Datadog SECL is a custom domain-specific language used to create Agent expressions and policies within Datadog Workload Protection. SECL allows security teams to define real-time threat detection rules by specifying conditions, operators, and patterns that security agents can monitor across hosts, containers, applications, and cloud infrastructure.

How SECL rules fit together

Think of SECL as a local filter: it runs inside the Agent on each host, watching kernel and OS events. When an event matches your SECL expression, the Agent raises a detection.

Datadog threat detection rules act as backend logic: they combine one or more Agent rules (using @agent.rule_id), add thresholds, suppress noise, and decide how alerts are routed.

In summary, the Agent rule finds raw behavior and the detection rule turns it into end-to-end attack stories.

SECL expression syntax

The standard format of a SECL expression is:

<event-type>.<event-attribute> <operator> <value> [<operator> <event-type>.<event-attribute>] ...

Using this format, an example rule for a Linux system looks like this:

open.file.path == "/etc/shadow" && process.file.path not in ["/usr/sbin/vipw"]

Operators

SECL operators are used to combine event attributes together into a full expression. The following operators are available:

Patterns and regular expressions

Patterns or regular expressions can be used in SECL expressions. They can be used with the in, not in, =~, and !~ operators.

Patterns on .path fields are used as Glob. * matches files and folders at the same level. **, introduced in 7.34, can be used at the end of a path to match all the files and subfolders.

Durations

You can use SECL to write rules based on durations, which trigger on events that occur during a specific time period. For example, trigger on an event where a secret file is accessed more than a certain length of time after a process is created. Such a rule could be written as follows:

open.file.path == "/etc/secret" && process.file.name == "java" && process.created_at > 5s

Durations are numbers with a unit suffix. The supported suffixes are “s”, “m”, “h”.

Platform specific syntax

SECL expressions support several platforms. You can use the documentation below to see what attributes and helpers are available for each.

• Linux
• Windows

Rule authoring tips

• Always set the operating system (OS).
• Anchor on ancestry to reduce noise. Use process.ancestors.file.name.
• Use durations (for example, > 5s, 10m, 2h) to target narrow execution windows.
• Use exact match (==) whenever possible as it results in the lowest noise.
• List membership (in [...]) is best for allowlists or controlled sets of values.
• Use a glob match (~"/path/*") for path families as it is safer and faster than regex.
• Use regex (=~) only when globs/lists can’t be used. Keep the regex expression as narrow as possible. As a rule of thumb, start with == or in [...]. Reach for regex only as a last resort.
• Use negation (not in [...], !~) to carve out exceptions explicitly (for example, trusted tools).
• Use CIDR operators (in CIDR, not in CIDR) for network boundaries.
• Name rules by behavior, with a format that follows What + Who + Context.
• Tag generously: team, app, env, MITRE, severity.

Avoid common mistakes

Example library

In Agent policy files, each rule includes an id and an expression. You can also add optional actions. See Variables and actions for more.

Linux

Access to sensitive files (allowlist safe tools)

rules:
  - id: access_sensitive_files
    expression: >-
      open.file.path in ["/etc/shadow", "/etc/sudoers"] &&
      process.file.path not in ["/usr/sbin/vipw", "/usr/sbin/visudo"]
    filters:
      - os == "linux"

NGINX or PHP spawning bash

rules:
  - id: nginx_php_spawn_bash
    expression: >-
      exec.file.path == "/usr/bin/bash" &&
      (
        process.ancestors.file.name == "nginx" ||
        process.ancestors.file.name =~ "php*"
      )
    filters:
      - os == "linux"

Suspicious IMDS access from container

rules:
  - id: suspicious_imds_access
    expression: >-
      connect &&
      network.destination.ip in ["169.254.169.254"] &&
      container.id != ""
    filters:
      - os == "linux"

Kernel module loads outside maintenance window

rules:
  - id: kernel_module_load
    expression: >-
      load_module &&
      process.user != "root" &&
      process.ancestors.file.name not in ["modprobe", "insmod"]
    filters:
      - os == "linux"

Sensitive file read shortly after start

rules:
  - id: sensitive_file_read_after_start
    expression: >-
      open.file.path == "/etc/secret" &&
      process.file.name == "java" &&
      process.created_at > 5s
    filters:
      - os == "linux"

Outbound to non-corporate IPs (CIDR allowlist)

rules:
  - id: outbound_non_corporate_ips
    expression: >-
      connect &&
      network.destination.ip not in [10.0.0.0/8, 192.168.0.0/16, 172.16.0.0/12]
    filters:
      - os == "linux"

Windows

Registry persistence through a run key

rules:
  - id: registry_run_key_persistence
    expression: >-
      set_key_value &&
      open_key.registry.key_path =~ "*\\Software\\Microsoft\\Windows\\CurrentVersion\\Run*"
    filters:
      - os == "windows"

Unsigned binary launching PowerShell

rules:
  - id: unsigned_binary_powershell
    expression: >-
      exec.file.path =~ "*\\WindowsPowerShell\\v1.0\\powershell.exe" &&
      process.parent.file.path !~ "*\\Program Files*" &&
      process.user_sid != "S-1-5-18"
    filters:
      - os == "windows"

Cross-platform

Crypto-miner indicators

rules:
  - id: crypto_miner_indicators
    expression: >-
      exec.args_flags in ["cpu-priority", "donate-level", ~"randomx-1gb-pages"] ||
      exec.args in [~"*stratum+tcp*", ~"*nicehash*", ~"*yespower*"]