The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides native support for executing arbitrary shell scripts and Atmos CLI commands through type: shell and type: atmos steps. This is a core functionality for CI/CD orchestration but presents a significant privilege surface if the underlying workflow files are compromised.
[EXTERNAL_DOWNLOADS]: The skill allows for the declaration and automated installation of tool dependencies such as Terraform, TFLint, and Checkov via the dependencies block. Reference documentation also provides examples of fetching external files using wget within shell steps.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface where external data can influence workflow logic. Specifically, when conditions using CEL expressions and require/assert steps can ingest data from the environment or HTTP resources to gate deployment logic.
Ingestion points: Workflow configuration files (YAML), environment variables, and external HTTP resources fetched during validation steps.
Boundary markers: No explicit markers are defined in the instructions to delimit or sanitize external data before it is evaluated in conditions or shell commands.
Capability inventory: The skill is capable of shell execution, command orchestration, network requests (type: http), and tool installation via the atmos-toolchain skill.
Sanitization: Relies on standard YAML parsing and CEL evaluation; no specific guidance is provided for escaping data interpolated into shell blocks.

atmos-workflows