Skip to main content

    Extract mode

    Deprecation notice

    As of Semgrep 1.65.0, extract mode has been deprecated and removed from Semgrep. This feature may return in the future.

    Extract mode enables you to run existing rules on subsections of files where the rule language is different than the language of the file. For example, running a JavaScript rule on code contained inside of script tags in an HTML document.

    Example of extract mode

    Without extract mode, writing rules to validate template, Markdown or configuration files which contain code in another language can be burdensome and require significant rule duplication.

    Let's take the following Bash rule as an example (a simplified version of the curl-eval rule from the Semgrep Registry):

    - id: curl-eval
    severity: WARNING
    - bash
    message: Evaluating data from a `curl` command is unsafe.
    mode: taint
    - pattern: |
    $(curl ...)
    - pattern: |
    `curl ...`
    - pattern: eval ...

    Usually, Semgrep uses this rule only against Bash files. However, a project might contain Dockerfiles or Python scripts that invoke Bash commands—without an extract mode rule, Semgrep does not run any Bash rules against commands contained in files of different languages.

    However, with extract mode, you can provide Semgrep with instructions on how to extract any Bash commands used in a Docker RUN instruction or as an argument to Python's os.system standard library function.

    - id: extract-docker-run-to-bash
    mode: extract
    - dockerfile
    pattern: RUN $...CMD
    extract: $...CMD
    dest-language: bash
    - id: extract-python-os-system-to-bash
    mode: extract
    - python
    pattern: os.system("$CMD")
    extract: $CMD
    dest-language: bash

    By adding the extract mode rules as shown in the previous code snippet, Semgrep matches Bash code contained in the following Python file and reports the contained Bash as matching against the curl-eval rule.

    from os import system

    if system('eval `curl -s ""`'):
    print("Command failed!")

    Likewise, if a query included a Dockerfile with an equivalent Bash command, Semgrep reports the contained Bash as matching against the curl-eval rule. See the following Dockerfile example that contains a Bash command:

    FROM fedora
    RUN dnf install -y unzip zip curl which
    RUN eval `curl -s ""`

    Extract mode rule schema

    Extract mode rules require the following usual Semgrep rule keys:

    • id
    • languages
    • One of pattern, patterns, pattern-either, or pattern-regex

    Extract mode rules also require two additional fields:

    • extract
    • dest-language

    Extract mode has two optional fields:

    • reduce
    • json

    The fields specific to extract mode are further explained in the sections below.


    The extract key is required in extract mode. The value must be a metavariable appearing in your pattern(s). Semgrep uses the code bound to the metavariable for subsequent queries of non-extract mode rules targeting dest-language.


    The dest-language key is required in extract mode. The value must be a language tag.


    The transform is an optional key in the extract mode. The value of this key specifies whether the extracted content is parsed as raw source code or as a JSON array.

    The value of transform key must be one of the following:


    Extract the matched content as raw source code. This is the default value.


    Extract the matched content as a JSON array. Each element of the array correspond to a line the resulting source code. This value is useful in extracting code from JSON formats such as Jupyter Notebooks.


    The reduce key is optional in extract mode. The value of this key specifies a method to combine the ranges extracted by a single rule within a file.

    The value of reduce key must be one of the following:


    Treat all matched ranges as separate units for subsequent queries. This is the default value.


    Concatenate all matched ranges together and treat this result as a single unit for subsequent queries.

    Limitations of extract mode

    Although extract mode supports JSON array decoding with the json key, it does not support other additional processing for the extracted text, such as unescaping strings.

    While extract mode can help to enable rules which try and track taint across a language boundary within a file, taint rules cannot have a source and sink split across the original file and extracted text.

    Not finding what you need in this doc? Ask questions in our Community Slack group, or see Support for other ways to get help.