Detection Rules

81 detection rules across 15 categories. All rules run at every tier — no features gated behind paid plans.

Rule IDs are stable identifiers you can use in policy severity_overrides and allowlists.

10Hostname

Homograph attacks, punycode, confusable characters, IDN spoofing

Rule IDSeverityDescription
non_ascii_hostnameCRITICALDetects hostnames containing non-ASCII characters that may be visual spoofs of legitimate domains.
punycode_domainHIGHDetects internationalized domain names encoded as punycode (xn--) that may disguise malicious domains.
mixed_script_hostnameCRITICALDetects hostnames mixing character scripts (Latin + Cyrillic) — a strong indicator of homograph attacks.
confusable_charactersHIGHIdentifies characters that are visually similar to common ASCII letters but are from different Unicode blocks.
idn_homographCRITICALFull internationalized domain name homograph detection using Unicode confusable mappings.
suspicious_tldMEDIUMFlags domains using TLDs commonly associated with abuse or phishing campaigns.
ip_address_hostnameMEDIUMDetects raw IP addresses used instead of domain names — common in C2 and phishing infrastructure.
excess_subdomainsLOWFlags URLs with unusually deep subdomain nesting, often used to obscure the actual domain.
lookalike_domainHIGHDetects domains that closely resemble popular services (typosquatting, character substitution).
newly_registered_domainLOWFlags domains matching patterns common in newly registered or disposable domain infrastructure.

3Path

Non-ASCII paths, homoglyphs, encoded traversal sequences

Rule IDSeverityDescription
non_ascii_pathHIGHDetects non-ASCII characters in URL paths that may be used to evade path-based security filters.
path_homoglyphHIGHIdentifies visually similar Unicode characters in URL paths that may disguise the real file being fetched.
encoded_path_traversalHIGHDetects percent-encoded directory traversal sequences (..%2f, %2e%2e/) in URL paths.

4Transport

Insecure HTTP, TLS downgrades, URL shorteners, data URIs

Rule IDSeverityDescription
insecure_httpMEDIUMFlags plain HTTP connections that transmit data without encryption — vulnerable to MITM attacks.
tls_downgradeHIGHDetects flags that disable TLS verification (-k, --insecure), exposing connections to interception.
url_shortenerMEDIUMDetects shortened URLs (bit.ly, t.co, etc.) that obscure the actual destination.
data_uri_schemeHIGHDetects data: URIs that embed executable content inline, bypassing external URL inspection.

5Terminal

ANSI injection, bidi overrides, zero-width chars, control sequences

Rule IDSeverityDescription
ansi_escape_injectionCRITICALDetects ANSI escape sequences that can manipulate terminal display, hide commands, or execute code.
bidi_overrideCRITICALDetects Unicode bidirectional override characters that reverse text display to hide true content.
zero_width_charactersHIGHDetects invisible Unicode characters (ZWSP, ZWNJ, ZWJ) that can alter command interpretation.
terminal_title_injectionHIGHDetects escape sequences that modify the terminal title bar, which can be used for social engineering.
control_character_injectionHIGHDetects non-printable control characters (except common whitespace) embedded in commands or pasted text.

8Command

Pipe-to-shell, dotfile persistence, archive bombs, code substitution

Rule IDSeverityDescription
pipe_to_interpreterHIGHDetects piping downloaded content directly to a shell interpreter — the classic supply-chain attack vector.
dotfile_modificationHIGHDetects commands targeting shell configuration dotfiles (.bashrc, .zshrc, .profile) for persistence.
archive_extraction_attackMEDIUMDetects suspicious archive extraction patterns (tar, zip) that may overwrite system files.
command_substitutionHIGHDetects backtick or $() command substitution in URLs or arguments that execute code during expansion.
background_executionMEDIUMDetects commands launched in background (&, nohup, disown) that persist beyond the current session.
base64_decode_executeHIGHDetects base64 decode piped to interpreter — through sudo/env wrappers and PowerShell -EncodedCommand.
cloud_metadata_accessHIGHDetects access to cloud instance metadata endpoints (169.254.169.254) used in SSRF and credential theft.
private_network_accessMEDIUMDetects commands targeting private/internal network ranges that may indicate lateral movement.

5Environment

Proxy hijacking and environment variable manipulation

Rule IDSeverityDescription
proxy_environment_variableHIGHDetects modification of HTTP_PROXY/HTTPS_PROXY variables that redirect all traffic through an attacker.
sensitive_env_exportHIGHDetects exporting sensitive environment variables (API keys, tokens) that may leak credentials to child processes.
code_injection_envCRITICALDetects environment variables containing executable code that gets evaluated by interpreters.
interpreter_hijackHIGHDetects PATH or interpreter variable changes that redirect command execution to attacker-controlled binaries.
shell_injection_envHIGHDetects environment variables with shell metacharacters that trigger injection when evaluated.

9Ecosystem

Git, Docker, pip, npm, Web3 — supply-chain attack surface

Rule IDSeverityDescription
git_credential_exposureHIGHDetects git operations that may expose credentials via URL-embedded tokens or credential helpers.
docker_privileged_runHIGHDetects Docker containers run with --privileged or dangerous volume mounts that escape isolation.
pip_install_untrustedMEDIUMDetects pip installing from URLs, git repos, or with --extra-index-url pointing to untrusted sources.
npm_lifecycle_scriptMEDIUMDetects npm packages with suspicious preinstall/postinstall scripts that execute arbitrary code.
web3_wallet_drainCRITICALDetects commands targeting cryptocurrency wallets, private keys, or seed phrases for theft.
supply_chain_registryHIGHDetects package installations from alternate registries or with dependency confusion patterns.
git_clone_typosquatHIGHDetects git clone of repositories with names similar to popular projects — potential typosquatting.
untrusted_docker_registryHIGHDetects pulling Docker images from untrusted or non-standard registries.
vet_not_configuredMEDIUMDetects package manager installations without supply-chain audit tooling (npm audit, pip-audit).

