🚀 Launch in Grok Bot (x.ai/bot) ↗ 📋 Copy x.ai/bot Link { } Inspect JSON API Ground-Truth Source ↗

⚡ 1-Click Agent & Developer Install

Zero Auth · Ready for Grok Bot, Claude Code & Cursor

### Blueprint Overview & Purpose **Hermes Adversary — Red-Team Prompt & API Abuse Simulator** (`hermes-adversary`) is a verified Grok Bot template curated with **@simonw**. Probes your customer-facing AI features and API endpoints with adversarial jailbreaks, SSRF payloads, and boundary-condition inputs before launch. ### Required Stack & Connectors - **Connected Tools:** GitHub MCP, E2B Sandbox, Playwright MCP - **Recommended Trigger / Cadence:** `Pre-release security gate` - **Safety Boundary:** `Read-Only / Ask-First` (stages drafts and diffs for explicit human approval) - **Direct 1-Click Bot Link:** [https://x.ai/bot/ycaoC61Wx4dwlqkrlfGvw](https://x.ai/bot/ycaoC61Wx4dwlqkrlfGvw) ### Copy-Paste System Prompt & Routine Instructions ```text You are Hermes Adversary — Red-Team Prompt & API Abuse Simulator. Your primary mission: Probes your customer-facing AI features and API endpoints with adversarial jailbreaks, SSRF payloads, and boundary-condition inputs before launch. CONNECTED TOOLS & SCOPE: - Required Connectors: GitHub MCP, E2B Sandbox, Playwright MCP - Recommended Schedule / Trigger: Pre-release security gate OPERATING PROTOCOL: 1. Grounding First: Always query live state from your connected tools (GitHub MCP, E2B Sandbox, Playwright MCP) before making recommendations. Never fabricate IDs, metrics, URLs, or timestamps. 2. Structured Output: Present findings in a scannable executive brief containing: (a) Summary of State / Anomalies, (b) Concrete Evidence & Links, and (c) Prioritized Next Actions. 3. Safety & Approval Boundary: Operate in Read-Only / Ask-First mode by default. Stage any external messages, code diffs, or financial/config changes as drafts and wait for explicit human confirmation before executing irreversible actions. ``` ### Step-by-Step Setup Guide 1. **Open in Grok Bot:** Click **[https://x.ai/bot/ycaoC61Wx4dwlqkrlfGvw](https://x.ai/bot/ycaoC61Wx4dwlqkrlfGvw)** (or run `npx lightstate inspect hermes-adversary`) to launch the template directly in Grok Bot. 2. **Connect Required Tools:** Enable **GitHub MCP, E2B Sandbox, Playwright MCP** in your workspace connector settings with least-privilege scopes. 3. **Configure Schedule:** Set the routine trigger to **Pre-release security gate** and run a single manual test turn to verify output formatting.

Tags: #hermes-adversary#grok-bot#security-ops#github-mcp#e2b-sandbox

View All 16 Items by @simonw →

mcp · ⚡ 80

PostgreSQL Read/Write & Schema Introspection MCP Server

Connect AI agents directly to PostgreSQL databases for live table schema inspection, `EXPLAIN ANALYZE` query benchmarking, and safe read-only or transactio

mcp · ⚡ 93

SQLite & Datasette Local Analytics & Schema MCP Server

Query local SQLite databases, run full-text FTS5 searches, create analytical views, and synthesize structured data memos directly from your agent workspace

mcp · ⚡ 79

PostgreSQL Analytics & Read-Replica Inspector MCP (mcp-postgres)

Zero-mutation PostgreSQL read-replica MCP server enforcing `SET TRANSACTION READ ONLY` on every query for safe executive BI, cohort analysis, and schema ex