web-research
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFECREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill's request script (
scripts/request) accesses a 1Password service account token by sourcing a file from a hardcoded absolute path:/Users/danielhowells/.codex/plugins/secrets/1password-service-account.env. While this is a common pattern for the author's local environment, it involves accessing a sensitive credential store. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and synthesize information from the open web, which is an untrusted source.
- Ingestion points: Web search results and page contents are retrieved via Exa and Tavily APIs (documented in
references/providers.md) and processed by the agent. - Boundary markers: The instructions do not define clear boundaries or provide safeguards to ignore potential instructions embedded in the retrieved web data.
- Capability inventory: The skill can execute shell scripts that perform network operations and read local configuration files.
- Sanitization: The skill lacks mechanisms to sanitize or escape potentially malicious instructions embedded in retrieved web pages before they are integrated into the agent's context.
Audit Metadata