The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses authoritative language to override default agent behaviors, mandating a specific persona ('Expert coding agent') and requiring exact responses such as 'ULTRAWORK MODE ENABLED!'. It explicitly forbids process narration, which can reduce user oversight during complex operations.\n- [COMMAND_EXECUTION]: Instructions rely on shell commands for automation and testing, including the use of curl, tmux, and git. It designates eval as the standard surface for executing concurrently dispatched code tasks.\n- [EXTERNAL_DOWNLOADS]: Fetches and utilizes the agent-browser tool from the Vercel Labs repository on GitHub when native browser automation is unavailable.\n- [PROMPT_INJECTION]: Potential surface for indirect prompt injection:\n
Ingestion points: Processes user-supplied prompts and local codebase files.\n
Boundary markers: None present to distinguish untrusted data from the core directives.\n
Capability inventory: Full shell access via eval and bash, network access via curl, and browser control tools.\n
Sanitization: No input validation or escaping mechanisms are specified for handling external data.

ultrawork