The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documents the installation of the Genkit CLI by fetching a script from a remote URL.
Evidence: SKILL.md contains the installation command curl -sL cli.genkit.dev | bash.
[REMOTE_CODE_EXECUTION]: The installation instructions for the developer tools involve piping a remote script directly into a shell interpreter.
Evidence: The command curl -sL cli.genkit.dev | bash is recommended for CLI installation in SKILL.md and references/getting-started.md.
[INDIRECT_PROMPT_INJECTION]: The skill describes an architecture where agents ingest external data and possess powerful capabilities, creating a potential surface for indirect prompt injection.
Ingestion points: User input provided during conversation turns, external file content read through the Filesystem middleware, and deliverables accessed via the Artifacts middleware.
Boundary markers: The documentation utilizes Handlebars templating and JSON schemas for structured data, but does not explicitly provide instructions for isolating or delimiting untrusted external content.
Capability inventory: The skill supports filesystem operations (read_file, write_file), automated tool execution loops, and network communication via model provider plugins.
Sanitization: No specific sanitization or validation logic for ingested external content is demonstrated in the provided examples.
[COMMAND_EXECUTION]: The skill provides instructions for executing the local application and developer environment using the Genkit CLI.
Evidence: SKILL.md provides the command genkit start -- go run . to launch the application with telemetry and the Developer UI.

developing-genkit-go