Third-Party Software Security Review — SCORPIOX CODE

---

1. Executive Summary

This report presents the findings of a comprehensive 13-agent automated security audit of SCORPIOX CODE, an AI-powered development tool and CLI platform written in pure C. The audit examined supply chain integrity, build provenance, network endpoints, TLS security, credential handling, file I/O, buffer safety, memory safety, command injection, privilege and access control, Windows deployment scope, telemetry, and install/distribution security.

Overall Risk Rating: LOW

The codebase demonstrates strong security engineering discipline across all domains. Out of 124 total findings, zero are critical severity, two are high (both in the install/distribution domain), 13 are medium, 32 are low, and 77 are informational (many of which are positive security controls). The safe-to-unsafe function ratio for buffer operations is an exceptional 3,493:1. All credential fields use empty-string defaults with runtime injection. All telemetry and tracking is disabled by default with no mandatory data collection. The primary area requiring attention is the install/distribution pipeline, which lacks cryptographic code signing and relies on a single distribution origin.

When scoped to Windows workstation deployment (the primary deployment target), findings reduce to 113 total (0 critical, 2 high, 12 medium, 27 low, 72 informational), as 11 findings relate exclusively to Linux-only binaries and container runtime features not present in the Windows distribution.

---

2. Deployment Scope

2.1 Windows Workstation (Primary Deployment Target)

The following binaries are compiled and distributed for Windows:

Additional Windows binaries (63 total): scorpiox-agent, scorpiox-askuserquestion, scorpiox-beam, scorpiox-bmp2png, scorpiox-claudecode-fetchtoken, scorpiox-claudecode-models, scorpiox-claudecode-refreshtoken, scorpiox-codex-fetchtoken, scorpiox-codex-models, scorpiox-codex-refreshtoken, scorpiox-conv, scorpiox-debug, scorpiox-dns, scorpiox-editfile, scorpiox-email, scorpiox-emit-session, scorpiox-frp, scorpiox-gemini-vertex, scorpiox-google-claude, scorpiox-google-fetchtoken, scorpiox-google-gemini, scorpiox-google-models, scorpiox-grep, scorpiox-hook, scorpiox-host, scorpiox-kql, scorpiox-logger, scorpiox-mcp, scorpiox-mirror-git, scorpiox-multiplexer, scorpiox-openai, scorpiox-otp, scorpiox-planmode, scorpiox-printlogs, scorpiox-pwshovertcp, scorpiox-readfile, scorpiox-rewind, scorpiox-sdk, scorpiox-search, scorpiox-server, scorpiox-server-email, scorpiox-server-fetchtoken, scorpiox-server-http2tcp, scorpiox-skills, scorpiox-systemprompt, scorpiox-tasks, scorpiox-tmux, scorpiox-transcript, scorpiox-usage, scorpiox-vault-git, scorpiox-vertex-models, scorpiox-voice.

2.2 Linux-Only Binaries (Not in Windows Distribution)

The following binaries are not compiled on Windows and their findings do not apply to Windows deployments:

2.3 macOS-Only Binaries

2.4 Windows-Filtered Findings

Of 124 total findings, 11 apply exclusively to Linux/macOS-only binaries, leaving 113 findings relevant to Windows deployment. The two HIGH findings (install script code signing) apply equally to all platforms including Windows.

---

3. Changes Since Last Audit

Previous Audit: 2026-04-30 | Commit fe4a27d | Folder output/2026-04-30_fe4a27d Key Changes:

• Total findings decreased by 31 (20% reduction) despite 10,959 more lines of project code
• Two new HIGH findings identified in the install-script audit (code signing gaps — likely present previously but not flagged by the prior audit scope)
• Medium findings decreased from 14 to 13
• Low findings significantly reduced from 44 to 32
• Informational findings reduced from 97 to 77, indicating consolidation of positive controls
• The wpf-launcher agent was removed from this audit cycle (14 → 13 agents)

---

4. Supply Chain & Dependencies

Agent: supply-chain | Risk: LOW | Findings: 0 Critical, 0 High, 0 Medium, 0 Low, 6 Info