8Config Security

AI config poisoning, prompt injection, MCP server validation

Rule IDSeverityDescription
config_injectionHIGHPrompt injection or skill activation triggers detected in AI config files.
config_suspicious_indicatorsMEDIUMSuspicious patterns (permission bypass, safety dismissal) in AI tool configs.
config_non_asciiMEDIUMNon-ASCII or invisible Unicode characters in AI tool configuration files.
mcp_insecure_serverHIGHMCP server configured with insecure HTTP connection instead of HTTPS.
mcp_untrusted_serverMEDIUMMCP server from untrusted or unknown source.
mcp_duplicate_serverMEDIUMDuplicate MCP server names that could cause conflicts or shadowing.
mcp_permissive_toolsMEDIUMMCP server configured with wildcard tool access.
mcp_shell_metachar_argsHIGHShell metacharacters in MCP server arguments enabling injection.

5Rendered Content

Hidden CSS, color-matched text, comment instructions, PDF hidden text

Rule IDSeverityDescription
css_hidden_contentHIGHContent hidden via CSS (display:none, visibility:hidden, opacity:0, font-size:0).
color_hidden_contentHIGHText hidden via similar foreground/background colors (contrast ratio < 1.5:1).
hidden_html_attributesMEDIUMHidden HTML attributes containing instructions invisible to users but readable by AI.
comment_instructionsMEDIUMHTML/Markdown comments containing instructions targeting AI agents.
pdf_hidden_textHIGHSub-pixel rendered PDF text (font-size < 1px) invisible to readers but parseable by LLMs.

2Cloaking

Server-side cloaking detection — different content for bots vs browsers

Rule IDSeverityDescription
server_cloakingHIGHServer returns different content to AI bots vs browsers.
clipboard_hidden_contentHIGHClipboard contains hidden content not visible in the UI.

1Policy

Custom blocklists and organizational security policies

Rule IDSeverityDescription
domain_blocklistCRITICALThe target domain matches a known-malicious entry in the configured blocklist.

10Credential Detection

AWS keys, GitHub PATs, Stripe/Slack/SendGrid tokens, private keys, entropy-based secrets

Rule IDSeverityDescription
aws_key_detectionCRITICALDetects AWS access key IDs and secret access keys in commands or arguments.
github_pat_detectionCRITICALDetects GitHub personal access tokens (ghp_, github_pat_) in commands.
stripe_token_detectionCRITICALDetects Stripe API keys (sk_live_, sk_test_) in commands or environment.
slack_token_detectionHIGHDetects Slack bot tokens (xoxb-), user tokens (xoxp-), and webhook URLs.
private_key_blockCRITICALDetects PEM-encoded private key blocks (RSA, EC, ED25519) being transmitted or exposed.
entropy_based_secretMEDIUMDetects high-entropy strings likely to be secrets or tokens using Shannon entropy analysis.
sendgrid_token_detectionHIGHDetects SendGrid API keys (SG.) in commands or environment variables.
anthropic_api_keyHIGHDetects Anthropic API keys (sk-ant-) in commands or environment variables.
gcp_service_accountHIGHDetects GCP service account JSON key files being exposed or transmitted.
npm_token_detectionHIGHDetects npm authentication tokens (npm_) in commands or .npmrc references.

5Data Exfiltration

curl/wget uploads, env var leaks, command substitution sending data to external hosts

Rule IDSeverityDescription
curl_data_uploadHIGHDetects curl uploading sensitive file contents via -d @file to remote servers.
curl_file_uploadHIGHDetects curl uploading files via -T/--upload-file to remote servers.
wget_post_fileHIGHDetects wget --post-file uploading local file contents to remote servers.
env_var_exfiltrationHIGHDetects sending sensitive env vars ($AWS_SECRET_ACCESS_KEY, etc.) to external hosts.
command_substitution_exfilHIGHDetects command output (whoami, hostname) exfiltrated via URL params or POST bodies.

3Code Scanning

Obfuscated payloads, dynamic code execution, secret exfiltration in JS/Python files

Rule IDSeverityDescription
dynamic_code_executionMEDIUMDetects dangerous function calls near b64decode or atob() patterns in scanned JS/Python files.
obfuscated_payloadMEDIUMDetects long base64-encoded strings decoded and executed in scanned files.
suspicious_code_exfiltrationMEDIUMDetects HTTP calls (fetch, requests.post) passing sensitive data in scanned files.

3Post-Compromise

Process memory scraping, Docker escalation, credential file sweeps — TeamPCP-inspired

Rule IDSeverityDescription
proc_memory_scrapingCRITICALDetects access to /proc/*/mem used to scrape secrets from running process memory.
docker_remote_escalationCRITICALDetects Docker socket access or remote API calls used to escalate host privileges.
credential_file_sweepHIGHDetects enumeration of known credential file paths (.aws/, .ssh/, .gnupg/) inspired by TeamPCP attack.