course-builder-incident-forensics

Warn

Audited by Gen Agent Trust Hub on Feb 28, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes multiple system commands including vercel, bun, rg, and node. It specifically runs bun src/index.ts from a local repository and uses node -e to execute inline JavaScript.
  • [CREDENTIALS_UNSAFE]: The skill instructs the agent to source .env.production.local. This loads highly sensitive production credentials and secrets (such as database connection strings) into the active shell environment. While a guardrail prohibits printing these secrets, they remain accessible to the agent and any commands it executes.
  • [REMOTE_CODE_EXECUTION]: The skill utilizes node -e to perform dynamic code execution. The generated script imports the @planetscale/database library and connects to a production database using the environment's DATABASE_URL to perform diagnostic queries.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through log processing. Ingestion points: Axiom queries (axiom query) and Vercel runtime logs (vercel logs). Boundary markers: None provided in the instructions. Capability inventory: Local command execution, file system access, and production database access. Sanitization: No sanitization or validation of log content is performed before the agent processes and interprets the data.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 28, 2026, 10:32 PM