# MRPS Agent Toolkit
> A curated collection of open-source tools and agent skills for coding agents.

This directory indexes verified open-source developer tools, CLI utilities, libraries, and agent skills. Every entry links to upstream sources and provides strict operational boundaries for coding agents.

Start here if you are an agent asked to audit a project: read AGENTS.md, then load
runner/SKILL.md. Everything else is reference.

## Core Documents
- [Agent Guidance](https://mrps.in/agent-toolkit/AGENTS.md): Read-only audit protocol, permission ladder, report contract, honesty rules.
- [Launcher Skill](https://mrps.in/agent-toolkit/runner/SKILL.md): The full audit workflow an agent follows — menu, gate, report format, ID scheme.
- [Capability Index](https://mrps.in/agent-toolkit/index.json): Machine-readable index of all capabilities, with instruction file, id prefix, and report slug per tool.
- [Project Credits](https://mrps.in/agent-toolkit/CREDITS.md): Full attribution, repositories, and licenses.

## Tools
- [MRPS Agent Launcher](https://mrps.in/agent-toolkit/tools/mrps-agent-runner.html): Read-only project auditor and runner for code, security, a11y, and UI checks. Category: orchestration. Kind: skill. License: MIT.
- [Impeccable](https://mrps.in/agent-toolkit/tools/impeccable.html): The design vocabulary and UI critique system for coding agents. Category: ui-ux. Kind: skill. License: Apache-2.0.
- [Obscura](https://mrps.in/agent-toolkit/tools/obscura.html): High-performance Rust-based headless browser for AI agents and web scraping. Category: testing. Kind: app. License: Apache-2.0.
- [Playwright](https://mrps.in/agent-toolkit/tools/playwright.html): Reliable end-to-end multi-browser testing framework for modern web applications. Category: testing. Kind: library. License: Apache-2.0.
- [axe-core](https://mrps.in/agent-toolkit/tools/axe-core.html): Automated accessibility testing engine for HTML and web applications. Category: accessibility. Kind: library. License: MPL-2.0.
- [Trivy](https://mrps.in/agent-toolkit/tools/trivy.html): Comprehensive security scanner for vulnerabilities, secrets, and misconfigurations. Category: security. Kind: cli. License: Apache-2.0.
- [Graphify](https://mrps.in/agent-toolkit/tools/graphify.html): Codebase knowledge graph generator with deterministic AST parsing. Category: code-analysis. Kind: skill. License: Apache-2.0.
- [REA](https://mrps.in/agent-toolkit/tools/rea.html): Universal reverse engineering MCP server for binaries, runtimes, and application behavior. Category: reverse-engineering. Kind: mcp-server. License: MIT.
- [Cloudflare Security Audit](https://mrps.in/agent-toolkit/tools/cloudflare-security-audit.html): Multi-phase source-level security audit skill with independently verified, machine-readable findings. Category: comprehensive. Kind: skill. License: MIT.

## Usage Instructions
- [ACCESSIBILITY.md](https://mrps.in/agent-toolkit/instructions/ACCESSIBILITY.md): WCAG 2.2 AA checks with axe-core, severity mapping, and required manual checks.
- [CODE-ANALYSIS.md](https://mrps.in/agent-toolkit/instructions/CODE-ANALYSIS.md): Codebase knowledge graphs with Graphify; linting delegated to the project.
- [COMPREHENSIVE.md](https://mrps.in/agent-toolkit/instructions/COMPREHENSIVE.md): Six-phase, source-level security audit with schema-validated, independently verified findings.
- [GLOBAL.md](https://mrps.in/agent-toolkit/instructions/GLOBAL.md): Orchestration, permission ladder, report contract, and finding ID scheme.
- [REVERSE-ENGINEERING.md](https://mrps.in/agent-toolkit/instructions/REVERSE-ENGINEERING.md): Authorized artifact analysis with REA; never patch binaries.
- [SECURITY.md](https://mrps.in/agent-toolkit/instructions/SECURITY.md): Dependency CVEs, secrets, and misconfiguration with Trivy — plus redaction rules.
- [TESTING.md](https://mrps.in/agent-toolkit/instructions/TESTING.md): End-to-end tests with Playwright; Obscura as headless infrastructure, not a scanner.
- [UI-UX.md](https://mrps.in/agent-toolkit/instructions/UI-UX.md): Visual hierarchy, spacing rhythm, responsive integrity, and UI critique evidence rules.