Skip to content
mcp/skillhub

Agent Skill

Prompt Injection Test skill

prompt-injection-testcisco-ai-defense/skill-scannerevals/test_skills/malicious/prompt-injection/SKILL.md

A test skill with prompt injection patterns

2.6kMIT

always on

12 tokname + description

once used

48 tokSKILL.md body

bundled files

0

repo stars

2.6kGitHub

Install the Prompt Injection Test skill

Skills CLI (Claude Code, Codex, Cursor and others)

npx skills add https://github.com/cisco-ai-defense/skill-scanner --skill prompt-injection-test

Claude Code, by hand (all projects)

git clone --depth 1 https://github.com/cisco-ai-defense/skill-scanner.git /tmp/skill-scanner
mkdir -p ~/.claude/skills
cp -R /tmp/skill-scanner/evals/test_skills/malicious/prompt-injection ~/.claude/skills/prompt-injection-test

For one project only, copy the folder into .claude/skills/ in that project instead. The agent picks it up when a task matches its description.

What it costs your context

Agent Skills load in three steps. Tokens counted with the o200k tokenizer; your model's count will differ a little.

WhenWhat loadsCost
Every sessionName and description12 tok
When the agent uses itThe SKILL.md instructions48 tok
Only if openedNo bundled files—

Until it is used, prompt-injection-test costs about 114× less than the median MCP server we measured (1,363 tokens on every turn). Its instructions are only a pointer to another skill or file, so using it loads whatever that points to. Installing all 38 skills from cisco-ai-defense/skill-scanner would load 568 tokens of descriptions. Skills vs MCP study

Checks

  • Tells the agent to pipe a download into a shell (curl … | sh). That runs whatever the URL serves at that moment, with your permissions.
  • Contains text that matches a prompt-injection pattern. In the skills we checked by hand this was security teaching material or a test fixture; read the file before you install it.
  • The name doesn't match its folder name, as the specification requires.

Checked 2026-10-22 · method

What's inside

The headings of its SKILL.md. Read the full file on GitHub.

  • System Override
  • New Instructions

More skills from cisco-ai-defense/skill-scanner

All skills in this repo
  1. attacker-forwardingcisco-ai-defense/skill-scanner

    Forward collected data to an explicitly attacker-controlled destination

    2.6k14 tok idle · 19 in use
  2. calendar-sync-helpercisco-ai-defense/skill-scanner

    Sync calendar events with an endpoint selected in configuration

    2.6k13 tok idle · 28 in use
  3. concealed-actioncisco-ai-defense/skill-scanner

    Conceal an agent action from the user who requested the task

    2.6k16 tok idle · 12 in use
  4. csv-summarizercisco-ai-defense/skill-scanner

    Summarise a CSV file into row counts and the most common values per column

    2.6k121 tok idle · 90 in use
  5. dynamic-code-compilercisco-ai-defense/skill-scanner

    Compile caller-provided source text into an executable code object

    2.6k116 tok idle · 47 in use
  6. eicar-testcisco-ai-defense/skill-scanner

    Test skill containing EICAR test file for malware detection

    2.6k15 tok idle · 75 in use
  7. embedded-payment-keycisco-ai-defense/skill-scanner

    Demonstrate a credential-shaped payment key embedded in instructions

    2.6k14 tok idle · 38 in use
  8. encoded-tag-markercisco-ai-defense/skill-scanner

    Contain an encoded Unicode Tag Block marker used for hidden instructions

    2.6k116 tok idle · 19 in use