脑力爆发
脑力爆发 是 AI Skill Hub 本期精选Agent工作流之一。综合评分 7.5 分，整体质量较高。我们推荐使用将其纳入你的 AI 工具库，帮助提升工作效率。
📚 深度解析
脑力爆发 工作流的设计遵循"最小配置，最大复用"原则：核心逻辑已经封装好，用户只需配置自己的 API Key 和业务参数即可快速上手。工作流内置错误处理和重试机制，在网络波动或 API 限速等情况下仍能稳定运行，适合作为生产环境的自动化基础设施。
在实际部署时，建议先在测试环境中运行 3-5 次，验证各个环节的输出结果符合预期，再部署到生产环境。AI Skill Hub 评分 7.5 分，是同类 Agent 工作流中的精选推荐。
📋 工具概览
脑力爆发 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排，将复杂的多步骤任务拆解为清晰的自动化流程，实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成，适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
📖 中文文档
脑力爆发 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排，将复杂的多步骤任务拆解为清晰的自动化流程，实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成，适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
- 可视化 Agent 工作流编排，无需编写复杂代码
- 支持多步骤自动化任务链，实现全流程无人值守
- 与外部 API、数据库和第三方服务无缝集成
- 内置错误处理与自动重试机制，保障稳定运行
- 提供可复用的自动化模板，快速在同类场景部署
- 自动化日常重复性工作，将精力集中于创造性任务
- 构建数据采集 → 处理 → 输出的完整自动化管线
- 实现跨平台、跨系统的数据流转和业务协同
# 方式一：npm 全局安装 npm install -g brainblast # 方式二：npx 直接运行（无需安装） npx brainblast --help # 方式三：项目依赖安装 npm install brainblast # 方式四：从源码运行 git clone https://github.com/DSB-117/brainblast cd brainblast npm install npm start
- 访问 GitHub 仓库获取工作流文件
- 在对应平台（Dify / Flowise / Make 等）中找到「导入工作流」功能
- 上传工作流文件
- 按照提示配置必要的环境变量和 API Key
- 运行测试确认流程正常后投入使用
# 命令行使用
brainblast --help
# 基本用法
brainblast [options] <input>
# Node.js 代码中使用
const brainblast = require('brainblast');
const result = await brainblast.run(options);
console.log(result);
# brainblast 配置说明 # 查看配置选项 brainblast --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量（覆盖配置文件） export BRAINBLAST_CONFIG="/path/to/config.yml"
brainblast
Research external APIs and SDKs before your AI agent starts coding — then enforce, in CI, that what got written matches.
---
AI coding agents start implementing before they actually know the systems they are integrating. They know the name of an SDK but not the version. They know an API exists but not that a required config step is mandatory, or that a setting is immutable after deploy, or that a fee recipient defaults to zero if omitted.
Brainblast runs first. It reads your requirements, identifies every external component, browses official docs and package registries, and produces a structured research report — with facts, risks, and answered questions — before any code is written.
The report travels with the project. Any coding agent can use it without repeating the research.
Two entry points, one product. Brainblast predicts the failure before any code exists — the /brainblast research skill, run inside your coding agent — then enforces that the fix stays shipped, forever — the brainblast npm CLI, run in CI. Same traps (Stripe, Privy/JWT, Bags/Solana fee-share, …), same report.json contract, two moments in the lifecycle:
```sh
Capabilities
Everything Brainblast does today, at a glance.
Research workflow - Auto-detects the requirements file from common spec names (requirements, prd, spec, brief, rfc, and more — any case, .md/.txt/.rst), or takes an explicit path. - Builds a component inventory — every external API, SDK, auth provider, database, payment processor, cloud platform, or chain in the spec, each tagged by how confident the identification is (named / implied / inferred). - Plans a source set per component — official docs, package registry, changelog, rate limits — before browsing. - Browses live sources, never recalls. Every fact comes from a URL fetched during the run, so it reflects the current docs, not stale training data. - Runs a questions loop — every open question that surfaces is answered from a live URL, or explicitly marked unresolvable with a note on where it looked. - Reviews its own coverage and flags gaps before finishing. - Re-reads the requirements against the research to surface wrong assumptions, missing constraints, underspecified choices, and decisions that are immutable after deploy. - Caches research per component, keyed by name@version. Re-runs are incremental — unchanged components are reused from .agent-research/cache/ and only new or version-bumped components are re-researched; --fresh forces a full re-research.
