prompt-injection-risk-linter

When to invoke

• You are building an agent that reads untrusted content (web pages, emails, tickets) and you want a pre-flight safety lint.
• You want to add an automated check to prompt templates before deployment.

Inputs needed

• --prompt path to a text file containing a system/developer prompt, or combined prompt template.
• Optional: --retrieved path to a text file with representative untrusted content.

Workflow

1. Detect common prompt-injection markers ("ignore previous instructions", requests to reveal hidden prompts, tool/credential exfiltration).
2. Check for missing boundaries (no explicit statement that retrieved content is untrusted; no tool-use constraints).
3. Emit a JSON report with severity, evidence snippets, and recommended mitigations.

Output format

• JSON report written to --output.

Guardrails

• Heuristics only: do not claim the prompt is safe/unsafe with certainty.
• Avoid printing full prompt contents to stdout.