The Agent Skills Directory

[SAFE]: The skill facilitates cursor design using local resources and limits external interactions to vendor-owned tools (sandbase_run, sandbase_inspect). These interactions are governed by clear instructions to minimize data transfer and avoid hard-coding parameters, which aligns with secure development practices.
[SAFE]: The workflow includes explicit requirements for user authorization before any paid service execution and requires manual verification of all outputs. These human-in-the-loop checkpoints significantly reduce the risk of autonomous malicious actions.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a potential attack surface for indirect prompt injection, as it is designed to ingest and process user-provided files and external media generated via SandBase. Ingestion points: SKILL.md (Workflow Step 1) and references/sandbase-api-map.md (Tool descriptions). Capability inventory: The skill uses sandbase_run for external generation. Sanitization: No explicit automated sanitization is described, but the skill mandates quality gates and manual verification of the final artifact to mitigate risks associated with untrusted data processing.

oversized-cursor