The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches repo health data from an external Sanity.io dataset. It includes a feature to generate "investigation prompts" for an agent based on this data, which could be exploited if the source data is attacker-controlled.
Ingestion points: External data is retrieved from the bench dataset using GROQ queries defined in references/groq-recipes.md.
Boundary markers: No boundary markers or instructions to ignore embedded commands are present when interpolating dataset content into prompts.
Capability inventory: The skill environment allows shell command execution (pnpm, gh, curl) and network requests to external APIs.
Sanitization: There is no evidence of sanitization or validation of the data retrieved from the external source before processing.
[CREDENTIALS_UNSAFE]: The instructions suggest running sanity debug --secrets to retrieve authentication tokens. While this is standard for the Sanity ecosystem, performing this action within an agent's context exposes sensitive credentials to the model.
[EXTERNAL_DOWNLOADS]: The skill uses curl to fetch monitoring data from the vendor's official API at mhfozd0z.api.sanity.io.
[COMMAND_EXECUTION]: The skill facilitates the execution of various development and operational commands, including pnpm for building and testing, and gh for triggering GitHub Actions workflows.

sanity-radar