Skip to content

GitHub Security Audit — smartmur (2026-03-02)

Status: ALL 4 agents completed. Full audit across 10 repos.

CRITICAL FINDINGS (fix immediately)

Across ALL repos

  • No branch protection on ANY repo — all 10 repos have unprotected default branches

k3s-cluster (3 critical)

  1. pull_request_target + write permissions in dependabot-automerge.yml — privilege escalation vector
  2. curl -sfL https://get.k3s.io | sh in ansible/playbook-k3s.yml — remote code execution risk
  3. StrictHostKeyChecking=no in scripts/00-create-proxmox-template.sh — MITM attack vector

homelab (13 critical)

  1. Branch protection DISABLED on main + auto-merge Dependabot = zero human review path
  2. pull_request_target + write permissions in dependabot-automerge.yml
  3. Hardcoded credentials in 6 docker-compose files: Keycloak (SUPERsecret), Guacamole (root_pass), Code-Server (password), Gitea (gitea), Frigate (password), CrowdSec (placeholder API key)
  4. chmod 666 /var/run/docker.sock in Docker Swarm scripts — full host compromise
  5. iptables -F + INPUT ACCEPT in swarm scripts — firewall completely disabled
  6. StrictHostKeyChecking no in swarm scripts
  7. curl | sh pattern in K3S deploy script

Forked repos (5 critical)

  1. Script injection in smart-sync.yml (claude-code-tresor + skill-factory) — github.event data interpolated into bash with write permissions
  2. Script injection in pr-into-dev.yml (skill-factory) — github.head_ref directly in bash = shell injection via branch name
  3. Script injection in claude-code-review.yml (skill-factory) — PR titles/labels in shell commands
  4. pull_request_target + write in pr-decline.yml (design-os + agent-os)
  5. Combined contents: write + id-token: write + unpinned actions in tresor/skill-factory

WARNING FINDINGS (fix soon)

Systematic across all repos

  • Zero SHA-pinned GitHub Actions anywhere (all use mutable tags like @v4, @v6)
  • || true suppresses security scan failures in homelab CI
  • Missing SECURITY.md in homelab
  • Missing explicit permissions blocks in CI workflows
  • No concurrency control in most CI workflows
  • id-token: write over-granted in tresor/skill-factory (6+ workflows)
  • CLAUDE_CODE_OAUTH_TOKEN passed to unpinned third-party actions

homelab-specific warnings

  • Grafana running as root (user: "0")
  • Docker socket mounted in Authentik, Komodo, Telegraf
  • Chrome remote debugging bound to 0.0.0.0:9222 (Hoarder)
  • Monitoring ports (Prometheus 9090, InfluxDB 8086, Loki 3100) exposed without auth
  • No set -euo pipefail in shell scripts
  • Unquoted variables throughout shell scripts
  • Hardcoded IPs everywhere

k3s-cluster warnings

  • Hardcoded dev path /Users/dre/Desktop/LABing/Projects/Notify
  • Kubeconfig written with mode 644 (should be 600)
  • Hardcoded IPs in scripts/ansible

POSITIVE FINDINGS

k3s-cluster (above average security)

  • Comprehensive .gitignore (one of the best seen)
  • Custom security_scrub.py pre-commit hook
  • detect-private-key pre-commit hook
  • Dependabot for github-actions + terraform
  • Branch protection enabled with required status checks
  • No secrets exposed in repo
  • All shell scripts use set -euo pipefail

homelab (mixed)

  • Strong .gitignore, gitleaks config, pre-commit hooks
  • .env.example templates for some services
  • Some services use proper env var substitution with ${VAR:?required}
  • no-new-privileges:true in several containers

Forked repos

  • No committed secrets in any fork
  • .env.example files all contain placeholders only

PRIORITY FIX ORDER

P0 — Today

  1. Enable branch protection on ALL 10 repos (most impactful single fix)
  2. Fix script injection in smart-sync.yml, pr-into-dev.yml, claude-code-review.yml
  3. Remove chmod 666 /var/run/docker.sock from homelab swarm scripts
  4. Remove iptables -F + INPUT ACCEPT from swarm scripts

