[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-93392":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":16,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":17,"compositeScore":18,"rankGlobal":10,"rankLanguage":10,"license":19,"archived":20,"fork":20,"defaultBranch":21,"hasWiki":20,"hasPages":20,"topics":22,"createdAt":10,"pushedAt":10,"updatedAt":23,"readmeContent":24,"aiSummary":25,"trendingCount":15,"starSnapshotCount":15,"syncStatus":26,"lastSyncTime":27,"discoverSource":28},93392,"ClaudeBrain","Encod3d-Sec\u002FClaudeBrain","Encod3d-Sec","Karpathy LLM based claude harness for PenetrationTesting \u002F Bugbounty using obsidian","",null,"Python",200,33,1,0,39,4,67.99,"MIT License",false,"main",[],"2026-07-22 04:02:09","# ClaudeBrain\n\n```\n   ____ _                 _      ____            _\n  \u002F ___| | __ _ _   _  __| | ___| __ ) _ __ __ _(_)_ __\n | |   | |\u002F _` | | | |\u002F _` |\u002F _ \\  _ \\| '__\u002F _` | | '_ \\\n | |___| | (_| | |_| | (_| |  __\u002F |_) | | | (_| | | | | |\n  \\____|_|\\__,_|\\__,_|\\__,_|\\___|____\u002F|_|  \\__,_|_|_| |_|\n\nPentest & bug-bounty knowledge base + AI automation harness\n\n```\n\n**An AI-powered penetration testing and bug-bounty knowledge base and automation harness for [Claude Code](https:\u002F\u002Fclaude.com\u002Fclaude-code).** It turns an Obsidian vault into an opinionated offensive-security workflow: a searchable wiki of 500+ hacking technique pages, per-vulnerability \"hunt\" skills, deterministic hooks that fire the right skill at the right moment, and a state-first engagement model that stops you (and the model) from repeating work.\n\n[![Version](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fversion-1.0.3-blue.svg)](#)\n[![License: MIT](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-green.svg)](LICENSE)\n[![Built for Claude Code](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fbuilt%20for-Claude%20Code-blueviolet.svg)](https:\u002F\u002Fclaude.com\u002Fclaude-code)\n[![Python 3.10+](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-3.10%2B-blue.svg)](https:\u002F\u002Fwww.python.org\u002F)\n[![Wiki pages](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fwiki-500%2B%20pages-brightgreen.svg)](wiki\u002F)\n[![LinkedIn](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLinkedIn-Tomas%20Zabukas-0A66C2.svg?logo=linkedin&logoColor=white)](https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Ftomas-zabukas\u002F)\n\n\n\nClaudeBrain is a red-team \u002F bug-bounty second brain built on Andrej Karpathy's [LLM Wiki](https:\u002F\u002Fgist.github.com\u002Fkarpathy\u002F442a6bf555914893e9891c11519de94f) pattern: a persistent, AI-maintained knowledge base the model synthesizes each new source into over time, instead of re-deriving from raw documents on every query. Concretely, an offensive-security wiki, an agentic hunt-skill library, and a Model Context Protocol (MCP) search layer, wired together so Claude Code always checks the knowledge base before it attacks and never repeats a dead end. Think HackTricks or PayloadsAllTheThings, but indexed for semantic search and driven by an autonomous AI agent.\n\n> **Authorized testing only.** Everything here assumes a legal engagement: a signed penetration test, a bug-bounty program in scope, or your own lab \u002F CTF. You are responsible for staying in scope and within the rules of engagement.