Prophet

Prophet is a policy-bound evidence system for defensive exposure prioritization. It helps a security team decide which exposure class to harden first, explain why that decision is reasonable, validate only in safe fixture or approved-sandbox modes, and hand off review artifacts to SOC and platform teams.

prioritize exposure → explain evidence → validate safely → hand off review artifacts

The product claim is deliberately narrow. Prophet does not discover zero-days, generate offensive payloads, test live infrastructure, or deploy controls autonomously. It produces an auditable "why this first" package from approved asset/SBOM metadata, public vulnerability context, policy checks, deterministic validation summaries, and safe handoff templates.

The full operator manual — every Make wrapper, validation-sprint command, and CLI contract — lives in docs/README_DETAILED.md and docs/CLI_REFERENCE.md.

Why this exists

When the vulnerability queue is larger than the team can clear, the same operational question recurs: what should be hardened first, and what evidence proves that decision was reasonable? CISA KEV/BOD pressure, EPSS signals, SBOM obligations, and leadership scrutiny all push on it. Prophet's wedge isn't another scanner or exposure graph — it's a repeatable defensive evidence loop:

• Prioritize — rank the exposure class that deserves attention first.
• Explain — show source basis, asset/SBOM basis, confidence, freshness, and assumptions.
• Validate safely — deterministic fixtures, localhost, or an explicitly approved sandbox only.
• Hand off — export evidence, audit, SIEM, and ticketing review templates with no production pushes.

What works today

Surface	Notes
Forecaster	Deterministic Python; no runtime LLM or external API dependency.
Asset import	Customer-safe CSV metadata with row-level cleanup reports and seedsets.
Asset-seeded OSINT	Policy-gated, fixture-backed public-source snapshots.
Contract validators	Reject payloads, credentials, live targets, procedural instructions, schema drift.
Exposure-class portfolio	Safe, non-operational defensive class recommendations for analyst review.
Sandbox runner	Deterministic localhost fixture validation for the edge-appliance profile.
Evidence export	Policy-bound JSON + Markdown evidence bundles from validated fixtures.
Integration handoff	Safe SIEM and ticketing review templates from validated evidence.
Policy linting	Validates pilot policies, allowed modes, source IDs, sandbox profiles, output paths.
React operator console	Fixture-backed end-to-end replay with human gate and validation status.

Live collection (PROPHET_ENABLE_VM_SCRAPER=1) is disabled by default and requires an approved isolated plan. Private lab research is not packaged — the public repo ships safe interfaces and fixtures only.

Architecture

approved asset/SBOM metadata        sanitized public context
            |                                |
   assets/ import + seedset          world-side/ forecaster
            |                                |
            +----------------+---------------+
                             v
                cyber-side/ defensive portfolio
                             v
                sandbox_runner/ fixture validation
                             v
                evidence/ bundle + audit trail
                             |
            +----------------+---------------+
            v                                v
 integrations/ review templates    prophet-console/ evaluator UI

The contracts are the product boundary: world-side/INTERFACE.md (candidate/forecast schemas), cyber-side/INTERFACE.md (defense artifact schema), cyber-side/validator.py (payload and live-target rejection), world-side/forecaster/models.py (forecast safety and schema validation).

Quickstart

Prove the defensive pilot loop from a fresh clone, no live collection, no code reading. Requires Python 3.9+.

./scripts/check-local-env.sh
./scripts/run-pilot-demo-smoke.sh

This runs policy linting, safe asset import, seeded OSINT, forecast refresh, sandbox validation, evidence + audit export, and SIEM/ticketing handoff — all fixture-backed and localhost-only — then verifies the output hash against scripts/pilot-demo-smoke.sha256.

Run the full Python contract suite:

make python-tests

Run the evaluator console (localhost-only control API + UI in one terminal):

(cd prophet-console && npm ci)
make console-demo   # UI at http://127.0.0.1:5173, control server at http://127.0.0.1:8787

Safety model

Prophet is built to be evaluated as a defensive decision-and-validation system, not an exploit-delivery system.

• No live infrastructure testing from the public console.
• No raw scraper output crosses into the main app.
• No credentials, IPs, or operational hostnames in committed artifacts.
• No payload bytes in accepted JSON contracts.
• OSINT snapshots use only approved source IDs; sandbox_runner uses only approved profiles.
• Production validation runs only in approved, isolated, vulnerable-by-design sandboxes.

See SECURITY.md for operating rules.

Repository map

assets/             Customer-safe asset/SBOM fixtures, import CLI, tests.
cyber-side/         Defensive portfolio fixtures, artifact contract, validator.
docs/               Pilot, validation, safety, release, and readiness docs.
evidence/           Policy-bound JSON/Markdown evidence bundles and audit helpers.
integrations/       SIEM, ticketing, and audit handoff template exporters.
policy/             Pilot policy schema, examples, linting, and tests.
prophet-console/    React operator console, localhost control server, browser tests.
sandbox_runner/     Deterministic localhost sandbox simulation profiles.
scripts/            Smoke, validation, release-safety, and operator helpers.
world-side/         Forecaster, source sanitization, fixtures, and outputs.

Stack: pure-Python standard library on the backend (no numpy/pandas/requests — the forecasting is heuristic and deterministic by design); React 19 + TypeScript + Vite for the console. Lab-only exploit validation scaffolding is not part of the public product tree.