The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and executes shell scripts (Bash) to interact with infrastructure components like databases (Postgres, MySQL, Mongo, Redis), cloud providers (AWS, GCP, Azure), and mainframes (IBM z/OS). This behavior is the primary intended purpose of the skill and is documented as a core feature for building independent verification harnesses.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface (Category 8) as it ingests and parses untrusted data from external systems, such as API responses, database rows, and CLI outputs. The instructions explicitly mitigate this risk by mandating structured data parsing using tools like jq to avoid unsafe interpolation and by recommending the use of atomic, isolated directories for test execution.
[EXTERNAL_DOWNLOADS]: The skill performs network operations using standard tools such as curl, wget, and ssh to interact with system interfaces and perform last-mile verification against authoritative backends (e.g., Okta API, Cloud provider CLIs, Vault). These operations are targeted at well-known services and are restricted to the intended scope of the verification task.
[CREDENTIALS_UNSAFE]: While the skill manages sensitive credentials (API tokens, SSH keys, database passwords), it adheres to safe practices by instructing the agent never to hardcode secrets and instead to resolve them through environment variables or local credential files.

parallel-verification