Skip to content

Security: fbailleux2/flowfabric

Security

SECURITY.md

Security Policy

Overview

FlowFabric is an enterprise workflow orchestration platform designed to handle business process data, including data about employees and organizational operations. Its security model is part of its core design, not an afterthought.

This document covers:

  • How to report vulnerabilities responsibly
  • What is in scope and out of scope
  • Known architectural security constraints
  • Our commitment on response timelines

Reporting a Vulnerability

Do not open a public GitHub issue for security vulnerabilities.

Report security issues by email to: franck.bailleux@gmail.com

If you do not receive an acknowledgment within 72 hours, follow up directly via a GitHub issue marked [SECURITY] — without disclosing the technical details.

What to include in your report

  • A clear description of the vulnerability
  • Steps to reproduce or a proof-of-concept (if applicable)
  • The component affected (jean-agent, jean-aggregator, FlowFabric Core, KFabric connector, etc.)
  • The potential impact in your assessment
  • Your name / handle if you want credit (optional)

We will acknowledge your report, keep you informed of our progress, and credit you publicly (if you wish) when a fix is released.


Response Timeline

Milestone Target
Acknowledgment 72 hours
Initial triage 7 days
Fix timeline communicated 14 days
Critical fix released 30 days (best effort)

For critical vulnerabilities with active exploitation, we aim to release a patch within 7 days.


Scope

In Scope

Component Notes
jean-agent Any vulnerability allowing unauthorized data capture, privilege escalation, tampering, or bypass of the collection policy
jean-aggregator Data injection, authentication bypass, pattern manipulation, unauthorized corpus feeding
FlowFabric Core Workflow execution bypass, human gate circumvention, state manipulation, DSL injection
FlowFabric Inbox Unauthorized approval, validation forgery, audit log tampering
KFabric connector Unauthorized corpus write, provenance spoofing, RAG poisoning
API / AuthN-AuthZ Authentication bypass, privilege escalation, token leakage, RBAC misconfiguration
Data governance enforcement Any mechanism that allows collection beyond the declared policy (e.g., ingesting email body content when policy excludes it)

Out of Scope

  • Vulnerabilities in third-party dependencies not introduced by FlowFabric (report directly to the upstream project)
  • Social engineering or phishing attacks against users
  • Denial-of-service attacks without a demonstrated security impact beyond availability
  • Issues requiring physical access to the endpoint
  • Theoretical attacks without a demonstrated realistic path to exploitation
  • Security issues in unreleased branches or experimental prototypes (unless they are the direct basis of a production component)

Known Architectural Security Constraints

These are documented known limitations, not undisclosed vulnerabilities. They are tracked as open work items.

jean-agent Endpoint Security (High Priority — Pre-Production Blocker)

jean-agent is a workstation-level process. The following properties are planned but not yet implemented:

Property Status Notes
Binary code signing Planned — Phase 2 Required before any enterprise deployment
MDM / EDR compatibility (Jamf, Intune, CrowdStrike, SentinelOne) Planned — Phase 2 Must be validated per target OS
Anti-tamper detection Planned — Phase 2 Agent must halt and alert on binary or IPC tampering
Local audit log (append-only, integrity-protected) Planned — Phase 2 Ground truth for compliance audits
Encrypted offline buffer Planned — Phase 2 Events buffered locally must be encrypted at rest
mTLS / rotating JWT for aggregator authentication Planned — Phase 2 No plain-text credentials on endpoint
Signed update mechanism with rollback Planned — Phase 2 Fleet-wide canary updates

Until these controls are implemented, jean-agent must not be deployed on production endpoints.

Human Gate Integrity

The human gate mechanism (FlowFabric Inbox) is the primary control ensuring no production change is made without human validation. Any vulnerability that allows a workflow to bypass or auto-approve a human gate is treated as critical severity regardless of the affected workflow scope.

Validation Fatigue Detection

The system is designed to detect and alert on patterns consistent with rubber-stamping (e.g., approving >80% of suggestions in <10 seconds over 7 days). This detection must not itself be gameable without triggering an alert.


Data and Privacy Security

FlowFabric handles employee behavioral data and business process data. The following rules apply to any security assessment:

  • Any vulnerability that allows collection beyond the declared policy (e.g., capturing email body content, keystrokes, or screen content) is treated as critical.
  • Any vulnerability that allows observed (unvalidated) data to contaminate the prescribed corpus is treated as high.
  • Any vulnerability that allows bypassing the DPIA-required data boundaries is treated as high.
  • Vulnerabilities enabling re-identification of pseudonymized employee data are treated as critical.

These classifications reflect the legal and organizational risk of the affected data, not just the technical impact.


Disclosure Policy

We follow coordinated disclosure:

  1. Reporter submits privately.
  2. We triage and reproduce.
  3. We develop and test a fix.
  4. We notify the reporter of the fix timeline.
  5. We release the fix.
  6. We publish a security advisory (GitHub Security Advisories) crediting the reporter (if they wish).

We ask reporters to allow us 30 days before public disclosure for non-critical issues, and 7 days for critical issues where active exploitation is suspected.


Severity Classification

We use a simplified four-level classification aligned with the data sensitivity of FlowFabric:

Level Criteria Examples
Critical Direct data breach risk, human gate bypass, collection policy violation, re-identification of employee data jean-agent keylogging bypass, human gate auto-approve, email body ingested contrary to policy
High Corpus integrity violation, authentication bypass, privilege escalation, audit log tampering Observed data contaminating prescribed corpus, RBAC bypass, provenance spoofing
Medium Indirect data exposure, partial policy bypass, information disclosure Aggregator leaking anonymized pattern metadata, API endpoint enumeration
Low Low-impact configuration issues, non-sensitive information disclosure Verbose error messages, non-sensitive header leakage

AI Act and GDPR Compliance Vulnerabilities

FlowFabric is subject to the EU AI Act (Annex III — employment context) and GDPR. We treat the following as security-equivalent issues, not merely compliance issues:

  • Any component that enables emotional inference about employees (prohibited, AI Act Art. 5.1.f since Feb 2, 2025)
  • Any mechanism enabling fully automated HR decisions without human review (GDPR Art. 22)
  • Any unauthorized modification of the legal basis documentation for a processing activity
  • Any bypass of the DPIA requirement gate for high-risk processing

These issues should be reported through the same channel as security vulnerabilities.


FlowFabric — SECURITY.md v0.1.0 — 2026-03-21 This policy applies to the FlowFabric project and all its components.

There aren't any published security advisories