The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions contain hardcoded passwords ('Abcd1234!') for seeded user accounts ('dennis@surge.app' and 'admin@surge.app') and a PostgreSQL connection string ('postgresql://postgres:postgres@localhost/surge_dev') with embedded credentials.
[COMMAND_EXECUTION]: The skill executes a shell function named 'dev' directly from the user's '~/.zshrc' file using 'zsh -ic', which triggers logic stored in the local shell environment.
[COMMAND_EXECUTION]: Dynamic code execution is performed using 'playwright-cli run-code' to execute arbitrary JavaScript within the browser context for UI emulations.
[DATA_EXFILTRATION]: The skill accesses a local database using the 'psql' utility with hardcoded credentials and manages browser session data via 'playwright-cli', providing a mechanism for data access and manipulation.
[PROMPT_INJECTION]: The agent navigates to and evaluates content from the external website 'hq.jason.surge-eng.com', creating an indirect prompt injection vulnerability if the site content is maliciously altered.
Ingestion points: Evaluation of browser DOM content via 'playwright-cli eval' and interaction with the live dashboard.
Boundary markers: No delimiters or instructions are provided to the agent to ignore embedded commands in the retrieved content.
Capability inventory: Shell execution (zsh, psql, mix, playwright-cli), repository access (git), and network navigation (Playwright).
Sanitization: No validation or sanitization of the external web content is implemented before it is processed by the agent.

dev