Introduction

This tool is designed to make it easy to signature potentially unique parts of RTF files.

It was written by David Cannings (@edeca) and released by PwC UK under the Apache 2.0 license.

To install, you'll need Python 3 and some basic libraries. These are handled automatically if you install using pip:

$ pip install rtfsig

Then run like:

$ rtfsig -f badfile.rtf -y output.yar

This will scan the file for potentially unique RTF tags, print details to screen and save a Yara rule to output.yar.

Please raise bugs as Github issues, and note this tool is in beta.

Output

Console

Basic output is shown on the console, which can be used to search VirusTotal (try a search like content:rsid7043998).

-> % rtfsig -f 0b06052d3b5954594cf0e28bd9c50d9110eb8fb78cb78c9a99686eb4ba3391df.hostile
INFO:root:Starting to parse file 0b06052d3b5954594cf0e28bd9c50d9110eb8fb78cb78c9a99686eb4ba3391df.hostile
INFO:root:Non-standard RTF magic marker, should be {\rtf1, often a sign of malicious docs
INFO:root:Found an RSID table in this document
INFO:root:Found 1 embedded image(s) with set height/width
INFO:root:Found 2 document information group tags
INFO:root:Interesting strings (higher chance of FP): \rsid7043998, \rsid7476075, insrsid7043998, \rsid10243744, \rsid7604251, insrsid10243744, {\author blue}, rsidroot10243744, \rsid9200135, tblrsid10243744, charrsid10243744, \picw1\pich1\picwgoal1\pichgoal1 , pararsid10243744, \rsid7238080, insrsid7476075, \rsid11666446, insrsid12343406, \rsid12343406, {\operator blue}
INFO:root:Found some unique strings!  Consider using vtgrep or deploying Yara rules

Debug output can be generated using -v which is helpful if you are reporting a bug.

Yara rules

The tool will automatically generate Yara rules if the -y option is passed. Two Yara rules are created, one which should generate low false positives (strict_rule) and one which may have a higher false positive rate (loose_rule).

It is recommended to review strings carefully and to change any of them to a sensible number, for example 3 of them.

An example rule generated from 0b06052d3b5954594cf0e28bd9c50d9110eb8fb78cb78c9a99686eb4ba3391df looks like:

rule loose_rule {
  meta:
    description = "RTF file matching known unique identifiers (higher chance of FP, adjust 'any of them' if required)"
    generated_by = "rtfsig version 0.0.2"

  strings:
    $ = "{\\author blue}" ascii
    $ = "\\rsid7238080" ascii
    $ = "pararsid10243744" ascii
    $ = "insrsid7043998" ascii
    $ = "\\rsid7043998" ascii
    $ = "rsidroot10243744" ascii
    $ = "\\rsid9200135" ascii
    $ = "\\rsid7604251" ascii
    $ = "insrsid7476075" ascii
    $ = "\\rsid10243744" ascii
    $ = "insrsid12343406" ascii
    $ = "{\\operator blue}" ascii
    $ = "insrsid10243744" ascii
    $ = "charrsid10243744" ascii
    $ = "\\rsid11666446" ascii
    $ = "\\rsid12343406" ascii
    $ = "\\picw1\\pich1\\picwgoal1\\pichgoal1 " ascii
    $ = "tblrsid10243744" ascii
    $ = "\\rsid7476075" ascii

  condition:
    uint32be(0) == 0x7b5c7274 and any of them
}

rule strict_rule {
  meta:
    description = "RTF file matching known unique identifiers (lower chance of FP)"
    generated_by = "rtfsig version 0.0.2"

  strings:
    $ = "\\rsid7043998\\rsid7238080\\rsid7476075\\rsid7604251\\rsid9200135\\rsid10243744\\rsid11666446\\rsid12343406" ascii

  condition:
    uint32be(0) == 0x7b5c7274 and any of them
}

Known limitations

• At present, documents containing lots of obfuscation (e.g. comments between control words and their values) may not be parsed correctly. Please raise an issue with sample files for further inspection.

Contributing

To setup a development environment, clone the git repository and run the following inside a virtualenv:

$ pip install -e ".[dev]"

Before submitting a pull request, please check all tests pass and there is 100% coverage of the core module.

This is as simple as running tox and checking the output:

$ tox
.. tool output ..

py37: commands succeeded
congratulations :)

Packaging:

$ python setup.py sdist bdist_wheel 

Check and upload to PyPI, signing with GPG:

$ twine check dist/*
$ twine upload dist/* --sign --identity FCEC8AAA140C74C826592AC357974C5B48A00D9B

Version history

• v0.0.1 (18th October 2019) - Initial version, supports RSID control words and generating Yara rules
• v0.0.2 (23rd October 2019) - Second beta, added support for unique image identifiers and document information
• v0.0.3 (23rd October 2019) - Third beta, added support for picture sizes
• v0.1.0 (19th September 2020) - First public release, packaged as a Python module for PyPI
• v0.1.1 (26th January 2024) - Bumped Jinja2 dependency to a current version
• v0.1.2 (7th January 2025) - Bumped Jinja2 dependency to a current version
• v0.1.3 (7th January 2025) - Tests fixed and integrated with GitHub actions
• v0.1.4 (5th January 2026) - Bumped Jinja2 dependency to a current version