The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install a CLI tool from external repositories associated with the author. Evidence: npx -y @mvanhorn/printing-press-library install atlas-obscura --cli-only and go install github.com/mvanhorn/printing-press-library/library/travel/atlas-obscura/cmd/atlas-obscura-pp-cli@latest.
[COMMAND_EXECUTION]: The core functionality relies on executing a local binary atlas-obscura-pp-cli with various subcommands. The use of the --agent flag includes --yes, which bypasses confirmation for operations.
[DATA_EXFILTRATION]: The CLI provides a --deliver webhook:<url> sink that allows command output to be POSTed to any external endpoint. Additionally, the feedback command can be configured via ATLAS_OBSCURA_FEEDBACK_ENDPOINT to send local diagnostic data to a remote server.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted community content from the Atlas Obscura website.
Ingestion points: Descriptions of places, categories, and destinations fetched via the places, destinations, and route commands (e.g., SKILL.md recipes).
Boundary markers: The instructions do not define delimiters or warnings to prevent the agent from treating instructions inside the place data as system commands.
Capability inventory: The CLI has the ability to write to files (file:<path>) and transmit data over the network (webhook:<url>).
Sanitization: There is no evidence of sanitization or safety filtering of the community-sourced descriptions before they enter the agent context.

pp-atlas-obscura