\n\nIf ClaudeBrain saves you time on an engagement, a [star](https:\u002F\u002Fgithub.com\u002FEncod3d-Sec\u002FClaudeBrain) helps other pentesters and bug-bounty hunters find it.\n\n---\n\n## Contents\n\n- [Features](#features)\n- [The knowledge base](#the-knowledge-base-500-pages-ship-with-the-repo)\n- [What ships vs what stays private](#what-ships-vs-what-stays-private)\n- [Requirements](#requirements)\n- [Quickstart](#quickstart)\n- [Plugins and MCP servers](#plugins-and-mcp-servers)\n- [Layout](#layout)\n- [The model underneath](#the-model-underneath)\n- [Safety and boundaries](#safety-and-boundaries)\n- [License](#license)\n\n---\n\n## Features\n\n- **Wiki-first methodology.** 500+ markdown technique pages indexed by `qmd` for semantic and keyword search over an MCP server (`wiki-search`). Every hunt skill queries the wiki *before* attacking, so knowledge compounds instead of scattering.\n- **Hunt skills** (`skills\u002Fhunt\u002Fhunt-*`): one per vulnerability class, XSS, SQLi, SSRF, IDOR, RCE, auth bypass, OAuth\u002FSAML federation, deserialization, cloud (AWS\u002FAzure\u002FGCP), Active Directory, API (OWASP API Top 10), LLM\u002FAI, request smuggling, cache poisoning, and more. Each is wiki-first, out-of-band-gated for blind bugs, and emits a uniform FIND finding schema.\n- **Deterministic automation (hooks).** Plain Python that fires on Claude Code lifecycle events:\n  - `hunt-trigger.py` (UserPromptSubmit) matches your prompt against `skills\u002Fhunt\u002Ftriggers.json` and loads the matching hunt skill.\n  - `recon-capture.py` (PostToolUse) fingerprints discovered tech against `scripts\u002Fplaybook.json`, routes to targeted tests, and auto-captures results into engagement state and PoC evidence.\n  - `engagement-init.py` (SessionStart) self-heals the engagement file set and injects a ranked next-move summary.\n  - `scope-guard.py` (PreToolUse) warns when a command targets an out-of-scope host or uses rules-of-engagement-forbidden tooling.\n- **State-first engagement model.** Each engagement lives under `targets\u002F\u003Cname>\u002F` (`state`, `loot`, `paths`, `scope`, `coverage`). Analyzers (`next_move.py`, `coverage.py`) rank what to do next and surface untested vulnerability classes so nothing in scope is skipped.\n- **Research loop** (`skills\u002Fresearch`) for CVE discovery on binaries, libraries, and repos, with its own persistent state under `raw\u002Fresearch\u002F`.\n- **Hard client-data boundary.** Every client specific stays under `targets\u002F` (git-ignored); `scripts\u002Fcheck-leaks.sh` gates tracked files before you ever push.\n\n---\n\n## The knowledge base (500+ pages ship with the repo)\n\nThe `wiki\u002F` corpus is the heart of ClaudeBrain and it is fully committed, clone it and you get the whole library, not an empty shell. It is a living offensive-security reference organized as:\n\n| Area | Covers |\n|---|---|\n| `wiki\u002Ftechniques\u002F` | Active Directory, cloud, web, network, Linux, macOS, exploit-dev, OSINT, cracking, red-team, mobile\u002FIoT, blockchain, methodology |\n| `wiki\u002Fpayloads\u002F` | Per-vulnerability-class payload arsenals the hunt skills pull from |\n| `wiki\u002Ftools\u002F` | Per-tool references (nmap, ffuf, nuclei, httpx, sqlmap, BloodHound, netexec, ...) |\n| `wiki\u002Fcheatsheets\u002F` | Quick-reference command sheets and default-credential tables |\n\nPages are cross-linked Obsidian-style and indexed for semantic search, so \"SSRF to cloud metadata\" or \"ESC1 ADCS\" resolves to the right page in one query. The wiki grows every engagement through the `learn` skill, which distills generic, client-free lessons back into it.