The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary focus is the execution of arbitrary shell commands through the type: shell step. It instructs users to run local scripts (e.g., ./tests/smoke.sh) and CLI tools to validate system states.
[DYNAMIC_EXECUTION]: The skill facilitates the execution of dynamic scripts using type: script with Python or Node.js interpreters. It provides patterns for embedding logic directly within Atmos configurations.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface where data from external HTTP endpoints is ingested and subsequently used as input for scripts.
Ingestion points: Data enters the agent context via the type: http step which captures response bodies from external URLs.
Boundary markers: There are no explicit boundaries or 'ignore' instructions for the captured HTTP response data, though it is scoped to specific step results.
Capability inventory: The skill enables shell commands, script execution (Python/Node.js), container runtime access, and native atmos CLI operations.
Sanitization: The documentation actively promotes a security best practice by instructing users to pass captured data through environment variables (env: HEALTH_RESPONSE: '{{ .steps.health.value }}') rather than direct string interpolation into script bodies, which mitigates simple code injection risks.

atmos-tests