The Agent Skills Directory

Indirect Prompt Injection Surface: The skill ingests data from local files such as .tmp/examples-start-logs/main_*.log, per-example logs, and example source files. This constitutes a surface where the agent processes content that could potentially contain instructions. The skill addresses this by enforcing explicit logical boundaries and forbidding the agent from executing or controlling any scripts or targets identified during analysis.\n- Restricted System Command Execution: The skill utilizes a set of standard system utilities, including git, find, ls, ps, sed, and rg, to perform discovery and text processing. These tools are restricted to read-only operations to ensure the agent does not modify the filesystem or influence active processes during its diagnostic tasks.\n- Defined Safety Constraints: The instructions include specific prohibitions against starting, stopping, or modifying background processes, and they explicitly forbid the agent from requesting elevated privileges. These constraints are intended to limit the agent's scope to informational analysis and prevent it from being used to alter the execution environment.

examples-run-analysis