The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes a file reactome-skill_audit_result_v1.json that contains self-referential claims about its own safety and audit status (e.g., "security: PASS", "grade: Production Ready"). This is a documented vector for attempting to influence or bypass security scanners through indirect injection by providing false authoritative context.
Ingestion points: reactome-skill_audit_result_v1.json is provided as part of the skill's file set.
Boundary markers: Absent; the content is presented as legitimate metadata.
Capability inventory: The skill uses requests and reactome2py to perform network operations against external API endpoints in scripts/reactome_tool.py.
Sanitization: Absent; the metadata is presented without filters.
[METADATA_POISONING]: The skill contains inconsistent and deceptive metadata designed to create a false sense of legitimacy. The skill_spec.yaml identifies the author as "K-Dense Inc." while SKILL.md identifies "AIPOCH". Additionally, the packaging plan in skill_spec.yaml references a file named risk03.md which is not present in the distribution, indicating inconsistent or misleading packaging practices.

reactome-skill