\n\n---\n\n## What ships vs what stays private\n\nThe wiki and the entire harness are public. Only client data and per-machine state are held back by [`.gitignore`](.gitignore):\n\n| Tracked (ships, safe to push) | Git-ignored (stays private) |\n|---|---|\n| `wiki\u002F` the full 500+ page corpus | `targets\u002F` client engagements and findings |\n| `skills\u002F`, `scripts\u002F`, `setup\u002F`, `docs\u002F`, `tests\u002F` | `CLAUDE.local.md` machine hostnames and paths |\n| `CLAUDE.md`, `README.md`, `LICENSE` | `session\u002F`, `raw\u002F` local working state |\n| `targets\u002FTARGETS.md` the generic engagement playbook | `.claude\u002F`, `.obsidian\u002F`, runtime stamps and caches |\n\n**The client-data boundary is a hard rule.** Hosts, credentials, findings, and engagement narrative live *only* under `targets\u002F`, which git ignores wholesale (with the single generic playbook whitelisted). Run `bash scripts\u002Fcheck-leaks.sh` before your first push, it scans tracked files for engagement markers, private IPs, and emails.\n\n---\n\n## Requirements\n\n- Linux or WSL, `bash`, `python3` (3.10+)\n- [Claude Code](https:\u002F\u002Fclaude.com\u002Fclaude-code) CLI\n- Node.js >= 18 and [bun](https:\u002F\u002Fbun.sh)\n- `qmd` (installed by the bootstrap via `bun install -g @qmd\u002Fcli`), which provides the `wiki-search` MCP server via `qmd mcp`\n\n## Quickstart\n\n```bash\ngit clone \u003Cthis-repo> ClaudeBrain\ncd ClaudeBrain\n\n# One-time, per-machine setup. NOTE: bootstrap.sh mutates ~\u002F.claude: it writes a\n# CLAUDE.md include, symlinks the hooks, registers the MCP servers, and installs\n# qmd + the official Claude plugins. Read it first; it is aggressive by design.\nbash setup\u002Fbootstrap.sh\n\n# Build the search index (re-run after adding wiki pages)\nqmd update\n\n# Restart Claude Code so it loads the hooks, skills, and MCP servers.\n\n# Start an engagement (pentest | bugbounty | ctf)\nbash setup\u002Fnew-engagement.sh acme pentest\n\n# Run the test suite\npython3 -m pytest -q\n\n# Before any push, run the leak gate\nbash scripts\u002Fcheck-leaks.sh\n```\n\n`bootstrap.sh` self-locates the vault; if it cannot, set `OBSIDIAN_VAULT` (and `QMD_VAULT` for the search index) to the repo root. Per-machine paths go in the git-ignored `CLAUDE.local.md`, copy [`CLAUDE.local.example.md`](CLAUDE.local.example.md) to create it. See [`docs\u002Fsetup.md`](docs\u002Fsetup.md) for the full walkthrough.\n\n---\n\n## Plugins and MCP servers\n\nClaudeBrain runs on Claude Code plus a set of plugins and MCP servers. `setup\u002Fbootstrap.sh`\ninstalls the core set for you; the rest are referenced by the workflow and installed separately.\n\n**Installed by `bootstrap.sh`:**\n\n- **qmd** - the semantic + keyword search engine over `wiki\u002F` (`bun install -g @qmd\u002Fcli`); registers the `wiki-search` MCP (`qmd mcp`).\n- **caveman** - terse-output mode; also registers the `caveman-shrink` MCP (a token-compressed `wiki-search`).\n- Official Claude Code plugins: **code-review**, **claude-md-management**, **skill-creator**, **frontend-design** (`claude plugins install \u003Cname>@claude-plugins-official`).\n\n**Install separately (referenced by the harness):**\n\n- **superpowers** (recommended) - the planning \u002F execution \u002F debugging workflow the CLAUDE.md loop routes to: `brainstorming` -> `writing-plans` -> `subagent-driven-development`, plus `systematic-debugging`, `dispatching-parallel-agents`, and `verification-before-completion`.