The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes output from the PlanetScale API, which represents an untrusted data source that could be manipulated to include malicious instructions.
Ingestion points: Output from pscale insights subcommands including queries, samples, errors, and recommendations.
Boundary markers: None identified; the instructions do not specify delimiters to isolate external data from the prompt context.
Capability inventory: Shell command execution via the pscale CLI for database inspection and schema modification.
Sanitization: No explicit sanitization or validation of the retrieved API data is defined.
[COMMAND_EXECUTION]: The skill uses the pscale CLI to perform various operations against a database branch, including data retrieval and dismissal of schema recommendations.
[DATA_EXFILTRATION]: The skill explicitly warns that query samples and error details can contain sensitive information such as SQL context, usernames, and addresses, creating a risk of data exposure if these outputs are logged or shared.
[PRIVILEGE_ESCALATION]: The skill mentions the use of the --force flag with pscale insights recommendations dismiss, which allows for the bypass of confirmation prompts during command execution.

pscale-insights