Skip to content
Get packs

Competition Prompt Injection

Guides an agent through analyzing prompt-injection and tool-boundary flaws in CTF-style sandbox challenges.

Source Download skill
Repository
zhaoxuya520/reverse-skill
Stars
43,969
Licence
MIT
Files
3 files

What it does

This skill is a specialist add-on for a CTF (capture-the-flag) sandbox workflow. Once installed and triggered by a parent skill called ctf-sandbox-orchestrator, it tells the agent how to investigate how untrusted text (retrieved documents, memory, planner state, tool outputs) can hijack an AI agent's behavior, such as changing tool arguments or leaking secrets. It asks the agent to map trust layers, prove one minimal exploit chain, and report which component failed, and to reply in Simplified Chinese unless English is requested.

Great for

  • Students or competitors working on AI agent security CTF challenges in a sandbox
  • Security researchers documenting prompt-injection or tool-argument abuse in a controlled lab
  • Teams already using the ctf-sandbox-orchestrator skill who want this downstream specialization

Skip it if

  • Attacking real, production or third-party agent systems, since the skill is designed for isolated competition sandboxes and gives no authorization or scope checks
  • Users who have not installed ctf-sandbox-orchestrator, because the SKILL.md says it should only run after that parent skill has set up sandbox assumptions
  • Anyone expecting a general prompt-writing or defensive hardening guide; the content is oriented to finding and proving exploits

Try asking

  • “Analyze why a retrieved document in my CTF sandbox changed the planner's next tool call”
  • “Trace how memory contamination in this agent chain led to a secret being exposed”
  • “Build a minimal reproduction transcript showing where untrusted text became a file path argument”
  • “Summarize which layer failed (retrieval, planner, executor, or output post-processing) in this challenge”
Step-by-step guide: how to use Competition Prompt InjectionInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 3 original files, the MIT licence and an attribution note.

    competition-prompt-injection.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip competition-prompt-injection.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip competition-prompt-injection.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Install and activate ctf-sandbox-orchestrator first; this skill will tell you to return to it otherwise
  • Keep all testing inside the designated sandbox and only against challenge targets you are authorized to test
  • Ask for the smallest reproducible transcript rather than broad exploration, as the skill recommends

What's inside

  • agents/openai.yaml
  • references/prompt-injection.md
  • SKILL.md

Redistributed unmodified from zhaoxuya520/reverse-skill under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.