framework-choice-advisor
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFECREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: Hardcoded credentials were detected in
evals/vp-mandate-one-framework-for-every-layer/task.mdwithin a simulatedapps/admin/cypress.env.jsonfile. The file contains plain-text values forADMIN_API_TOKEN,OWNER_PASSWORD, andBILLING_WEBHOOK_SECRET. These are used in a test scenario to evaluate whether an agent can correctly identify and flag insecure practices. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to analyze untrusted repository data such as
package.json, lockfiles, and configuration files to generate decision records. This process relies on the agent quoting 'observed project signals' without explicit sanitization or filtering instructions for the ingested content, creating a standard surface for indirect prompt injection via malicious repository metadata.
Audit Metadata