AGENT PROMPT SECURITY

Paste your agent's prompt. Watch it get autopsied.

Find secrets, injection surface, over-permissioning before your agent ships — in seconds.

SECRET INJECT GUARDRAILS JSON A–F E-01 E-02 AUTOPSY FIG. 01
LOCAL MODEPAX-01-01
CHARS 0 / 12,000
AUTOPSY / 01 PROMPT
Your prompt is processed transiently and never stored. No client-side key — the API key never leaves the server.
REF / 03 SAMPLES LOAD SAMPLE synthetic sample
FINDINGS / 04 CLASSES
No report yet. Run AUTOPSY or load the demo.
SCORE / 02 EXPOSURE
–EXPOSURE
GRADE–
CRITICAL0
HIGH0
MEDIUM0
LOW0

How it works

METHOD / 03 STEPS
01

Worker receives the prompt; nothing stored.

02

DeepSeek probes secrets and injection.

03

Get an evidence-backed report.

Scan the web surface too → Talon Audit