The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to autonomously resolve "Configuration Errors" by installing missing software dependencies. An example in references/EXAMPLES.md shows the agent installing the 'pandas' library to recover from a failure. This automated installation capability allows the agent to execute code from remote packages that are not pre-verified.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to trigger based on and process untrusted inputs such as error messages, exceptions, and tracebacks from external processes.
Ingestion points: Error strings and system outputs monitored by the triggers defined in SKILL.md.
Boundary markers: No explicit delimiters or instructions are provided to help the agent distinguish between legitimate error data and potential embedded instructions.
Capability inventory: The agent has the authority to perform environment modifications, including installing software and changing task logic.
Sanitization: There are no documented mechanisms for sanitizing or validating the content of the error messages captured from external tools.
[COMMAND_EXECUTION]: The skill includes instructions to use npx commands to add additional components from the developer's ecosystem. This facilitates the execution of shell commands to modify the agent's environment and capabilities.

error-recovery