Skills
Navigate
skills/k-dense-ai/scientific-agent-skills/deepspot-m/Gen Agent Trust Hub

deepspot-m

Warn

Audited by Gen Agent Trust Hub on Oct 24, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Downloads the deepspotm and histolab packages from PyPI and fetches gated model weights and configuration from Hugging Face (ratschlab/DeepSpotM).\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data in the form of histology slide images (SVS, PNG) and coordinate manifests (CSV), creating an attack surface.\n
  • Ingestion points: Loads slide files via openslide and Pillow, and manifest CSVs via pandas (found in references/whole_slide.md).\n
  • Boundary markers: None used to delimit untrusted data from instructions.\n
  • Capability inventory: Authorized for Bash, Write, and Edit tool usage.\n
  • Sanitization: Employs regex for filename validation and verifies spatial bounds and metadata.\n- [METADATA_POISONING]: Metadata includes a last-reviewed date of '2026-09-30', which is in the future, providing a deceptive claim about the skill's review history.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 24, 2026, 09:46 PM