verify
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill automatically executes
npx -y skills@latest update -ywhen a local timestamp file (.shiplight-agent-skills-last-update) is missing or older than 24 hours. This pattern downloads and executes code from the NPM registry at runtime. - [CREDENTIALS_UNSAFE]: For applications requiring authentication, the skill instructs the agent to save session cookies and localStorage data into the user's home directory at
~/.shiplight/<site_url>/storage-state.json. These files contain sensitive authentication tokens used to bypass login screens in subsequent sessions. - [DATA_EXFILTRATION]: The
upload_html_reporttool transmits local browser session artifacts, including video recordings, execution traces, and generated HTML reports, to the Shiplight cloud service to generate shareable URLs. - [COMMAND_EXECUTION]: The skill facilitates starting local development servers by executing commands such as
npm run devoryarn devto ensure the application is available for verification. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by ingesting the DOM tree and screenshots of web pages via the
inspect_pagetool. - Ingestion points: Browser DOM content and visual screenshots processed from the target application under test.
- Boundary markers: None identified; instructions do not specify delimiters to separate application content from agent instructions.
- Capability inventory: The skill can execute shell commands for dev servers, perform remote package updates via
npx, and exfiltrate data to the cloud. - Sanitization: No evidence of sanitization or filtering of the ingested web content before it is processed by the agent.
Audit Metadata