Per-component output - Each component file is structured identically: Facts (each with a source URL), Assumptions, Inferences, Risks (rated CRITICAL / HIGH / MEDIUM / LOW, biased toward silent failures), and Resolved questions.
Handoff report - A single final-report.md opening with an Executive Summary (what's being built, a Ready / Caution / Blocked verdict, the top risk, the one irreversible decision, the biggest spec gap) and a Risk Heatmap (component × severity counts, with CRITICAL/HIGH risks named). - Followed by the components table, what a coding agent must know before starting, required pre-coding decisions, requirements corrections, and the specific failure modes the run prevents. - Auto-injects a pointer to the report into the project's agent-instructions file (CLAUDE.md, or AGENTS.md on Codex) as an idempotent, marker-delimited, reversible block — so the research travels to the next coding session with no copy-paste. - Emits a machine-readable report.json alongside the prose — a stable, versioned (schemaVersion: "1.0") schema with components, severity-tagged risks, pre-coding decisions, and requirements corrections, so other tools and CI gates can build on a contract instead of parsing prose. - Gates CI. A --ci mode runs non-interactively (no prompts, documented defaults), and a dependency-free gate script turns report.json into an exit code — fail the build if any CRITICAL risk remains (--fail-on=critical|high|…) or the verdict is blocked.
Deterministic auditor — npx brainblast - Published to npm as brainblast@0.2.0 with SLSA provenance attestation — npx brainblast . runs it with no install, and you can verify the build came from this repo's CI, not a laptop. - A Node/TypeScript static auditor in packages/core that scans code offline (no network, no LLM) for the first built-in integration traps: Stripe webhook raw-body signature verification, Privy/JWT signature + aud + iss verification, and the Bags/Solana fee-share creator-inclusion trap (the same zero-revenue misconfiguration the research example below caught). - Emits CI-readable checks[] and checkTotals into report.json, and can generate behavioral contract tests that fail on the vulnerable fixtures and pass on the fixed ones — the durable guardrail that keeps a fixed trap fixed. - Loads project-local .agent-research/rules/*.yaml rules as data, without executing scanned code or allowing project rules to shadow bundled rules.
Safety - Prompt-injection resistant by design. Browsed docs are treated as untrusted data; imperative content ("ignore previous instructions", "run this") is quoted and flagged, never propagated as fact or action. - Reaches gated docs when needed via gstack's cookie import.
Platforms & install - Runs on Claude Code, OpenClaw, Codex (native skill + adapter block), and any agent with web access via a generic prompt. - Exposes /brainblast and /brainblast-update slash commands. - Secure installer: pins to a tagged release and verifies the SHA-256 of every file before writing it, checks the gstack dependency, and re-installs idempotently (BRAINBLAST_REF=latest or a specific version). - Ships two complete committed example runs and a release self-check (scripts/validate.sh).
Prerequisites
Brainblast is a workflow that runs inside a host agent. It needs a browser engine to fetch live docs.
| Host agent | What you need |
|---|---|
| **Claude Code / OpenClaw** | [gstack](https://github.com/garrytan/gstack) (provides the browse engine), plus its requirements: Git, [Bun](https://bun.sh) v1.0+, and Node.js on Windows |
| **Codex** | Built-in web access — no extra dependency |
| **Generic agent** | Any agent with web browsing |
Install gstack first if you are on Claude Code or OpenClaw. Paste this into Claude Code and it does the rest:
Install gstack: run git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setupEnforce — statically scan the code that got written, gate the build on what it finds
npx brainblast . ```
See it for real: examples/bags-api/ is a complete committed run against the Bags API (Solana token launch), including the final report. It caught a permanent, silent, zero-revenue misconfiguration an agent would have shipped.
Install
curl -fsSL https://raw.githubusercontent.com/DSB-117/brainblast/v0.2.0/install.sh | shThe installer pins to a tagged release, verifies SHA-256 checksums before writing any file, and auto-detects Claude Code, OpenClaw, and Codex. If gstack is missing, it warns you with the exact command to fix it. (It installs the Brainblast skill, but it does not install gstack for you — that is a one-time prerequisite above.)
Or tell your agent:
Install Brainblast by running: curl -fsSL https://raw.githubusercontent.com/DSB-117/brainblast/v0.2.0/install.sh | sh
For the bleeding edge instead of a pinned release, prefix with BRAINBLAST_REF=main.
