The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute a local Python script (check_environment.py) or a shell wrapper (check-environment.sh) to perform system diagnostics. This execution is scoped to a shared library directory and is the primary, intended function of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and parses JSON output generated by an environment inspection script. This process represents a potential attack surface if the script targets an external project directory using the --project flag, as content within that directory could influence the diagnostic report. However, the skill instructions provide a rigid structure for output (Verdict, Check table, Actions, and Limits), which serves as a boundary to prevent the agent from treating report data as new instructions.
Ingestion points: Data enters the agent context from the JSON output of the check_environment.py script executed via the shell.
Boundary markers: The instructions define a strict format for presentation (verdict line, check table, verbatim action list) and explicitly state the skill should only report what the new output shows, not memory or user claims.
Capability inventory: The skill has the capability to execute shell commands (python3, bash) and read local file paths.
Sanitization: The agent is instructed to parse structured JSON and map specific fields to a table format, providing a layer of validation over the ingested data.

environment-check