What problem does it solve? When tackling CTF challenges involving AI agents, prompt injection, cloud infrastructure, containers, or CI/CD pipelines, it is hard to systematically connect untrusted input to tool side effects or reconcile checked-in manifests with live runtime state. This Skill provides a structured methodology for that analysis under established sandbox assumptions. ## Core Features & Use Cases - Agentic Path Analysis: Map instruction layers, retrieval boundaries, memory files, and tool gates to prove a minimal chain from untrusted content to tool side effects. - Cloud and Container Reconciliation: Compare compose or Kubernetes manifests against live mounts, environment variables, logs, and traffic across build-time, deploy-time, and runtime. - Supply-Chain Provenance Tracing: Track the chain from source to dependency resolution, build, publish, and runtime consumer, focusing on version drift and registry pulls. - Use Case: In a CTF challenge where a prompt-injected agent triggers an unintended tool call, use this Skill to build a compact evidence block linking the retrieved chunk, planner transition, and final tool arguments. ## Quick Start After the ctf-sandbox-orchestrator skill is active, ask the agent to trace the prompt-to-tool flow or deployment drift in this challenge using competition-agent-cloud.