The project has a minimal, well-controlled supply chain. Three vendored C libraries are compiled from source with no pre-built binaries:

Key Findings:

• Zero build-time downloads — no FetchContent or ExternalProject in CMake
• No pre-built binaries in source tree
• An npm/Bun dependency (WhatsApp bridge, ~97 transitive packages) exists but is excluded from the Windows release
• Custom minimal Mbed TLS configuration (sx_mbedtls_config.h) reduces attack surface
• Windows release contains only native executables, DLLs, and a CA certificate bundle — zero JavaScript, TypeScript, Python, or Node.js artifacts
• All vendored libraries are at their latest stable versions

---

5. Build System & Code Provenance

Agent: build-provenance | Risk: LOW | Findings: 0 Critical, 0 High, 1 Medium, 3 Low, 4 Info

The build system uses CMake with C11 and demonstrates strong security hardening:

Findings:

---

6. Network Endpoints

Agent: network-endpoints | Risk: LOW | Findings: 0 Critical, 0 High, 2 Medium, 4 Low, 1 Info

The codebase contains 50+ hardcoded network endpoints. All external API endpoints use HTTPS. The majority are intentional product features (AI provider integrations, distribution URLs, search engine connectors).

Findings:

---

7. TLS/SSL Security

Agent: tls-security | Risk: LOW | Findings: 0 Critical, 0 High, 0 Medium, 3 Low, 8 Info

TLS posture is well-architected with a centralized sx_curl_set_tls() helper enforcing certificate verification by default.

Findings:

---

8. Hardcoded Credentials

Agent: credential-hardcode | Risk: LOW | Findings: 0 Critical, 0 High, 0 Medium, 1 Low, 3 Info No hardcoded credentials found. All 13 credential-type fields use empty-string placeholders with runtime injection via environment variables or config files.

---

9. File I/O & Data Handling

Agent: file-io | Risk: LOW | Findings: 0 Critical, 0 High, 1 Medium, 4 Low, 7 Info

The codebase has strong file I/O hygiene: centralized sx_fopen() macro ensures FD_CLOEXEC, sx_mkdir() uses 0700 permissions, mkstemp() for temp files, and chmod 0600 for sensitive config.

Findings:

---

10. Buffer Safety

Agent: buffer-safety | Risk: LOW | Findings: 0 Critical, 0 High, 0 Medium, 3 Low, 7 Info

Buffer safety discipline is exceptional:

Findings:

---

11. Memory Safety

Agent: memory-safety | Risk: LOW | Findings: 0 Critical, 0 High, 1 Medium, 2 Low, 7 Info

No use-after-free or double-free vulnerabilities were confirmed. The project uses sx_strdup() (abort-on-OOM wrapper) and consistent realloc patterns with temporary variables.

Findings:

---

12. Command Injection

Agent: command-injection | Risk: LOW | Findings: 0 Critical, 0 High, 1 Medium, 3 Low, 6 Info

The codebase demonstrates strong command injection awareness with CJ-* tags in comments indicating prior remediation. Critical paths use fork()+execvp() with argv arrays or is_safe_shell_arg() validation.

Findings:

---

13. Privilege & Access Control

Agent: privilege-access | Risk: LOW | Findings: 0 Critical, 0 High, 0 Medium, 3 Low, 11 Info Threat Model Context: The container runtime (scorpiox-unshare) runs trusted first-party agents on a private network — not untrusted workloads from the internet. Severity ratings reflect this deployment model. Findings:

---

14. Windows Deployment Analysis

Agent: windows-deployment | Risk: LOW | Findings: 0 Critical, 0 High, 3 Medium, 3 Low, 6 Info

The Windows build produces 63 executables using MinGW static linking with vendored libraries.