Uninstall
```sh
Usage
Write a requirements file, then run:
/brainblast requirements.mdOr just /brainblast — Brainblast auto-detects common spec filenames (requirements.md, prd.md, spec.md, brief.md, rfc.md, etc., case-insensitive, .md/.txt/.rst). If it finds exactly one match it uses it silently; if it finds several it asks you to pick.
Re-runs are incremental: Brainblast caches each component's research keyed by name@version and only re-researches what changed — a new component, or a version bump. Pass /brainblast --fresh (or set BRAINBLAST_FRESH=1) to ignore the cache and re-research everything.
Brainblast will:
- Read the requirements and list every external component
- Build a source plan (docs, registry, changelog, rate limits) for each
- Browse each source and extract facts, assumptions, inferences, and risks
- Answer every open question it encounters — no question is left unresolved if a URL can answer it
- Review coverage and flag gaps
- Re-read the requirements and flag wrong assumptions, missing constraints, and immutable decisions
- Produce a final handoff report for the coding agent
Continuous integration
Brainblast can gate a pipeline on its own findings — block a merge until a human has dealt with every CRITICAL risk. Two pieces:
--cimode runs Brainblast non-interactively: it never asks a question, picks documented defaults (e.g. a deterministic requirements-file precedence), and writesreport.json. Invoke your agent headless —/brainblast requirements.md --ci, or setBRAINBLAST_CI=1.- The gate —
scripts/brainblast-gate.sh— turns thatreport.jsoninto an exit code:
```sh
Troubleshooting
BROWSE_MISSING when running /brainblast gstack's browse engine is not installed. Run the gstack install command from Prerequisites, then retry.
"Multiple requirements files found" Brainblast found more than one candidate (e.g. both prd.md and spec.md) and will ask which to use. Pass one explicitly: /brainblast prd.md.
"No requirements file found" No file with a recognised spec name exists. Brainblast will show any .md files in the project root and ask which to use. If there are none, create a file describing what you are building and pass it explicitly.
Checksum mismatch during install The installer refuses to write files whose SHA-256 does not match the tagged release. If you see this, you may be behind a proxy that rewrites content, or the ref is mistyped. Verify the ref and retry.
高质量的AI工作流项目
- 需要 brainblast 解决具体问题的开发者与运营人员
- 先在测试环境跑通最小用例，再接入生产数据
- API key 直接提交到 git 仓库（请用 .env 并加入 .gitignore）
- 云端托管：可放在 Vercel / Railway / Fly.io 等 PaaS 平台
⚡ 核心功能
- 可视化 Agent 工作流编排，无需编写复杂代码
- 支持多步骤自动化任务链，实现全流程无人值守
- 与外部 API、数据库和第三方服务无缝集成
- 内置错误处理与自动重试机制，保障稳定运行
- 提供可复用的自动化模板，快速在同类场景部署
- 需要 brainblast 解决具体问题的开发者与运营人员
- 先在测试环境跑通最小用例，再接入生产数据
- API key 直接提交到 git 仓库（请用 .env 并加入 .gitignore）
👥 适合人群
🎯 使用场景
- 自动化日常重复性工作，将精力集中于创造性任务
- 构建数据采集 → 处理 → 输出的完整自动化管线
- 实现跨平台、跨系统的数据流转和业务协同
⚖️ 优点与不足
- +MIT 协议，可免费商用
- +大幅减少重复性人工操作
- +可视化流程，清晰直观
- +可扩展性强，支持复杂场景
- −初始配置和调试需投入一定时间
- −强依赖外部服务的稳定性
- −复杂场景需具备一定技术基础
AI Skill Hub 为第三方内容聚合平台，本页面信息基于公开数据整理，不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后，再部署至生产环境，并做好必要的安全评估。
✅ MIT 协议 — 最宽松的开源协议之一，可自由商用、修改、分发，仅需保留版权声明。
🔗 相关工具推荐
❓ 常见问题 FAQ
经综合评估，脑力爆发 在Agent工作流赛道中表现稳健，质量良好。如果你已有明确的使用需求，可以直接上手体验；如果还在评估阶段，建议对比同类工具后再做决策。
| 原始名称 | brainblast |
| 原始描述 | 开源AI工作流：Predict the silent integration traps an AI agent would ship (zero-revenue config。⭐6 · TypeScript |
| Topics | aicijwtsecuritysolanatypescript |
| GitHub | https://github.com/DSB-117/brainblast |
| License | MIT |
| 语言 | TypeScript |
收录时间：2026-06-11 · 更新时间：2026-06-11 · License：MIT · AI Skill Hub 不对第三方内容的准确性作法律背书。