The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-generated content from Google Play reviews, which is a known vector for indirect prompt injection where an attacker might embed instructions within a review to influence the agent.
Ingestion points: User-generated review bodies, titles, and display names are ingested via gplay reviews list, gplay reviews view, and gplay reviews history (SKILL.md).
Boundary markers: The documentation instructs the agent to treat content as untrusted data and provides behavioral rules, though no technical delimiters (like XML tags or JSON schemas) are explicitly enforced in the prompt interpolation process.
Capability inventory: The skill includes the gplay reviews reply command, which allows the agent to perform outward-facing writes by posting public replies (SKILL.md).
Sanitization: The documentation suggests an enforcement backstop using the GPLAY_READONLY=1 environment variable to disable mutating commands, which acts as a safety measure.
[COMMAND_EXECUTION]: The skill operates by executing the gplay CLI tool and its various subcommands (list, view, history, reply) to interact with the Google Play API and Cloud Storage.

gplay-reviews