Findings: Windows Build Security Controls:

---

15. Telemetry and Tracking

Agent: telemetry-tracking | Risk: LOW | Findings: 0 Critical, 0 High, 0 Medium, 0 Low, 3 Info Verdict: No tracking by default. All telemetry features are disabled out-of-the-box. No mandatory data collection exists. No network calls at startup. When explicitly enabled, collected data includes: session ID, provider/model, hostname, username, OS/arch, project name, git branch, token counts. Session telemetry additionally includes full conversation content — this should be noted for data governance. Recommended Corporate Configuration (already default):

USAGE_TRACKING=0
EMIT_SESSION_TRACKING=0
EMIT_SESSION_API_URL=
USAGE_API_URL=
UPDATE_URL=
HTTP_RELAY=0

Findings:

---

16. Install Script & Distribution Security

Agent: install-script | Risk: MEDIUM | Findings: 0 Critical, 2 High, 4 Medium, 3 Low, 8 Info

This is the highest-risk area of the audit. The install/distribution system uses curl|bash (Linux/macOS) and iwr|iex (Windows) patterns.

Findings:

---

17. Consolidated Risk Matrix

---

18. Conclusion & Recommendations

Overall Assessment

SCORPIOX CODE demonstrates strong security engineering across its pure-C codebase. The overall risk rating is LOW for Windows workstation deployment.

Strengths:

• Exceptional buffer safety discipline (3,493:1 safe-to-unsafe ratio)
• Zero hardcoded credentials — proper secrets-from-environment pattern
• No telemetry or tracking by default — fully opt-in
• No network calls at startup — privacy-respecting design
• Consistent use of CreateProcessA on Windows (no system() calls)
• Static linking eliminates DLL hijacking vectors
• Centralized TLS helper enforces certificate verification
• Strong compiler hardening (stack protector, FORTIFY_SOURCE, RELRO, NX stack, CFI)
• All vendored libraries at current stable versions
• Clean git provenance — single organizational author

Priority Recommendations:

HIGH PRIORITY — Code Signing: Implement cryptographic code signing for distributed binaries using cosign, minisign, or GPG with an offline-stored private key. This is the most impactful improvement available.

HIGH PRIORITY — Distribution Infrastructure: Separate the signing infrastructure from the distribution server. Implement multi-party approval for releases.

MEDIUM PRIORITY — API Log Redaction: Add a configuration toggle to disable API body logging, and document that api.log may contain sensitive content.

MEDIUM PRIORITY — Memory Leak Fix: Fix the DKIM buffer leak in sxmail_smtp.c to prevent memory exhaustion on busy mail servers.

LOW PRIORITY — Defense-in-Depth: Extend seccomp blocklist and capability dropping in the container runtime for additional hardening (Linux-only).

Risk Acceptance

For Windows workstation deployment, the following findings are recommended for risk acceptance:

• Container runtime findings (P-01 through P-03) — Linux-only, not applicable
• INIT_TOOLS command injection (CJ-01) — Linux-only scorpiox-init
• --privileged mode — intentional power-user feature
• Opportunistic STARTTLS with VERIFY_OPTIONAL — standard SMTP practice
• -ExecutionPolicy Bypass in updater — expected for self-managed tooling

---

19. Appendix: Audit Methodology

Audit Agents

Scope

• Target: clang repository at commit c9d5a5e4758b2c5e7f5d22de4e54cd51e159253c
• Files analyzed: 235 first-party C source files, 276 vendor files (excluded from findings)
• Lines of code: 408,011 total (159,094 project / 248,531 vendor)
• Allocation sites audited: 686
• Network endpoints catalogued: 50+

Threat Model

The audit was conducted with the understanding that:

• The container runtime runs trusted first-party agents on a private network — not untrusted workloads
• --privileged mode is an intentional feature, not a vulnerability
• Hardcoded IPs are configuration choices, not credential leaks
• Severity ratings reflect actual exploitability in the intended deployment context

Tools & Techniques

• Static analysis via automated pattern matching (grep, AST-based)
• Configuration review of embedded defaults and template files
• Binary inventory and cross-platform compilation analysis
• Network endpoint cataloguing from source code
• Git history and provenance analysis (1,295 commits)
• Dependency tree analysis (CMake, npm/Bun)
• Compiler and linker flag verification