The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted microscopy metadata, annotations, and filenames from remote OMERO servers, creating a potential surface for indirect prompt injection.
Ingestion points: Data is ingested from remote hosts via the omero-py BlitzGateway and CLI tools, specifically within scripts/inventory.py and scripts/export_image_metadata.py.
Boundary markers: Documentation in references/metadata.md explicitly notes that remote filenames are untrusted. SKILL.md mandates an operating contract that requires bounded requests and explicit data scopes.
Capability inventory: The skill possesses capabilities for local file system writes (via atomic_write_json in scripts/omero_common.py) and execution of OMERO CLI commands.
Sanitization: The skill implements a json_safe utility to enforce maximum lengths for strings and collections, and uses scrubbed_error to ensure credential values are not exposed in logs or exception messages.

omero-integration