The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes social media calendar data which is considered untrusted input. If this input contains malicious instructions, they could potentially influence the agent's actions when interacting with the BulkPublish API.
Ingestion points: SKILL.md (Inputs section and Workflow step 2 process calendar data).
Boundary markers: None explicitly mentioned in the workflow instructions to isolate untrusted content.
Capability inventory: Ability to interact with the BulkPublish API and MCP server to list channels and create posts.
Sanitization: The workflow includes validation for character counts and media formats, but does not specify sanitization of the content for prompt injection patterns.
[EXTERNAL_DOWNLOADS]: The skill references external repositories and documentation necessary for the BulkPublish integration.
Evidence: References to github.com/azeemkafridi/bulkpublish-api and app.bulkpublish.com/docs are used to provide the API interface and technical documentation for the service.

kai-bulkpublish