The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill takes user-provided parameters such as <REQ-CODE>, --be-container, and --user and interpolates them directly into shell command templates without sanitization. If these inputs contain shell metacharacters (e.g., ;, |, `, $(...)), they could alter the behavior of the resulting script.
Ingestion points: User input for <REQ-CODE>, --since, --be-container, and --user parameters.
Boundary markers: None identified; user input is directly placed into templates.
Capability inventory: The skill has the capability to write .sh files to the local filesystem (scratchpad/).
Sanitization: No validation or escaping is specified for the input parameters before they are used in commands.
[DYNAMIC_EXECUTION]: The skill generates executable shell scripts (.sh) by populating predefined templates with user-supplied data. Although the agent is explicitly restricted by disable-model-invocation: true and RULE-01 from executing these scripts itself, the generation of local executable artifacts based on untrusted input remains a low-risk concern.
[COMMAND_EXECUTION]: The primary function of the skill is to define and generate complex shell commands involving docker logs and grep for use on production servers (hcm-bonbon-sv03). The instructions provide a specific structure for executing these operations, including piping and output redirection.

check-approve