\n- **ponytail** (optional) - \"lazy senior dev\" mode that pushes for the simplest working solution.\n- **context7** (optional MCP) - up-to-date library \u002F API docs, used for vendor-default and dependency lookups.\n- **gsd** (optional) - the `pause-work` session-end helper.\n- **burp-mcp** (optional MCP) - drives Burp Suite for the `hunt-burp` workflow.\n\nInstall plugins from Claude Code's plugin marketplace (`\u002Fplugin`), and MCP servers with `claude mcp add`.\nEverything degrades gracefully: the hooks fail open and the hunt\u002Fwiki skills work without the optional\nplugins, but the documented planning loop assumes `superpowers`.\n\n## Layout\n\n```\nCLAUDE.md         top-level instructions loaded by Claude Code\nwiki\u002F             500+ page technique corpus (ships; semantic + keyword indexed)\nskills\u002F           hunt-* skills, research \u002F disclosure, hooks\u002F, meta-skills\nscripts\u002F          next_move, coverage, find-lint, check-leaks, index \u002F lint tooling\nsetup\u002F            bootstrap, install-hooks, install-skills, new-engagement \u002F research, templates\ndocs\u002F             workflows, page-types, setup, sharing (client-data boundary), conventions\ntests\u002F            pytest suite for the automation\ntargets\u002F          engagements (git-ignored; client data lives ONLY here)\n```\n\nStart with [`docs\u002Fworkflows.md`](docs\u002Fworkflows.md) for the day-to-day flow and [`docs\u002Fsharing.md`](docs\u002Fsharing.md) for the client-data boundary rules.\n\n---\n\n## The model underneath\n\nClaudeBrain is a *harness*, the intelligence it orchestrates is a large language model (Claude). If you want to understand what an LLM actually is under the hood, tokens, attention, training, and inference, the clearest from-scratch implementations on the internet are Andrej Karpathy's:\n\n- **[nanoGPT](https:\u002F\u002Fgithub.com\u002Fkarpathy\u002FnanoGPT)** a minimal, readable GPT training and finetuning codebase; the canonical \"here is a transformer, end to end.\"\n- **[nanochat](https:\u002F\u002Fgithub.com\u002Fkarpathy\u002Fnanochat)** a full ChatGPT-style pipeline (pretraining, supervised finetuning, RL, and inference with a web UI) in one hackable repo.\n- **[Neural Networks: Zero to Hero](https:\u002F\u002Fkarpathy.ai\u002Fzero-to-hero.html)** the video series that builds an LLM line by line, from `micrograd` up to a GPT.\n\nThose repos show you the engine; ClaudeBrain shows you how to point that engine at a target and keep it disciplined.\n\n---\n\n## Safety and boundaries\n\n- Client and engagement data (hosts, credentials, findings) lives **only** under `targets\u002F`, which is git-ignored. Never write it into `wiki\u002F`, `docs\u002F`, `skills\u002F`, or commit messages.\n- `scripts\u002Fcheck-leaks.sh` scans tracked files for engagement markers and flags emails and private IPs before you publish. Run it before any push.\n- The hunt skills enforce out-of-band confirmation for blind vulnerability classes, no inference-only findings.\n\n---\n\n## License\n\nMIT, see [LICENSE](LICENSE).\n","ClaudeBrain 是一个面向渗透测试与漏洞赏金猎人的 AI 自动化知识库与工作流引擎，基于 Claude Code 大模型构建。它将 Obsidian 笔记库转化为结构化的红队知识中枢，内置 500+ 技术页面、按漏洞类型组织的可触发‘狩猎技能’、确定性上下文钩子及状态感知的交互模型，支持语义检索与防重复决策。项目采用 Karpathy 提出的 LLM Wiki 持久化知识合成范式，强调知识沉淀而非单次查询重算。适用于授权渗透测试、合规漏洞挖掘及 CTF 实战等需知识复用与流程自动化的安全场景。",2,"2026-07-17 02:30:09","CREATED_QUERY"]