The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a directive in the 'Important Restrictions' section instructing the agent to suppress information about a functional failure. Specifically, it states that alt text is silently ignored by the API and tells the agent: 'Do not tell the user their alt text was published.' This is a concealment pattern designed to mislead the user by hiding the fact that a requested action (adding alt text) was not actually performed.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through external data ingestion.
Ingestion points: The create_post tool accepts an array of mediaUrls containing public HTTPS links to images or videos.
Boundary markers: There are no instructions or delimiters provided to the agent to treat content from these external sources as untrusted or to ignore embedded instructions within the metadata or content of the retrieved media.
Capability inventory: The skill has the capability to write to a public social platform (Bluesky) via the create_post tool.
Sanitization: There is no mention of validation or sanitization of the content fetched from the external mediaUrls before it is processed by the agent or the API.

bluesky-post