The Agent Skills Directory

[COMMAND_EXECUTION]: The skill reproduction and verification steps rely on executing npx playwright test through a shell, incorporating variables provided via the user arguments.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted test files and metadata, which serves as a vector for malicious instructions to influence the agent's actions.\n
Ingestion points: The skill receives a test file path and test name as input from the user via the $ARGUMENTS field in SKILL.md.\n
Boundary markers: There are no explicit markers or instructions provided to isolate user-supplied data from the skill's operational commands.\n
Capability inventory: The agent is capable of executing shell commands and performing file system writes, allowing malicious content in a test file to potentially compromise the environment during execution.\n
Sanitization: The skill lacks validation or sanitization of the file paths, creating a surface for command injection if the input contains shell-sensitive characters like semicolons or pipes.\n- [DYNAMIC_EXECUTION]: The skill involves modifying existing code and subsequently executing it to verify fixes. This process executes code at runtime that may be derived from or influenced by untrusted external sources.

fix