The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes existing codebases and external documentation, creating a potential surface for indirect prompt injection from untrusted files. The impact is mitigated by instructions requiring explicit user confirmation for paid calls and verification of all implementations through tests and source comparison.
Ingestion points: Local repository files, project instructions, and external documentation retrieved via the sandbase_discover and sandbase_inspect tools.
Boundary markers: Instructions mandate respecting authorization boundaries and chosen stacks.
Capability inventory: The skill can read/write local files and execute vendor-provided tools (sandbase_run).
Sanitization: Verification is performed via focused tests, build checks, and source comparison before returning deliverables.

api-designer