The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install a CLI tool using npx or go install from the author's repository (github.com/mvanhorn/printing-press-library). This results in the execution of remote code during the setup process.
[EXTERNAL_DOWNLOADS]: The skill downloads binary files and library code from NPM (@mvanhorn/printing-press-library) and GitHub repositories to function.
[DATA_EXFILTRATION]: The CLI tool includes a --deliver webhook:<url> flag that allows the agent to route command results to an external URL. Additionally, the feedback command can be configured to send local logs to a remote endpoint via the WANDERLUST_GOAT_FEEDBACK_ENDPOINT environment variable.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from community-driven external sources including Reddit, Wikipedia, Wikivoyage, and OpenStreetMap (OSM).
Ingestion points: The sync-city, near, and reddit-quotes commands fetch data from these external platforms into the agent's context.
Boundary markers: The instructions do not specify any delimiter or boundary markers to prevent the agent from interpreting instructions embedded within the third-party data.
Capability inventory: The skill uses Read Bash to execute the wanderlust-goat-pp-cli binary and can write output to files or webhooks.
Sanitization: The reddit-quotes command explicitly returns raw text snippets to avoid LLM summarization, which potentially leaves unprocessed external content in the prompt context.
[COMMAND_EXECUTION]: The skill relies on executing the wanderlust-goat-pp-cli binary through a shell interface, taking user-supplied parameters such as addresses, criteria, and identity descriptions as arguments.

pp-wanderlust-goat