LLM Mart Basic

Use when a PR has review feedback and the user wants it addressed — pull the review comments, triage each one, apply the changes it warrants, then commit, push, and post a humanized reply per comment. Closes the loop between a review and the follow-up commit; it does not re-revie

Use when the user wants to record an architectural decision — drafting an Architecture Decision Record (ADR) or RFC, documenting a design choice and its trade-offs, or capturing why an approach was taken. Detects the repo's existing ADR location and template style (MADR or Nygard

Use when the user reports an error, bug, or unexpected behavior in this repo and wants help diagnosing it. Five phases — reproduce, diagnose root cause (read-only), write a failing test, fix, verify against the full suite. Also known as `klaussy-debug`.

Use when the user wants to upgrade the project's dependencies safely — bump versions, read changelogs for breaking changes, and verify the suite still passes. Upgrades incrementally and stops on the first break; it does not add new dependencies (that's a design decision to raise

Use when the user wants documentation written or updated — docstrings, API docs, a README section, or a doc comment on a tricky piece of code. Documents selectively: what a reader genuinely can't infer from the code, and nothing they can. Writes prose, not code changes. Also know

Use when the user wants lint, format, and type errors fixed in the current changes. Reads CLAUDE.md for the repo's lint/format/type-check commands, runs each, and fixes only style/format/type issues — no behavior changes. Also known as `klaussy-fix`.

Use when the user is tired of approving the same routine dev work ("stop asking me yes", "allow the normal dev tools", "grant permissions"). Detects the repo's stack and writes a curated allow-list into the agent's own local permission file so the basics stop prompting — reading,

Use whenever prose, a comment, a doc, a PR or commit body, or a file's text should read like a human engineer wrote it instead of an AI — "humanize this", "make it sound less like a bot", "does this read AI-written?", or before shipping prose a human will read. Rewrites in four p

Use when the user pastes a ticket, design doc, or task description and wants it implemented. Multi-phase flow — understand, investigate (in plan mode), plan, implement, verify. Enforces strict scope rules and writes failing tests first for bug fixes. Also known as `klaussy-implem

Use when the user wants a new git worktree created for a task. Picks a kebab-case branch name with a fix/feat/chore/docs/refactor prefix, runs `git worktree add` from the configured base branch, and reports the new path. Also known as `klaussy-new-worktree`.

Use when reviewing a staged diff or an about-to-commit/push change for last-mile issues — silent failures, leaked secrets, debug leftovers, blatant correctness landmines, and excessive/narrating comments. Reports findings on the changed lines only; it does not refactor or rewrite

Use when the user wants the current change QA'd and PR-ready evidence captured. Classifies the diff and runs the verification that actually fits it — screen recordings and screenshots for UI/frontend changes, endpoint or e2e runs for backend, command output for a CLI, tests for a

Use when the user wants to restructure code while preserving behavior exactly. Establishes a passing test baseline first, then makes incremental moves that each leave the suite green. Refuses to change behavior and structure in the same step. Also known as `klaussy-refactor`.

Use when the user wants to cut a release — bump the version, update the changelog from conventional commits, and tag. Detects where the version lives, derives the next version from the commits since the last tag, and stages the release locally; it does not push or publish unless

Use when designing or analyzing a controlled experiment — falsifiable hypothesis, sample size from an MDE, reading significance/CI/power, CUPED, or rescuing tests that won't go significant. NOT recurring metric tracking (that is `analytics`), NOT north-star/KPI trees (that is `kp

Use when making a web UI conform to WCAG 2.2 Level AA — axe-core or Lighthouse a11y violations, keyboard operability, focus management, ARIA roles/names/live regions, contrast, tap-target size. NOT palette or visual intent (that is `design`), NOT test-runner setup (that is `testi

Use when running or fixing paid acquisition on Google or Meta — campaign structure (Performance Max, Demand Gen, Search, Advantage+), platform-fit creative, budget/scaling rules, break-even ROAS math, and Consent Mode v2 / CAPI tracking gaps. NOT the page the ad clicks into (that

Use when measuring whether an LLM or agent system actually got better and gating merges on it: golden sets, fixing an inflated LLM-as-judge, scoring RAG (faithfulness, contextual recall) or agent trajectories (tool correctness, completion), or picking an eval framework. NOT build

