The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's examples demonstrate the installation of external Python packages using pip install to resolve environment issues, which involves fetching software from remote registries.
[COMMAND_EXECUTION]: The error recovery protocol defined in SKILL.md and references/EXAMPLES.md instructs the agent to fix environment configuration errors by executing system commands, such as installing dependencies or modifying settings.
[PROMPT_INJECTION]: The skill evaluates untrusted data from error messages, exceptions, and tracebacks to determine its recovery logic, making it susceptible to indirect prompt injection attacks.
Ingestion points: Error messages and stack traces identified as triggers in SKILL.md.
Boundary markers: The protocol lacks delimiters or specific instructions to the agent to treat error content as data only, potentially allowing embedded malicious instructions to be executed.
Capability inventory: The agent is authorized to perform environment modifications, package installations, and alternate logic paths as part of its recovery strategies.
Sanitization: No sanitization or validation logic is defined to check the content of error messages before the agent acts upon them.

error-recovery