title: Ad-Hoc Non-Loopback HTTP File Server Launched in Linux Home or AI Agent Root id: 6c8f3b21-4d92-4a1e-9b75-10a100000001 status: experimental description: Detects ad-hoc static HTTP file servers (python http.server, npx serve, http-server, miniserve, php -S) started in /root or /home without explicit loopback binding, which exposes hidden .claude, .openclaw, .codex, .cursor, and .hermes session directories. Requires a Linux process_creation source that populates CurrentDirectory alongside CommandLine (such as Sysmon for Linux EventID 1 or Linux auditd EXECVE and CWD records correlated by shared audit event ID via Laurel/Auditbeat), or an explicit home/dotfile directory argument on CommandLine. author: Odin by Brandefense date: 2026/11/18 references: - https://docs.python.org/3/library/http.server.html - https://docs.redhat.com/en/documentation/red_hat_enterprise_linux/9/pdf/security_hardening/auditing-the-system_security-hardening - https://attack.mitre.org/techniques/T1552/001/ tags: - attack.credential_access - attack.t1552.001 - attack.collection - attack.t1213 logsource: category: process_creation product: linux detection: selection_cli: CommandLine|contains: - ' -m http.server' - 'SimpleHTTPServer' - 'http-server' - 'npx serve' - 'miniserve' - ' -S 0.0.0.0:' selection_workdir: CurrentDirectory|startswith: - '/root' - '/home/' selection_explicit_dir_arg: CommandLine|contains: - ' /root' - ' /home/' - '/.claude' - '/.openclaw' - '/.codex' - '/.cursor' - '/.hermes' condition: selection_cli and (selection_workdir or selection_explicit_dir_arg) falsepositives: - Intentional air-gapped lab file transfers restricted by host firewall rules (enforce explicit --bind 127.0.0.1). level: high --- title: Unauthenticated Web Server Access to Exposed AI Agent Session and Credential Dotfiles id: 6c8f3b21-4d92-4a1e-9b75-10a100000002 status: experimental description: Detects HTTP 200 access to exposed Claude Code, OpenClaw, OpenAI Codex, Cursor IDE, and Hermes Agent session transcripts, subagent compaction logs, and credential stores while excluding bundled Hermes OSS skill documentation. author: Odin by Brandefense date: 2026/11/18 references: - https://docs.anthropic.com/en/docs/claude-code/overview - https://attack.mitre.org/techniques/T1552/001/ tags: - attack.credential_access - attack.t1552.001 - attack.collection - attack.t1213 logsource: category: webserver detection: selection_status: sc-status: 200 selection_uris: cs-uri-stem|contains: - '/.claude/.credentials.json' - '/.claude/projects/' - 'subagents/agent-acompact-' - 'subagents/agent-aside_question-' - '/.openclaw/agents/' - '/.openclaw/identity/device.json' - '/.codex/sessions/' - '/.codex/auth.json' - '/.cursor/mcp.json' - '/agent-transcripts/' - '/.hermes/sessions/' - '/.hermes/state-snapshots/' - '/.hermes/profiles/' condition: selection_status and selection_uris falsepositives: - Authorized forensic artifact mirrors gated behind reverse-proxy mTLS authentication. level: critical --- title: Autonomous AI Agent Runtime Spawning Cloud IMDS Probe or Python Socket Reverse Shell id: 6c8f3b21-4d92-4a1e-9b75-10a100000003 status: experimental description: Detects AI coding or gateway runtimes (OpenClaw, Claude Code, Codex CLI, Hermes) spawning inline Python socket reverse shells, headless Chrome CDP port 9222 enumeration, or cloud IMDS credential queries observed in exposed trajectories. author: Odin by Brandefense date: 2026/11/18 references: - https://attack.mitre.org/techniques/T1059/006/ - https://attack.mitre.org/techniques/T1552/005/ tags: - attack.execution - attack.t1059.006 - attack.credential_access - attack.t1552.005 logsource: category: process_creation product: linux detection: selection_parent: ParentCommandLine|contains: - 'openclaw' - 'claude' - 'codex' - 'hermes' selection_payload: CommandLine|contains: - '169.254.169.254/metadata' - 'import socket' - '127.0.0.1:9222/json/list' - '/dev/tcp/' condition: selection_parent and selection_payload falsepositives: - Authorized cloud security posture validation run inside isolated sandbox containers. level: high