Use when bounding an LLM agent that already runs — scoping its task domain, gating tools to least privilege, defending against prompt injection in untrusted web/email/RAG text, requiring human approval on irreversible actions, capping runtime and cost, or triaging what it already

Use when a creative goal must become a finished media file: pick and order generative-media models per modality — AI voiceover, image-to-video clips, score — then glue them with ffmpeg (mux, duck, loudnorm, concat). NOT still-image generation/editing (that is `replicate-images`);

Fourteen posts of being wrong in production, compressed to checkboxes

Healthy nodes, a quiet network, 300 restarts in three days, and a latency budget measured in milliseconds

Discovery worked. Ping worked. Every TCP connection timed out, and later the tunnel only worked when someone had a terminal open.

Every VM came back. The cluster did not. Declarative systems converge on config, and the datapath isn't config.

A surprising share of AI-in-the-terminal failures aren't the AI. They're zsh, and a version of bash from 2006.

A Claude Code plugin turns standalone project configuration into a namespaced, installable extension that teams and communities can update as one unit.

None of the safety came from the model. It came from six boring habits.

Skills package instructions and references. Subagents run work in a separate context and return results. They solve different problems and can be composed deliberately.

Six hours in, one step left, everything green, and the incident that didn't happen

CLAUDE.md carries persistent project context. Skills load reusable procedures when relevant. Separating stable facts from task-specific workflows keeps both easier to maintain.

Twenty minutes recovering secrets that never existed, and the one sentence from a human that ended it

An API request routing a model's tool call through an approval gate to a remote MCP server

31 config keys, two audits, and why the first one was wrong in both directions

The official MCP Registry stores standardized server metadata rather than package code. Publishers verify a namespace, describe installation or remote access, and submit immutable versions.

Everyone looks at the Dockerfile. The file that actually leaked the key was the project file.

Remote MCP authorization uses established OAuth standards, but secure integration still requires issuer validation, least-privilege scopes, protected token handling, and server-side enforcement.

"Copy it over and switch the reference" is two steps, and the outage lives in the one nobody checks

stdio fits local processes and prototypes. Streamable HTTP fits hosted services and shared integrations. The right choice follows where the capability runs and who must reach it.

The most important rule wasn't about what I could change. It was about what I was allowed to display.

Tools perform operations, resources expose readable context, and prompts provide reusable templates. Choosing the correct primitive makes an MCP server easier to understand and govern.

Run autonomous hunt loop on a target — scope check → recon → rank surface → hunt → validate → report with configurable checkpoints. Usage: /autopilot target.com [--paranoid|--normal|--yolo]

Build an exploit chain — given bug A, finds B and C to combine for higher severity and payout. Knows common chain patterns: IDOR→ATO, SSRF→cloud metadata, XSS→ATO, open redirect→OAuth theft, S3→bundle→secret→OAuth. Usage: /chain

Active vulnerability hunting. Two-track dispatcher — asks Red Team vs WAPT, hands off to hunt-dispatch skill and sibling commands. Usage: /hunt target.com | /hunt *.target.com | /hunt targets.txt [--vuln-class X] [--source-code P] [--chrome]

On-demand intelligence fetch for a target — CVEs, disclosed reports, new features. Pulls NVD/GitHub-Advisory CVEs + bundled disclosed reports + hunt memory context. Usage: /intel target.com

Inspect or rotate the autopilot ledger JSONL files (findings.jsonl, negatives.jsonl). Caps file size and keeps N rotated backups so memory does not grow unbounded.

Pick up a previous hunt on a target — shows hunt history and untested surface from the autopilot ledger. Usage: /pickup target.com

Run full recon pipeline on a target — subdomain enum (Chaos API + subfinder), live host discovery (dnsx + httpx), URL crawl (katana + waybackurls + gau), gf pattern classification, nuclei scan. Outputs to recon/<target>/ directory. Usage: /recon target.com

Optional manual note on a target or the last confirmed finding. Capture is automatic during autopilot; this is for extra context. Usage: /remember

Write a submission-ready bug bounty report. Generates H1/Bugcrowd/Intigriti/Immunefi format with CVSS 3.1 score, proof of concept, impact statement, and remediation. Run /validate first. Usage: /report

Mandatory pre-flight scope check — verify an asset is in scope BEFORE any HTTP touch. Deterministic (deny-wins, default-deny) via engine/scope.py against the engagement's scope.md. Blocks out-of-scope testing. Usage: /scope <asset> [<asset> ...]

