skills/blink-new/claude/pg-boss/Gen Agent Trust Hub

pg-boss

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill utilizes standard, well-known Node.js packages including pg-boss, pg, and date-fns-tz from the NPM registry.
  • [COMMAND_EXECUTION]: The debugging section provides shell commands that use the railway CLI and node -e to interact with environment variables and perform direct database queries.
  • [DATA_EXFILTRATION]: Instructions suggest running railway variables to retrieve sensitive environment variables such as DATABASE_URL and DATABASE_PUBLIC_URL. While intended for developer debugging, these commands provide a method to access and display credentials.
  • [INDIRECT_PROMPT_INJECTION]: The skill describes a job processing system that ingests external data into the agent's context through job payloads.
  • Ingestion points: Job data is passed into the queue via boss.send() and retrieved in worker handlers via boss.work() (e.g., userId, email addresses).
  • Boundary markers: The provided code snippets do not implement specific delimiters or instructions to ignore embedded commands within the job data.
  • Capability inventory: The skill allows database interaction via the PgBoss client and arbitrary task execution within worker functions (e.g., processJob(job)), which could include network or file operations.
  • Sanitization: The example code demonstrates functional usage but does not explicitly show sanitization or validation of the job payload content.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 04:31 PM