Real-time risk scores and redaction for digital insiders.
Static and AST passes return a single risk score in seconds. No install required.
Static regex and Python AST passes detect debug prints leaking env vars, hard-coded credentials, wallet keys, and unsafe network or subprocess calls.
A FastAPI endpoint runs every tool output through the same regex ruleset as the auditor. Matched secrets are replaced with [REDACTED by Estes] before they reach your model context.
Map every finding to a watsonx.governance-aligned policy. Track skill provenance and enforce org-wide allowlists from a single console.
The same weather_tool skill, run twice. The leak on the left is a single print() away from polluting your model context.
# tool output → captured by agent → model context def get_weather(city): api_key = os.environ["OPENWEATHER_API_KEY"] print(f"DEBUG key={api_key}") return requests.get(url, params={ "key": api_key, }).json()
# intercepted at runtime → context stays clean POST /redact › in "DEBUG key=sk_live_4tQk29ZxbN8m..." › out "DEBUG key=sk_live_***REDACTED***" # compliance ledger { "rule": "OPENAI_KEY", "severity": "high", "skill": "weather_tool", "action": "redact" }