The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches untrusted data from user-supplied URLs, creating an attack surface where malicious instructions embedded in a webpage could influence the agent's behavior.
Ingestion points: Multiple commands (/ad-hooks, /ad-angles, /ad-script, /hook-audit, /image-ad-prompt, /ugc-brief) use the WebFetch tool to retrieve content from external URLs provided in the command arguments.
Boundary markers: The skill instructions do not specify any delimiters (like XML tags or markdown blocks) or "ignore embedded instructions" warnings for the agent when processing fetched content.
Capability inventory: The agent has access to the Bash tool (allowing shell command execution), Read (allowing file access), and WebFetch/WebSearch (allowing network operations).
Sanitization: There are no instructions to sanitize or validate the external content before it is interpolated into the agent's reasoning process.
[DATA_EXFILTRATION]: The skill sends data to a vendor-controlled external API.
The commands /ad-prompts, /ad-teardown, and /hook-bank transmit user search queries and a user-provided API key (HOOKADS_API_KEY) to hookads.app. This is documented as the primary mechanism for accessing the full creative library.
[COMMAND_EXECUTION]: The skill utilizes the shell environment to perform data fetching tasks.
Commands including /ad-prompts, /ad-teardown, and /hook-bank use the Bash tool to execute curl commands. These commands are strictly scoped to interacting with the vendor's REST API at hookads.app.

ad-maker