muchdogesec/txt2detection

# txt2detection [](https://codecov.io/gh/muchdogesec/txt2detection) ## Overview  A command line tool that takes a txt file containing threat intelligence and turns it into a detection rule. ## The problems To illustrate the problem, lets walk through the current status quo process a human goes through when going from idea (threat TTP) to detection rule: 1. read and understand threat using their own research, aided by external sources (blogs, intel feed, etc.) * problems: lots of reports, threats described in a range of ways, reports contain differing data 2. understand what logs or security data can be used to detect this threat * problems: log schemas are unknown to analyst, TTPs often span many logs making it hard to ensure your detection rule has full coverage 3. convert the logic created in step 1 into a Sigma detection rule to search logs identified at step 2 * problems: hard to convert what has been understood into a logical detection rule (in a detection language an analyst might not be familiar with) 4. modify the detection rule based on new intelligence as it is discovered * problems: this is typically overlooked as people create and forget about rules in their detection tools ## The solution Use AI to process threat intelligence, create and keep them updated. txt2detection allows a user to enter some threat intelligence as a file to considered be turned into a detection. ## Usage ### Setup Install the required dependencies using: # clone the latest code git clone https://github.com/muchdogesec/txt2detection cd txt2detection # create a venv python3 -m venv txt2detection-venv source txt2detection-venv/bin/activate # install requirements pip3 install -r requirements.txt pip3 install . ### Set variables txt2detection has various settings that are defined in an `.env` file. To create a template for the file: cp .env.example .env To see more information about how to set the variables, and what they do, read the `.env.markdown` file. Then test your configoration python3 txt2detection.py \ check-credentials It will return a response to show what API keys are working ============= Service Statuses =============== ctibutler : authorized ✔ vulmatch : authorized ✔ LLMS: openai : authorized ✔ deepseek : unsupported – gemini : unsupported – openrouter : unsupported – anthropic : unsupported – Not all services need to be configured, if you have no intention of using them. ### Run python3 txt2detection.py MODE \ ARGUEMENTS There are 3 modes in which you can use txt2detection: * `file`: A text file, usually a threat report you want to create rules from the intel held within * `text`: A text prompt that describes the rule you want to create * `sigma`: An existing Sigma Rule you want to convert into a STIX bundle #### File (`file`) / Text Input (`text`) Use this mode to generate a set of rules from an input text file; * `--input_file` (required, if not using `--input_text`, file path): the file to be converted. Must be `.txt` * `--input_text` (required, if not using `--input_file`, string): a text string that will be analysed to create a rule by the AI if you don't want to use a file. Generally you want to be as descritive as possible with this text (consider it similar to an LLM propmpt). Very short `input_text`s will often cause the AI to fail at rule generation (you will see `Unparsable output returned by LLM model`). * `--name` (required): name of file, max 72 chars. Will be used in the STIX Report Object created. Note, the Indicator object names/titles are generated by AI * `--report_id` (optional, default random uuidv4): Sometimes it is required to control the id of the `report` object generated. You can therefore pass a valid UUIDv4 in this field to be assigned to the report. e.g. passing `2611965-930e-43db-8b95-30a1e119d7e2` would create a STIX object id `report--2611965-930e-43db-8b95-30a1e119d7e2`. If this argument is not passed, the UUID will be randomly generated. * `--tlp_level` (optional, default `clear`): Options are `clear`, `green`, `amber`, `amber_strict`, `red`. * `--labels` (optional): whitspace separated list of labels. Case-insensitive (will all be converted to lower-case). Allowed `a-z`, `0-9`. Must use a namespaces (`NAMESPACE.TAG_VALUE`). e.g.`"namespace.label1" "namespace.label_2"` would create 2 labels. Added to both report and indicator objects created and the rule `tags`. * note: you can use reserved namespaces `cve.