Skip to content

Security: oyesanyf/OshoosiClaw

SECURITY.md

Security Policy

Supported Versions

OshoosiClaw releases follow Semantic Versioning (vMAJOR.MINOR.PATCH). Security updates and critical patches are actively maintained for the following branches and release lines:

Version Line Status Security Patch Support
main branch Active Continuous daily patches and vulnerability remediation
v1.2.x (Current) Active Critical vulnerabilities, parser bugs, and mesh security fixes
v1.1.x Maintenance Critical security fixes only
< v1.1.0 End-of-Life Unsupported; upgrade to v1.2.x immediately

Reporting a Vulnerability

OshoosiClaw is a critical endpoint detection and response system. Security bugs — particularly agent bypasses, false-negative evasion, privilege escalation, or consensus poisoning — are treated as critical security emergencies.

Important

Do NOT open public GitHub issues or forum discussions for suspected security vulnerabilities.
Public disclosure exposes production environments before mitigations can be distributed.

Preferred Reporting Channel: GitHub Security Advisories

Submit reports directly to the maintainers through GitHub's encrypted advisory portal:
👉 Open a Private Security Advisory

Alternative Channel: Encrypted Direct Email

If you cannot access GitHub Advisories, email your disclosure to:
📧 security@oshoosiclaw.dev (or maintainer @oyesanyf)
🔑 PGP Public Key: Available on GitHub at https://github.com/oyesanyf.gpg

Please include:

  1. Component / Subsystem: Affected crate, binary, script, or API endpoint.
  2. Environment & Version: Operating system (Windows/Linux), OshoosiClaw version / git commit hash.
  3. Vulnerability Type: Agent bypass, memory corruption, privilege escalation, DLP leak, or consensus spoofing.
  4. Steps to Reproduce & Proof of Concept (PoC): Minimal reproduction code, network trace, or test script.
  5. Impact Assessment: Consequences of exploitation in an enterprise deployment.

Response Timeline & SLA Commitments

The maintenance team commits to the following Service Level Agreements for reported vulnerabilities:

Incident Stage Target Window Action
Initial Acknowledgement Within 48 hours Confirm receipt and establish a private coordination channel.
Triage & Reproducibility Within 5 business days Validate reproducibility and assign initial CVSS v3.1 / v4.0 severity.
Critical Patch Release Within 14 calendar days Release hotfix binary, signed MSI, and GitHub Security Advisory.
High Severity Patch Within 30 calendar days Release patch in the regular release train.
Public Coordinated Disclosure Coordinated Published after users and downstream swarms have patched.

Vulnerability Scope

In-Scope Vulnerabilities (Eligible for Coordinated Disclosure)

Category Description & Examples
Agent Bypass & False Negatives Novel evasion techniques that bypass Ring-0/ETW sensors, SiQuery telemetry, or decision models without detection.
Privilege Escalation Escaping AppContainer / CanaryCage sandbox, Win32 Job Object confinement, or Linux namespace barriers.
Mesh & Consensus Poisoning Forging ML-DSA-65 signatures, byzantine peer equivocation, or corrupting Merkle audit ledgers.
Data Loss Prevention (DLP) Exfiltration Techniques that evade RedactX-v3 OpenJEV hybrid logit scoring, secret unwrappers, or crown-jewel Merkle chunks.
AI Agentic Exploitation (ASI01–ASI10) Indirect prompt injections escaping agent boundary, unauthorized tool invocation, or memory trajectory hijacking.
Cryptographic Flaws Broken lattice implementations in ML-KEM-768 / ML-DSA-65, TFHE ciphertext leakage, or weak RNG seeds.
Supply Chain & Dependencies Exploitable CVEs in workspace crates identified via cargo audit.

Out-of-Scope

  • Attacks requiring physical device theft with unencrypted bitlocker and compromised physical TPM.
  • Cosmetic styling issues in the WebUI that do not impact security state.
  • Bugs in upstream, third-party standalone tools (e.g. Sysmon, Hayabusa) without an OshoosiClaw integration vulnerability.
  • Denial of Service attacks originating from an attacker who already possesses root/SYSTEM privileges on the target host.

Security Architecture & Invariant Principles

  • Zero Shell Hot-Paths: Telemetry, inspection, and enforcement paths execute exclusively via native Win32/ETW/eBPF APIs without spawning external shells (cmd.exe, powershell.exe, /bin/sh).
  • Multi-Voter Consensus: High-impact containment (Isolate, Tarpit) requires multi-voter consensus (detector_count >= 2, confidence $\ge 0.90$) and respects human-in-the-loop gates.
  • Fail-Closed Isolation: Boundary violations in AppContainer profiles or Job Objects immediately fail closed and terminate child handles.
  • Air-Gapped by Default: External WAN relays (Nostr/DuckDNS) and remote LLMs are strictly opt-in; all threat analysis operates fully local.
  • Signed Authenticode & SLSA Attestations: All Windows executables, libraries, and MSI packages are Authenticode signed with native signtool.exe (SHA-256 + RFC 3161 timestamp) and verified via SLSA v1.0 DSSE attestations.
  • Automated WDSI Submission: All release artifacts are submitted to Microsoft Security Intelligence.

Security Hall of Fame & Acknowledgments

We believe in recognizing security researchers who responsibly report vulnerabilities. Contributors will be credited in our GitHub Security Advisories, release notes, and our public security acknowledgments page (unless anonymity is requested).

There aren't any published security advisories