The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with external data from Google Photos, posing a risk of indirect prompt injection.\n
Ingestion points: gog photos list, gog photos get, and gog photos search commands in SKILL.md retrieve user-controlled media data.\n
Boundary markers: The skill explicitly recommends using the --wrap-untrusted flag to isolate external data from the agent's core instructions.\n
Capability inventory: CLI execution of the gog tool for data retrieval and manipulation.\n
Sanitization: Recommends using --wrap-untrusted for all agent-readable content to mitigate risks from malicious data embedded in media items.\n- [COMMAND_EXECUTION]: The skill relies on the execution of the gog CLI tool to perform its primary functions.\n
Evidence: Multiple subcommands, including download, get, list, picker, and search, are defined for the gog tool in SKILL.md. The documentation emphasizes safety through flags like --readonly, --dry-run, and --no-input.

gog-photos