The Agent Skills Directory

[SAFE]: The skill implements best-practice security architecture for data migration. It strictly separates control logic from data movement by ensuring object payloads never enter the model's context, instead using presigned URLs and direct server-side copies to move bytes between endpoints.
[SAFE]: Robust safety gates are defined in the instructions, requiring explicit user approval for any operation that involves minting write-capable bearer URLs, overwriting existing destination keys during collision discovery, or modifying bucket security settings such as Object Lock or retention policies.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential attack surface for indirect prompt injection as it ingests untrusted data from bucket listings and object metadata.
Ingestion points: Object keys and metadata retrieved during inventory and collision discovery phases using s3_list_objects_v2 and s3_head_object (SKILL.md).
Boundary markers: The instructions do not specify explicit delimiters for metadata, but they limit sampled output to 50 rows in the chat interface to reduce exposure.
Capability inventory: Significant capabilities exist including s3_copy_object, s3_get_presigned_url, and s3_complete_multipart_upload (SKILL.md).
Sanitization: The skill explicitly forbids reading object bodies into the model and instructs the agent to only sample and summarize metadata, which significantly reduces the risk of the model following malicious instructions embedded in object data.

b2-migration