The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests metadata and object listings from external B2 and S3 storage environments via tools like b2_list_buckets, b2_list_keys, and s3_list_objects_v2. While these external sources could theoretically contain malicious instructions (e.g., in bucket names or metadata), the skill effectively mitigates this risk by requiring explicit human-in-the-loop confirmation for every write or delete capability, such as b2_delete_key or s3_delete_object.
[SAFE]: The skill includes robust "Safety gates" and "Byte path" instructions that prevent the AI from printing sensitive secrets, tokens, or object contents into the chat interface. It ensures that data movement happens directly between the client and B2 using presigned URLs rather than routing through the model context.

b2-incident-response