The Agent Skills Directory

[PERSISTENCE_MECHANISMS]: The skill instructions in references/change-tracking.md direct the agent to modify the ~/.claude/settings.json configuration file to add a SessionStart hook. This hook is designed to automatically execute a local binary (notfair-change-watch) whenever a new session begins, which constitutes a persistence mechanism.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the Google Ads API, such as search terms and campaign names, which could potentially contain malicious instructions meant to influence the agent's actions.
Ingestion points: Data is retrieved via the runScript tool using GAQL queries (as defined in SKILL.md).
Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard potentially malicious content within the API responses.
Capability inventory: The agent has permissions to execute JS scripts via runScript, perform bulk keyword mutations, and modify account budgets and bids.
Sanitization: The provided instructions do not include mechanisms for sanitizing or validating the content of the API data before it is incorporated into the agent's reasoning process.

google-ads