` and `attack.` when creating labels to perform external enrichment using Vulmatch and CTI Butler. All Indicators will be linked to these objects (AI enrichments link individual rules). Created tags will be appended to the list of AI generated tags. * note: you cannot use the namespace `tlp.` Use the `--tlp_level` flag instead. * `--created` (optional, `YYYY-MM-DDTHH:MM:SS`): by default all object `created` times will take the time the script was run. If you want to explicitly set these times you can do so using this flag. Pass the value in the format `YYYY-MM-DDTHH:MM:SS` e.g. `2020-01-01T00:00:00` * `--use_identity` (optional, default txt2detection identity): can pass a full STIX 2.1 identity object (make sure to properly escape). Will be validated by the STIX2 library. The ID is used to create the Indicator and Report STIX objects, and is used as the `author` property in the Sigma Rule. * `--license` (optional): [License of the rule according the SPDX ID specification](https://spdx.org/licenses/). Will be added to the rule. * `--reference_urls` (optional): A list of URLs to be added as `references` in the Sigma Rule property and in the `external_references` property of the Indicator and Report STIX object created. e.g `"https://www.google.com/" "https://www.facebook.com/"` * `--external_refs` (optional): txt2detection will automatically populate the `external_references` of the report object it creates for the input. You can use this value to add additional objects to `external_references`. Note, you can only add `source_name` and `external_id` values currently. Pass as `source_name=external_id`. e.g. `--external_refs txt2stix=demo1 source=id` would create the following objects under the `external_references` property: `{"source_name":"txt2stix","external_id":"demo1"},{"source_name":"source","external_id":"id"}` * `--ai_provider` (required): defines the `provider:model` to be used to generate the rule. Select one option. Currently supports: * Provider (env var required `OPENROUTER_API_KEY`): `openrouter:`, providers/models `openai/gpt-4o`, `deepseek/deepseek-chat` ([More here](https://openrouter.ai/models)) * Provider (env var required `OPENAI_API_KEY`): `openai:`, models e.g.: `gpt-4o`, `gpt-4o-mini`, `gpt-4-turbo`, `gpt-4` ([More here](https://platform.openai.com/docs/models)) * Provider (env var required `ANTHROPIC_API_KEY`): `anthropic:`, models e.g.: `claude-3-5-sonnet-latest`, `claude-3-5-haiku-latest`, `claude-3-opus-latest` ([More here](https://docs.anthropic.com/en/docs/about-claude/models)) * Provider (env var required `GOOGLE_API_KEY`): `gemini:models/`, models: `gemini-1.5-pro-latest`, `gemini-1.5-flash-latest` ([More here](https://ai.google.dev/gemini-api/docs/models/gemini)) * Provider (env var required `DEEPSEEK_API_KEY`): `deepseek:`, models `deepseek-chat` ([More here](https://api-docs.deepseek.com/quick_start/pricing)) * `--create_attack_navigator_layer` (boolean, default `false`): passing this flag will generate a [MITRE ATT&CK Navigator layer](https://mitre-attack.github.io/attack-navigator/) for MITRE ATT&CK tags. Note, Sigma currently supports ATT&CK Enterprise only. Note, in this mode, the following values will be automatically assigned to the rule * `level`: the AI will be prompted to assign, either `informational`, `low`, `medium`, `high`, `critical` * `status`: will always be `experimental` in this mode #### Sigma rule input (`sigma`) Use this mode to turn a Sigma Rule into a STIX bundle and get it enriched with ATT&CK and Vulmatch. Note, in this mode you should be aware of a few things; ### A note on observable extraction txt2detection will automatically attempt to extract any observables (aka indicators of compromise) that are found in the created or imported rules to turn them into STIX objects joined to the STIX Indicator object of the Rule. In `txt2detection/observables.py` you will find the observable types (and regexs used detection) currently supported. ### Output The output of each run is structured as follows; . ├── logs │ ├── log-.log │ ├── log-.log │ └── log-.log └── output └── bundle-- ├── rules │ ├── rule--.yml │ └── rule--.yml ├── data.json # AI output, useful for debugging └── bundle.json # final STIX bundle with all objects ## Examples See `tests/manual-tests/README.md` for some example commands. ## License [Apache 2.0](/LICENSE).