competition-prompt-injection

1# Competition Prompt Injection
2 
3Use this skill only as a downstream specialization after `$ctf-sandbox-orchestrator` is already active and has established sandbox assumptions, node ownership, and evidence priorities. If that has not happened yet, return to `$ctf-sandbox-orchestrator` first.
4 
5Use this skill when the challenge is primarily about trust boundaries inside an agentic system.
6 
7Reply in Simplified Chinese unless the user explicitly requests English.
8 
9## Quick Start
10 
111. Identify the first untrusted content that becomes model-visible.
122. Map the chain from retrieval, memory, or transcript into planner or executor behavior.
133. Record the exact point where text becomes a tool argument, file path, network target, or secret request.
144. Prove one minimal exploit chain before exploring variants.
155. Keep prompt snippets and tool transitions in compact evidence blocks.
16 
17## Workflow
18 
19### 1. Map The Control Stack
20 
21- Track system, developer, user, retrieved, memory, planner, and tool-response layers separately.
22- Distinguish claimed capability from runtime-exposed capability.
23- Note what the model can actually call, read, or mutate.
24 
25### 2. Prove The Boundary Crossing
26 
27- Reproduce one chain from untrusted text to changed planner behavior, changed tool args, or secret exposure.
28- Keep the decisive transcript compact: source chunk, rewritten planner state, final tool invocation.
29- Prefer the smallest transcript that still demonstrates the bug.
30 
31### 3. Report By Boundary
32…