The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external datasets and file inputs to generate SQL queries, creating a surface for indirect prompt injection.
Ingestion points: SKILL.md instructs the agent to "Inspect the available files, runtime, versions, inputs" as the first step of its workflow.
Boundary markers: The skill explicitly instructs the agent to "parameterize user input" in the Focus checks section, which serves as a defensive boundary against injection-style attacks.
Capability inventory: The agent is authorized to produce implementation artifacts (SQL code) and utilize SandBase tools (sandbase_run) for external data or inference.
Sanitization: The instructions include mandatory parameterization of user input and validation of totals against independent queries.
[SAFE]: The skill uses vendor-specific tools (sandbase_discover, sandbase_inspect, sandbase_run, sandbase_run_get, sandbase_runs, sandbase_account) provided by the author 'sandbaseai'. The workflow includes security-conscious steps such as cost estimation and explicit user confirmation before any paid external calls.

sql-queries