The Agent Skills Directory

[SAFE]: The skill implements a governance dashboard for platform teams. It aggregates cost and performance data from internal services using a specific Busabase-backed architecture. The analysis found no evidence of prompt injection, unauthorized data exfiltration, or persistence mechanisms.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from external records (routes and settings). However, the instructions explicitly prohibit the use of LLMs for anomaly detection or decision-making, which are handled by deterministic, rule-based functions.
Ingestion points: Data is ingested via busabase-provider.js from the routes, services, models, and settings bases.
Boundary markers: The SKILL.md defines a strict boundary forbidding LLM invocation for processing this data.
Capability inventory: The app can perform record updates (records.changeRequest) and serve static content via a Hono server (server.js).
Sanitization: Content is sanitized using escapeHtml in the frontend (app.js) and validated through JSON schemas.
[EXTERNAL_DOWNLOADS]: The skill uses well-known and trusted libraries including busabase-sdk, hono, and @hono/node-server. All dependencies are pinned to specific versions in package.json and are standard for the developer platform context.

kelly-llm-gateway