feat: Add Crucible agentic challenge attack notebook

[rdheekonda]

Summary

• Adds examples/airt/agentic_red_teaming_attacks.ipynb demonstrating automated adversarial attacks against three Crucible agentic challenges (toolshed, webwhisper, vaultguard)
• Uses TAP, GOAT, and Crescendo attack strategies with diverse transforms (perturbation, role-play, authority exploitation)
• Includes real-time console() output, contains_crucible_flag scoring, and recommended mitigations

Test plan

• Run notebook end-to-end against platform.dreadnode.io
• Verify all 6 attack cells execute and produce results
• Confirm contains_crucible_flag scorer validates captured flags

🤖 Generated with Claude Code

Generated Summary:

• Added a new Jupyter Notebook for Agentic AI Red Teaming utilizing the AIRT framework.
• Implemented automated adversarial attacks against multiple challenges in the Dreadnode Crucible platform.
• Included various attack methodologies: TAP (beam search), GOAT (graph exploration), Crescendo (progressive escalation).
• Defined targets and scoring mechanisms for each challenge:
  • toolshed: Focus on DevOps Tool Misuse.
  • webwhisper: Address Indirect Prompt Injection.
  • vaultguard: Engage in Multi-Agent Defense Bypass.
• Provided environmental setup instructions and required API key configurations.
• Added multiple code cells demonstrating setup for targets, transformations, and scoring.
• Highlighted complexities such as input filters, system prompts, and security measures in the attacks.
• Enhanced documentation for each challenge outlining objectives and defenses in a structured format.

This summary was generated with ❤️ by rigging