mcpField
C
competition-prompt-injection

competition-prompt-injection

github.com/zhaoxuya520Security43.7k93 installsUpdated 5w ago
Repository

About competition-prompt-injection

A downstream security skill for analyzing prompt-injection challenges in agentic systems. It helps security researchers trace how untrusted content crosses into planning or tool execution and document a minimal exploit chain.

Key features

  1. 01Maps system, user, retrieved, memory, planner, and tool layers
  2. 02Traces untrusted text into tool arguments or secret requests
  3. 03Distinguishes instruction drift from actual side effects
  4. 04Preserves compact evidence of the exploit chain

Use cases

  1. 01Investigating retrieved text that alters an agent's tool use
  2. 02Documenting a prompt-boundary failure with a minimal transcript

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/zhaoxuya520/reverse-skill --skill competition-prompt-injection

After installing, type/competition-prompt-injectionin Claude Code to confirm it loaded.

Bundled filesagents/openai.yamlreferences/prompt-injection.md

Details

From the project README

Competition Prompt Injection

Use 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.

Use this skill when the challenge is primarily about trust boundaries inside an agentic system.

Reply in Simplified Chinese unless the user explicitly requests English.

Quick Start

  1. Identify the first untrusted content that becomes model-visible.
  2. Map the chain from retrieval, memory, or transcript into planner or executor behavior.
  3. Record the exact point where text becomes a tool argument, file path, network target, or secret request.
  4. Prove one minimal exploit chain before exploring variants.
  5. Keep prompt snippets and tool transitions in compact evidence blocks.

Workflow

1. Map The Control Stack
  • Track system, developer, user, retrieved, memory, planner, and tool-response layers separately.
  • Distinguish claimed capability from runtime-exposed capability.
  • Note what the model can actually call, read, or mutate.
2. Prove The Boundary Crossing
  • Reproduce one chain from untrusted text to changed planner behavior, changed tool args, or secret exposure.
  • Keep the decisive transcript compact: source chunk, rewritten planner state, final tool invocation.
  • Prefer the smallest transcript that still demonstrates the bug.
3. Report By Boundary
  • State which layer failed: retrieval, summarizer, planner, executor, tool normalization, or output post-processing.
  • Separate instruction drift from actual side effect.

Read This Reference

  • Load references/prompt-injection.md for the checklist, evidence layout, and common prompt-boundary pitfalls.

What To Preserve

  • Original malicious chunk or prompt
  • Intermediate summary or planner drift if it matters
  • Final tool args, file paths, or exposed secret surface

Comments

Markdown supported