Please report vulnerabilities privately through GitHub security advisories ("Report a vulnerability" on the repository) or by email to hello@carloscar.com.
During 0.x only the latest minor release receives fixes.
- Payload URLs: claim-check pointers may contain presigned URLs. Consumers only fetch URLs on an
allow-list (S3 endpoints by default, not other hosts under AWS's domains such as load balancers or
EC2 instances;
PayloadFetch.allowed_url_hosts) over HTTPS, or on the configuredendpoint_url(its scheme, host and port), with a size cap and without following redirects, and only for pointers that record the payload's SHA-256, to prevent server-side request forgery through crafted messages. - Payload buckets: pointers with a bucket and key are read with the consumer's credentials.
Grant consumers
s3:GetObject(ands3:DeleteObjectwithdelete_offloaded_payloads=True) on the payload buckets only, or list them inPayloadFetch.allowed_buckets, so that a crafted message cannot have them read or delete other objects. - Integrity: offloaded payloads are verified against the size and SHA-256 recorded in the pointer; SQS MD5 digests of message bodies and attributes are verified.
- Secrets: fanferro never logs message bodies or credentials; presigned URLs are redacted from
HTTP error messages.
Credentials.__repr__hides the secret key. - Queue policies: subscriptions add a statement allowing only the subscribed topic (
aws:SourceArn) to send to the queue. - Untrusted input: message bodies are parsed by the configured codecs; prefer validating codecs (pydantic, msgspec) for messages from other trust domains.