The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides commands to scan and crawl public Substack content, which represents an ingestion point for untrusted data. This data could contain hidden instructions designed to influence the agent's behavior.
Ingestion points: The feed scan, feed crawl, and feed crawl-all commands in SKILL.md retrieve data from external Substack URLs.
Boundary markers: There are no specified delimiters or instructions for the agent to treat crawled data as untrusted or to ignore embedded instructions.
Capability inventory: The skill grants the agent capabilities to create, update, publish, and delete posts, providing a significant impact surface if an indirect injection is successful.
Sanitization: No sanitization or validation mechanisms are mentioned for the external content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill defines a set of interactions with the substackctl CLI tool, instructing the agent to execute shell commands to manage posts, sections, and profiles.

substackctl