The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from blockchain queries and Dune documentation searches which could contain malicious instructions.
Ingestion points: Data enters the agent's context through the output of commands like dune query run, dune query run-sql, dune dataset search, dune execution results, and dune docs search.
Boundary markers: The skill does not provide instructions for using delimiters or "ignore" instructions for the data retrieved from the CLI are provided in the skill instructions.
Capability inventory: The skill uses Bash(dune:*) which allows creating, updating, and archiving queries, visualizations, and dashboards. It also has Bash(curl:*) permissions.
Sanitization: The skill instructs the agent to redact API keys from output, but no other sanitization of retrieved data is mentioned.- [EXTERNAL_DOWNLOADS]: The skill relies on the dune CLI tool, which is identified as a vendor-managed resource that is automatically installed on first use. This is standard operational behavior for this vendor skill and is documented neutrally.- [CREDENTIALS_UNSAFE]: The skill manages authentication via Dune API keys. It employs safe secret management practices, recommending the use of environment variables or local configuration files (~/.config/dune/config.yaml) and explicitly instructing the agent to redact keys from user-facing output.- [COMMAND_EXECUTION]: The skill facilitates the execution of bash commands using the dune CLI and curl for interacting with the Dune API and performing on-chain analytics.

dune