The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the 'first-artifact.html' template from the vendor's repository (latitude-dev). This is an expected retrieval of resources from the service provider's infrastructure.
[COMMAND_EXECUTION]: Employs standard shell utilities including curl, jq, sed, and printf to automate the configuration of environment variables and binary permissions necessary for the telemetry setup.
[CREDENTIALS_UNSAFE]: Accesses local configuration files such as .env, ~/.claude/settings.json, and ~/.hermes/.env to check for existing credentials. It uses specific shell patterns with jq and sed to handle LATITUDE_API_KEY values securely, ensuring they are not printed to terminal output or chat logs.
[INDIRECT_PROMPT_INJECTION]: The skill features a data ingestion point by collecting telemetry data from the Latitude CLI and embedding it into an HTML artifact.
Ingestion points: Trace data, including model inputs and outputs, is retrieved via the Latitude CLI in Step 8.
Boundary markers: The data is placed within a script-based JSON blob in the generated HTML file. The rendering logic includes an explicit escaping function.
Capability inventory: The skill uses file write/append operations and network downloads to generate the local artifact.
Sanitization: The included first-artifact.html template implements a JavaScript esc() function that replaces special HTML characters (&, <, >, ", ') with their respective entities to prevent execution of embedded scripts during rendering.

latitude-setup