openai-api
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/check-versions.shexecutes shell commands includingnpm listandnpm viewto verify package versions against the npm registry. - [INDIRECT_PROMPT_INJECTION]: The skill templates, particularly
templates/cloudflare-worker.tsandtemplates/structured-output.ts, are vulnerable to indirect prompt injection due to lack of input protection. - Ingestion points: User-provided strings are ingested via the
messageparameter intemplates/cloudflare-worker.tsand thetextsarray intemplates/structured-output.ts. - Boundary markers: Absent. The templates interpolate user input directly into the
messagesarray of the OpenAI API call without using delimiters or system-level instructions to ignore embedded commands. - Capability inventory: The skill uses
fetchto communicate with the OpenAI API and includes scripts that can write files to the local system (e.g.,fs.writeFileSyncin multiple templates). - Sanitization: Absent. There is no evidence of input validation, escaping, or filtering of user-provided content before it is sent to external LLM endpoints.
Audit Metadata