Show ranked attack surface for a target from its recon manifest + hunt memory. Deterministic backing is `cbh surface <target>` (reads recon/<target>/manifest.json); LLM layer adds ledger signal. Usage: /surface target.com

Meme coin and token security scan — checks for rug pull vectors (hidden mint, honeypot, fee manipulation, LP lock bypass, authority retention, bonding curve exploits, fake renounce, sandwich amplification). Manual 8-class grep audit (with an optional automated scanner if present). Usage: /token-scan <contract_path_or_dir> [--chain solana]

Quick 7-Question Gate triage on a finding before writing a report. Kills N/A submissions before they happen. Faster than /validate — for quick go/no-go decisions. Usage: /triage

Validate a finding — runs 7-Question Gate + 4-gate checklist. Kills weak findings before report writing. Prevents N/A submissions that hurt validity ratio. Usage: /validate

Smart contract security audit — runs through 10 bug class checklist (accounting desync, access control, incomplete path, off-by-one, oracle errors, ERC4626, reentrancy, flash loan, signature replay, proxy/upgrade). Applies pre-dive kill signals first. Generates Foundry PoC template for confirmed findings. Usage: /web3-audit <contract.sol>

Crabbox is a single CLI (`crabbox`). Commands are top-level, not nested under a

`crabbox actions` prepares a leased box from your repository's own GitHub

See [Runtime adapter stack](../features/runtime-adapter-stack.md) for the

`crabbox admin` groups trusted operator controls for coordinator-backed leases and the cloud resources behind them. Use it to inspect every lease the broker tracks, reconcile expired leases against live cloud state, force-release or delete a backing server, print provider IAM pol

`crabbox artifacts` turns a desktop lease into durable QA evidence: it collects

The fastest way to put Volcengine Ark in your terminal and your AI agent — go from prompt to generated media, multimodal answer, or deployed endpoint in a sin…

本地私有、开源的自进化跨平台 AI 内容发现 Agent：先理解你，再主动从 B站、小红书、抖音、YouTube、X、知乎、Reddit、微博等平台与开放 Web 寻找内容。（支持 deepseek harness 插件） | Local-first open-source cross-platform AI cont…

Persistent memory for AI coding agents — one verified kb_search replaces the grep/find/ls orientation loop. Cross-repo, CPU-only, zero token spend.

AI 时代的伯克希尔：基于 Claude Code / Codex 的价值投资研究框架。巴菲特·芒格·段永平·李录四大师方法论 + 多Agent并行研究。| AI-era Berkshire: a value investing research framework built for Claude Code / Co…

The batteries-included, No-Code FinOps automation platform, with the AI you trust.

Open-source 3D AI agent framework — GLB/glTF avatars with LLM brains, memory, emotions, and autonomous payments. MCP server · x402 · Solana/EVM · Three.js. Embe…

XLSX parser for LLMs, RAG, LangChain, LangGraph, CrewAI, Claude, MCP — turns Excel (.xlsx) into citation-ready JSON with formulas, charts, dependency graphs, an…

Hermes-Relay — Your Hermes AI agent, in your pocket — chat, voice, and control.

A minimalist, terminal-native coding agent written in C.

AI-powered OSINT agent with interactive REPL, MCP server, and CLI. 19 tools. Works with Claude, GPT-4, or local models. For authorized security research only.

AI pair programming in your terminal — one static binary, sub-ms startup, any model

Where data access meets operational intelligence

Build your own security agents. Open-source framework for agents with live, read-only access to your infrastructure, with no path to widen it. Reasons across AW…

Multi-workspace terminal aggregator with Claude Code AI integration

Go implementation of AI coding agent

Harness engineering beginner tutorial, from 0 to 1

Generate images directly in DeepSeek Harness chats

A smarter, self-hosted AI assistant — multi-user, multi-agent.

Turn any research paper into a commercialization report — 6 AI agents, TRL/MRL scoring, patent landscape, market intelligence, verified citations. DeepSeek / Op…

Power BI CLI - semantic models (.NET TOM) and PBIR reports for token-efficient AI agent usage, built for Claude Code