The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for capturing external HTTP response data which is then passed to script and shell execution steps, creating a potential surface for indirect prompt injection.\n
Ingestion points: Data is ingested from external endpoints using the type: http step documented in references/patterns.md.\n
Boundary markers: The skill does not define explicit boundary markers for the ingested data, though it advises using environment variables for safer data handling.\n
Capability inventory: The skill has access to type: script (Python, Node.js), type: shell, and type: container steps, which allow for code and command execution.\n
Sanitization: Examples demonstrate using environment variables and structured JSON parsing to process input data, which reduces the risk of direct code injection compared to string interpolation.\n- [DYNAMIC_EXECUTION]: The skill enables the creation and execution of scripts at runtime to perform validation logic.\n
The type: script step documented in SKILL.md allows embedding Python or Node.js code directly within Atmos configuration files.\n
This execution is a primary function of the skill for performing integration and smoke tests.\n- [COMMAND_EXECUTION]: The skill supports executing arbitrary shell commands through the type: shell step.\n
The references/patterns.md file shows examples of running local shell scripts like ./tests/smoke.sh to validate deployment status.

atmos-tests