Promptguard

Detect prompt injection attacks in text. Returns risk score and detected patterns.

Security Scan

The package and SKILL.md are internally consistent: it implements a local FastAPI service that scans text with regex patterns for prompt-injection indicators and does not request unrelated credentials or perform hidden network/file actions.

Purpose & Capability

Name/description, required binaries (python), declared Python packages, and the included source code all match a simple local prompt-injection detector. Nothing in the files indicates unrelated capabilities (cloud access, system configuration, or secrets handling).

Instruction Scope

SKILL.md instructs to install FastAPI/uvicorn/pydantic, run uvicorn to start the server, and POST text to /v1/scan. The code implements exactly that endpoint and only inspects the provided text; it does not read files, environment variables, or external endpoints.

Install Mechanism

The declared packages are standard PyPI packages (fastapi, uvicorn, pydantic), which is appropriate. The install 'kind' is listed as 'uv' in metadata (unusual/ambiguous), but the SKILL.md shows pip instructions and the code is pure Python — this looks low risk but you may want to confirm how your platform's installer resolves 'uv'.

Credentials

No environment variables, credentials, or config paths are requested. The skill does not attempt to access unrelated secrets or system config.

Persistence & Privilege

always is false and the skill does not try to modify other skills or system-wide settings. It runs as a local HTTP service and requires explicit startup; it does not persist elevated privileges.

Assessment

This skill appears to be a straightforward, local prompt-injection scanner. Before installing: (1) confirm how your platform implements the install 'uv' step (ensure it installs from trusted PyPI sources); (2) don't expose the server publicly without authentication because the provided service accepts arbitrary text and has no auth built in; (3) keep dependencies up to date and run in an isolated environment if you are concerned about supply-chain risk; (4) test with representative inputs to validate detection coverage and false positives/negatives for your use case.

A security API that scans text for common prompt injection patterns and returns a risk score. Designed for AI agents that process untrusted text input from external sources.

What It Detects

• Instruction override attempts
• HTML comment injection
• Zero-width unicode characters
• Delimiter-based attacks
• Role switching tokens
• System prompt extraction attempts

Installation

pip install fastapi uvicorn pydantic

Usage

Start the server:

uvicorn promptguard.app:app --port 8000

Then send a POST request:

curl -X POST http://localhost:8000/v1/scan \
  -H "Content-Type: application/json" \
  -d '{"text": "What is the weather in London today?"}'

Response (clean text):

{
  "risk_score": "0",
  "patterns_detected": [],
  "input_length": 38
}

Request

Field	Type	Required	Description
text	string	yes	Text to scan (1-100,000 chars)

Response

Field	Type	Description
risk_score	decimal	0.0 (safe) to 1.0 (high risk)
patterns_detected	list	Names of detected patterns
input_length	integer	Length of input text