The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data through scraping and enrichment via SandBase tools, which presents an attack surface for indirect prompt injection.
Ingestion points: External data sources, search results, web scraping, and data enrichment payloads (SKILL.md, references/sandbase-api-map.md).
Boundary markers: The skill explicitly references 'authorization boundaries' and mandates that the agent preserve the user's chosen stack and source material.
Capability inventory: The skill utilizes sandbase_run for executing external capabilities (SKILL.md).
Sanitization: The workflow requires verifying real output with tests, calculations, or source comparison, and stating material assumptions to prevent blind obedience to potentially poisoned inputs.
[DYNAMIC_EXECUTION]: The skill uses a discovery-based model (sandbase_discover and sandbase_inspect) to identify and execute external capabilities at runtime via sandbase_run.
Evidence: The workflow involves polling asynchronous work and submitting schema-defined arguments discovered live, rather than hard-coding endpoints.
Mitigation: Significant guardrails are present, including mandatory cost estimation, unit price displays, and explicit user confirmation before any paid or external call is executed.

revops