪些 IDE 或 Agent?

该技能兼容 Cursor, Windsurf, VS Code, Trae, Claude Code, OpenClaw, Aider, Codex, OpenCode, Goose, Cline, Roo Code, Kiro, Augment Code, Continue, GitHub Copilot, Sourcegraph Cody, and Amazon Q Developer。可使用 Killer-Skills CLI 一条命令通用安装。

安装步骤

  1. 1. 打开终端

    在你的项目目录中打开终端或命令行。

  2. 2. 执行安装命令

    运行:npx killer-skills add benchflow-ai/benchflow/launch-prep。CLI 会自动识别 IDE 或 AI Agent 并完成配置。

  3. 3. 开始使用技能

    launch-prep 已启用,可立即在当前项目中调用。

! 参考页模式

此页面仍可作为安装与查阅参考,但 Killer-Skills 不再把它视为主要可索引落地页。请优先阅读上方评审结论,再决定是否继续查看上游仓库说明。

Imported Repository Instructions

The section below is supporting source material from the upstream repository. Use the Killer-Skills review above as the primary decision layer.

Supporting Evidence

launch-prep

安装 launch-prep,这是一款面向AI agent workflows and automation的 AI Agent Skill。支持 Claude Code、Cursor、Windsurf,一键安装。

SKILL.md
Readonly
Imported Repository Instructions
The section below is supporting source material from the upstream repository. Use the Killer-Skills review above as the primary decision layer.
Supporting Evidence

BenchFlow Launch Prep

Prepares a release: docs/labs/code review → CI gate → e2e smoke → CHANGELOG → version bump → PR.

Arguments passed: $ARGUMENTS


Dispatch

No args — show release state: current version (grep '^version' pyproject.toml), [Unreleased] section of CHANGELOG.md, branch and dirty files (git status --short). Recommend next step.

check — run CI gate, report pass/fail, no edits:

bash
1.venv/bin/ruff format --check src tests && .venv/bin/ruff check src tests 2.venv/bin/ty check 3.venv/bin/python -m pytest tests/ -q 4uv lock --check

patch / minor / major — follow Steps 0–6 below.


Step 0 — Pre-flight

  1. Version: grep '^version' pyproject.toml, compute new version, ask user to confirm before proceeding. Use their value if they override.
  2. Branch: git branch --show-current — if main, stop: "Switch to a feature branch or: git checkout -b release/v<NEW_VERSION>". Any non-main branch is fine.

Step 1 — Docs, labs, and code review (3 subagents in parallel)

A — user-facing docs: Read README.md and all docs/*.md. Evaluate as a first-time external user: complete install instructions, working examples, accurate CLI flags and API signatures, no broken relative links. Flag anything confusing or stale.

B — labs/: For each experiment: README matches code, no broken links, no stale src/benchflow API calls, no notebook tracebacks. Grep for TODO, FIXME.

C — src/ code smell: Grep src/ and tests/ for TODO, FIXME, HACK, XXX (blocker vs. debt). Grep for hardcoded secrets (api_key\s*=\s*["'], etc.). Check __init__.py for debug symbols.

Synthesize: Blockers / Polish / Clean. Stop on blockers.


Step 2 — CI gate

Mirrors .github/workflows/test.yml exactly. Stop on any failure.

bash
1.venv/bin/ruff format src tests # note: ty check takes no path arg 2.venv/bin/ruff check src tests 3.venv/bin/ty check 4.venv/bin/python -m pytest tests/ -q 5uv lock --check

If ruff format changed files: git diff --name-only, then git add <those files only> — not git add ..


Step 3 — e2e smoke test

bash
1source .env 2>/dev/null || true 2.venv/bin/python -m pytest -m live tests/test_smoke.py -v

If Docker is unavailable, warn and ask to skip or abort — do not skip silently. Expected: test_hello_world_smoke passes with reward > 0 and non-empty trajectory.


Step 4 — CHANGELOG draft

If [Unreleased] is non-empty: show it, ask to confirm.

If empty: generate from git log —

bash
1git log $(git describe --tags --abbrev=0)..HEAD --oneline --no-merges

Map prefixes: feat: → Added, fix: → Fixed, chore:/refactor:/build: → Changed. Omit docs: and test: unless notable. Show draft and wait for user approval before writing anything.


Step 5 — Write CHANGELOG, pyproject.toml, and uv.lock

CHANGELOG: Insert approved entries as ## <NEW_VERSION> — $(date +%Y-%m-%d) immediately after ## [Unreleased], leaving [Unreleased] empty. Use Edit, not Write.

pyproject.toml + uv.lock: Edit only version = "..." under [project] (line 3). Do not touch target-version (ruff) or python-version (ty) — those are Python version pins. Then run uv lock so the editable benchflow package entry in uv.lock matches the new version. Do not edit __init__.py; it reads version from importlib.metadata automatically.


Step 6 — Commit and PR

bash
1git add CHANGELOG.md pyproject.toml uv.lock # plus any ruff-formatted files from Step 2 2git commit -m "chore: release v<NEW_VERSION>" 3git push -u origin HEAD 4gh pr create --title "chore: release v<NEW_VERSION>" --body "$(cat <<'EOF' 5## Release v<NEW_VERSION> 6 7See CHANGELOG.md for details. 8 9## Checklist 10- [ ] CI gate passes 11- [ ] e2e smoke test passes 12- [ ] CHANGELOG updated 13- [ ] Version bumped in pyproject.toml 14- [ ] uv.lock refreshed 15- [ ] Merge and tag after review 16 17🤖 Generated with [Claude Code](https://claude.com/claude-code) 18EOF 19)"

Return the PR URL.


Step 7 — Post-merge (remind user, do not run)

bash
1git checkout main && git pull 2git tag v<NEW_VERSION> && git push origin v<NEW_VERSION> 3 4# GitHub release — substitute the actual version for <NEW_VERSION> in the awk pattern 5awk '/^## <NEW_VERSION> /{flag=1; next} /^## [0-9]/{flag=0} flag' CHANGELOG.md > /tmp/release_notes.md 6gh release create v<NEW_VERSION> --title "v<NEW_VERSION>" --notes-file /tmp/release_notes.md 7 8# PyPI publish — separate step (requires UV_PUBLISH_TOKEN or ~/.pypirc) 9uv build && uv publish

相关技能

寻找 launch-prep 的替代方案 (Alternative) 或可搭配使用的同类 community Skill?探索以下相关开源技能。

查看全部

openclaw-release-maintainer

Logo of openclaw
openclaw

Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞

333.8k
0
AI

widget-generator

Logo of f
f

为prompts.chat的信息反馈系统生成可定制的插件小部件

149.6k
0
AI

flags

Logo of vercel
vercel

React 框架

138.4k
0
浏览器

pr-review

Logo of pytorch
pytorch

Python中具有强大GPU加速的张量和动态神经网络

98.6k
0
开发者工具