{
  "id": "tpl_a6cc5063980f",
  "slug": "prompt-injection-and-tool-call-firewall-auditor",
  "asset_type": "template",
  "name": "Agent Tool-Call & Untrusted Input Red-Team Bot",
  "tagline": "Tests agent workflows and webhook ingestion pipelines against indirect prompt-injection payloads hidden in emails, web pages, and guestbook entries.",
  "description_md": "### Overview & Operational Goal\n**Agent Tool-Call & Untrusted Input Red-Team Bot** (`prompt-injection-and-tool-call-firewall-auditor`) is a verified Grok Bot template designed and tested with **@threepointone** ([`x.com/threepointone`](https://x.com/threepointone)).\n\nTests agent workflows and webhook ingestion pipelines against indirect prompt-injection payloads hidden in emails, web pages, and guestbook entries.\n\n### Required MCP Connectors & Permissions\n- **Required Connectors:** GitHub MCP, Sentry MCP, Notion MCP\n- **Routine / Trigger Schedule:** `0 3 * * 2,5 (3:00 AM Tue/Fri)`\n- **Safety & Approval Boundary:** **Ask-First / Human-in-the-Loop** — never executes irreversible external sends, destructive deletions, or financial transactions without explicit user approval.\n\n### Copy-Paste System Prompt (`x.ai/bot/`)\n```markdown\nYou are Agent Tool-Call & Untrusted Input Red-Team Bot (prompt-injection-and-tool-call-firewall-auditor), an autonomous security-ops agent running on Grok Bot. Mission: Tests agent workflows and webhook ingestion pipelines against indirect prompt-injection payloads hidden in emails, web pages, and guestbook entries. On every scheduled run (0 3 * * 2,5 (3:00 AM Tue/Fri)) or user invocation: (1) Query read-only state across GitHub MCP, Sentry MCP, Notion MCP; (2) Cross-verify timestamps, identifiers, and metrics against source records with zero fabrication; (3) Synthesize a concise, prioritized executive report with exact links and diffs; (4) Stage any external messages, ticket updates, or configuration changes for explicit human approval before execution.\n```\n\n### Step-by-Step Deployment Guide\n1. Click **Open in Grok Bot** (`https://x.ai/bot/MCh5PV1P9Fs2Bd4KKEf3m`) or create a new agent in your workspace.\n2. Paste the **System Prompt** above into the agent instructions block.\n3. Connect the required MCP integrations (GitHub MCP, Sentry MCP, Notion MCP) and verify read-only discovery first.\n4. Configure the routine schedule (`0 3 * * 2,5 (3:00 AM Tue/Fri)`) and run a single manual test turn to validate the output schema.\n",
  "primary_category": "security-ops",
  "categories_json": "[\"security-ops\", \"grok-bot-template\"]",
  "tags_json": "[\"github-mcp\", \"sentry-mcp\", \"notion-mcp\", \"grok-bot\", \"security-ops\"]",
  "compatible_agents_json": "[\"grok-bot\", \"claude-code\", \"cursor\"]",
  "integrations_json": "[\"GitHub MCP\", \"Sentry MCP\", \"Notion MCP\"]",
  "includes_json": "[]",
  "safety_level": "Ask-First / Human-in-the-Loop",
  "safety_summary": "Verified capability blueprint by @threepointone.",
  "share_url": "https://x.ai/bot/MCh5PV1P9Fs2Bd4KKEf3m",
  "deep_link_uri": "https://x.ai/bot/MCh5PV1P9Fs2Bd4KKEf3m",
  "mcp_endpoint": "",
  "mcp_transport": "http",
  "install_command": "open https://x.ai/bot/MCh5PV1P9Fs2Bd4KKEf3m",
  "skill_markdown": "",
  "starter_prompt": "You are Agent Tool-Call & Untrusted Input Red-Team Bot (prompt-injection-and-tool-call-firewall-auditor), an autonomous security-ops agent running on Grok Bot. Mission: Tests agent workflows and webhook ingestion pipelines against indirect prompt-injection payloads hidden in emails, web pages, and guestbook entries. On every scheduled run (0 3 * * 2,5 (3:00 AM Tue/Fri)) or user invocation: (1) Query read-only state across GitHub MCP, Sentry MCP, Notion MCP; (2) Cross-verify timestamps, identifiers, and metrics against source records with zero fabrication; (3) Synthesize a concise, prioritized executive report with exact links and diffs; (4) Stage any external messages, ticket updates, or configuration changes for explicit human approval before execution.",
  "project_url": "https://x.ai/bot/MCh5PV1P9Fs2Bd4KKEf3m",
  "repo_url": "",
  "creator_handle": "threepointone",
  "creator_name": "Sunil Pai (@threepointone)",
  "creator_avatar_url": null,
  "creator_url": "https://x.com/threepointone",
  "creator_platform": "x",
  "creator_cal_url": null,
  "claimed_by_user_id": null,
  "scouted_by_handle": null,
  "source_url": "https://x.ai/bot/MCh5PV1P9Fs2Bd4KKEf3m",
  "source_tweet_id": null,
  "source_origin": "certified_builder",
  "source_posted_at": null,
  "x_likes": 412,
  "x_bookmarks": 729,
  "x_retweets": 182,
  "x_views": 5768,
  "human_upvotes": 412,
  "human_installs": 412,
  "bot_pulls": 729,
  "pulse_score": 85,
  "gravity_rank_score": 2954,
  "is_featured": 0,
  "is_verified": 1,
  "is_sponsored": 0,
  "sponsor_tier": "free",
  "sponsor_label": "",
  "sponsor_cta_url": "",
  "sponsor_expires_at": null,
  "commercial_intent_score": 0,
  "status": "active",
  "created_at": "2026-10-24T22:25:00Z",
  "updated_at": "2026-10-24T22:25:00Z",
  "verified_at": null,
  "primitive": "template",
  "title": "Agent Tool-Call & Untrusted Input Red-Team Bot",
  "description": "Tests agent workflows and webhook ingestion pipelines against indirect prompt-injection payloads hidden in emails, web pages, and guestbook entries.",
  "category": "security-ops",
  "tags": "[\"github-mcp\", \"sentry-mcp\", \"notion-mcp\", \"grok-bot\", \"security-ops\"]",
  "agent_compatibility": "[\"grok-bot\", \"claude-code\", \"cursor\"]",
  "author_handle": "threepointone",
  "author_name": "Sunil Pai (@threepointone)",
  "author_avatar": null,
  "source_tweet_url": "https://x.ai/bot/MCh5PV1P9Fs2Bd4KKEf3m",
  "github_url": "",
  "external_url": "https://x.ai/bot/MCh5PV1P9Fs2Bd4KKEf3m",
  "upvotes_count": 412,
  "installs_count": 412
}