crissyfield/troll-aExtracts secrets (passwords, API keys, tokens) from WARC web archives using Gitleaks rules. Supports HTTP, S3, local files, and compressed archives with concurrent processing.
Troll-A is a command line tool for extracting secrets such as passwords, API keys, and tokens from WARC (Web
ARChive) files. Troll-A is an easy-to-use, comprehensive, and fast solution for finding
secrets in web archive.
*.megawarc.warc.zst files).c7g.12xlarge. This can be further improved by
narrowing down the WARC records to process, via the --filter option.Troll-A is distributed as prebuilt binaries, as a Docker image, or in source form.Troll-A is available on Github's container
registry and can be used as follows:
docker run --rm ghcr.io/crissyfield/troll-a [flags] [url]
Troll-A is also available in binary form for macOS and Linux on the
releases page.
[!NOTE] Unlike the Docker image, the prebuilt binaries are compiled using Go's Stdlib regular expressions and are therefore noticeably slower. If native binaries are preferred and performance is crucial, it is recommended to build the binaries from source.
For better performance, it is recommended to build Troll-A from source, as this allows to use the optimized
regular expression engine provided by go-re2. For this to work, the
RE2 dependency must be installed first.
# Install dependencies
brew install re2
# Install with RE2 activated
go install -tags re2_cgo github.com/crissyfield/[email protected]
# Install dependencies
sudo apt install -u build-essential libre2-dev
# Install with RE2 activated
go install -tags re2_cgo github.com/crissyfield/[email protected]
Usage:
troll-a [flags] [url]
This tool allows to extract (potential) secrets such as passwords, API keys, and tokens
from WARC (Web ARChive) files. Extracted information is output as structured text org
JSON, which simplifies further processing of the data.
"url" can be either a regular HTTP or HTTPS reference ("https://domain/path"), an Amazon
S3 reference ("s3://bucket/path"), a file path (either "file:///path" or simply "path"),
or a dash ("-") to read from STDIN. If "url" is omitted data is read from STDIN. If the
input data is compressed with either GZip, BZip2, XZ, or ZStd it is automatically
decompressed. ZStd with a prepended custom dictionary (as used by "*.megawarc.warc.zstd")
is also handled transparently.
This tool uses rules from the Gitleaks project (https://gitleaks.io) to detect secrets.
Flags:
-c, --custom stringArray additional custom rule to apply. Secrets that match the
given regular expression (using RE2 syntax) will also be
reported. Can be specified multiple times.
-e, --enclosed only report secrets that are enclosed within their context
-f, --filter string filter for the target URL of each WARC record. Only WARC
records that match the given regular expression (using RE2
syntax) will be checked for secrets. An empty filter will
match everything.
-h, --help help for troll-a
-j, --jobs uint detect secrets with this many concurrent jobs (default 8)
-s, --json output detected secrets as JSON
-p, --preset rules-preset rules preset to use. This could be one of the following:
all: All known rules will be applied, which can
result in a significant amount of noise for
large data sets.
most: Most of the rules are applied, skipping the
biggest culprits for false positives.
secret: Only rules are applied that are most likely
to result in an actual leak of a secret.
none: No rules at all are applied. This can be used
in combination with custom rules via the
--custom/-c switch.
No other values are allowed. (default secret)
-q, --quiet suppress success message(s)
-r, --retry retry-strategy retry strategy to use. This could be one of the following:
never: This strategy will fail after the first fetch
failure and will not attempt to retry.
constant: This strategy will attempt to retry up to 5
times, with a 5s delay after each attempt.
exponential: This strategy will attempt to retry for 15
minutes, with an exponentially increasing
delay after each attempt.
always: This strategy will attempt to retry forever,
with no delay at all after each attempt.
No other values are allowed. (default never)
-t, --timeout duration fetching timeout (does not apply to files) (default 30m0s)
-v, --version version for troll-a
Common Crawl maintains a free, open repository of web crawl data that can be used by anyone. The Common Crawl corpus contains petabytes of data collected regularly since 2008.
For example, to extract secrets from all of the 3.35 billion pages of the November/December 2023 crawl
(called CC-MAIN-2023-50), you can do this:
# Download the list of all 90.000 WARC paths
curl -sSL -O https://data.commoncrawl.org/crawl-data/CC-MAIN-2023-50/warc.paths.gz
# Iterate through all paths using 64 scanning jobs, output matches as JSON
gzcat warc.paths.gz | \
xargs -I{} -- troll-a -e -s -j64 https://data.commoncrawl.org/{} > secrets.json