The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains a hardcoded authentication token in the 'TS — Header' example (Authorization: Bearer AAAAAAAAAAAAAAAAAAAAANRILgAAAAA...). While potentially a public guest token, hardcoding such values is a poor security practice.
[CREDENTIALS_UNSAFE]: The skill provides explicit instructions for extracting sensitive session identifiers from the browser environment, including the retrieval of session cookies (document.cookie) and CSRF tokens (X-Csrf-Token).
[DATA_EXFILTRATION]: The skill automates the process of making authenticated network requests using fetch with credentials: 'include'. This capability allows for the automated extraction of user-specific data from any website the user is currently logged into in the browser session.
[COMMAND_EXECUTION]: The workflow describes the generation of executable TypeScript and YAML files based on captured network traffic. These generated scripts are then executed locally via npm run build and the opencli command, representing a dynamic script generation and execution risk.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from arbitrary URLs via browser_navigate and browser_network_requests. This data is then used to construct executable adapters without evidence of sanitization or boundary markers.
Ingestion points: Network request logs and page content from external URLs (SKILL.md).
Boundary markers: None identified.
Capability inventory: File writing, browser evaluation, authenticated network requests, and local command execution.
Sanitization: None identified; captured API data is interpolated directly into code templates.

opencli-oneshot