P1 — This week

  1. Pin all GitHub Actions to SHA hashes across all repos
  2. Replace pull_request_target with pull_request or add proper guards
  3. Externalize hardcoded credentials in homelab docker-compose files to .env
  4. Fix StrictHostKeyChecking=no → accept-new everywhere
  5. Remove || true from security scan steps
  6. Add permissions: contents: read to all CI workflows

P2 — Next week

  1. Add SECURITY.md to homelab
  2. Expand Dependabot to cover Docker images
  3. Add shellcheck to CI for shell-based repos
  4. Restrict exposed monitoring ports to localhost
  5. Quote all shell variables, add set -euo pipefail

dotfiles (Grade: B+)

  • CRITICAL: pull_request_target + write in dependabot-automerge.yml
  • WARNING: Actions pinned by tag not SHA, no permissions block in ci.yml, curl|sh for Homebrew
  • GOOD: Excellent .gitignore, security_scrub.py, SECURITY.md, detect-private-key hook, clean shell configs

home_media (Grade: D+)

  • CRITICAL: No .gitignore file at all — nothing prevents .env commits
  • CRITICAL: No CI/CD workflows — zero automated testing/scanning
  • CRITICAL: Docker socket mounted in tsdproxy + dockhand (host root equiv)
  • CRITICAL: Termix (web terminal) + Dockhand (Docker UI) have zero authentication
  • WARNING: All images use :latest tags, no health checks, no resource limits, ports bound to 0.0.0.0
  • GOOD: Tailscale-only architecture, secrets externalized to .env, read-only media mounts

FIX STATUS (all pushed)

Repo Commit Branch Fixes Applied
home_media db1d631 master .gitignore, CI workflows, dependabot
homelab 490eddc main docker.sock chmod removed, iptables -F removed, SSH StrictHostKeyChecking→accept-new, CI hardened, SECURITY.md added
k3s-cluster cb402f4 main Actions SHA-pinned, StrictHostKeyChecking→accept-new, CI permissions+concurrency
dotfiles c942ebd main Automerge fetch-metadata guard, Actions SHA-pinned, CI permissions
design-os 6fbfdf1 main pull_request_target→pull_request, input validation, Actions SHA-pinned
agent-os ab75fa0 main pull_request_target→pull_request, input validation, Actions SHA-pinned
claude-code-tresor ee6032e main Script injection fixed (5 workflows: smart-sync, claude-code-review, main-branch-guard, release-orchestrator, ci-quality-gate), Actions SHA-pinned (9 workflows)
claude-code-skill-factory d5440d9 dev Script injection fixed (7 workflows: smart-sync, pr-into-dev, claude-code-review, dev-to-main, release-orchestrator, ci-quality-gate, ci-commit-branch-guard), Actions SHA-pinned (17 workflows)

Additional fixes (wave 2)

Repo Commit Fixes
homelab 688198d Externalized creds in 6 compose files (Keycloak, Guacamole, Code-Server, Gitea, Frigate, CrowdSec) + .env.example files
homelab d19e600 Restricted Prometheus/InfluxDB/Loki/Headscale metrics to 127.0.0.1
homelab 7dbfa8b Added Docker image Dependabot
home_media fd7bf00 Expanded Docker Dependabot to cover all compose dirs
k3s-cluster 8e5fa71 Added Docker image Dependabot

Branch protection (wave 3)

  • [x] 10 repos protected: homelab, k3s-cluster, home_media, claude-code-skill-factory, claude-code-tresor, agent-os, design-os, dotfiles, Lighthouse-AI, Smoke
  • [ ] 6 private repos blocked: claude-superpowers, hommie, zabbix_enchancement, vmware-audit, TrainingWheels, MurzDev_env — GitHub Free doesn't support branch protection on private repos (requires Pro)

Audit performed by Claude Code with 4 parallel research agents Fixes applied by 6 parallel fix agents across 2 sessions Completed: 2026-03-02