The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze HTML snapshots via agent-browser snapshot -i and screenshots via agent-browser screenshot from the application under test. This creates a surface where malicious instructions embedded in the application's UI or metadata could influence the agent's behavior.
- Ingestion points: Application HTML snapshots and screenshots (SKILL.md).
- Boundary markers: None identified in the skill instructions to delimit or warn about untrusted application content.
- Capability inventory: Shell access for browser automation, starting development servers, and executing generated scripts (SKILL.md).
- Sanitization: No explicit instructions for sanitizing or validating the content retrieved from the browser.
[DYNAMIC_EXECUTION]: The instructions direct the agent to generate and execute code at runtime. Specifically, it tells the agent to "Write a temporary seed script, run it before browser tests, then clean up after and delete the script" (SKILL.md).
[COMMAND_EXECUTION]: The skill utilizes shell commands through a generalPurpose subagent to control the agent-browser CLI tool and manage development server processes such as bun run dev (SKILL.md).

qa-testing