The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to navigate to and process data from external URLs via browser_navigate and browser_network_requests. This creates an ingestion surface for untrusted data that could contain malicious instructions.
Ingestion points: Browser navigation to arbitrary URLs (Step 1) and network request capture.
Boundary markers: None specified in the instructions.
Capability inventory: browser_navigate, browser_network_requests, browser_evaluate (JS execution), and file writing for adapter generation.
Sanitization: None provided for the data extracted from network responses.
[DYNAMIC_EXECUTION]: The templates use page.evaluate to execute dynamic JavaScript code within the browser context. This is used to simulate fetch requests and extract data from the page DOM at runtime.
[EXTERNAL_DOWNLOADS]: The skill requires the @jackwener/opencli package, which is an external dependency used to build the CLI adapters.
[CREDENTIALS_UNSAFE]: One of the TypeScript templates (TS — Header) includes a hardcoded Bearer token. Although this appears to be a commonly known public token for guest access, hardcoding authentication tokens in templates is a sensitive practice.
[DATA_EXFILTRATION]: The skill demonstrates how to extract sensitive authentication data from the browser environment, specifically accessing document.cookie and X-Csrf-Token to facilitate authenticated API requests.
[COMMAND_EXECUTION]: The skill provides instructions for the user or agent to run shell commands such as npm run build and opencli for testing and registration purposes.

qiaomu-opencli-oneshot