The Agent Skills Directory

[DYNAMIC_EXECUTION]: The documentation for Bitrix Agents (CAgent) in rules/agents.md explicitly notes that the name field of an agent is PHP code executed using eval(). This is a core mechanism of the Bitrix framework for scheduling tasks.
[INDIRECT_PROMPT_INJECTION]: A vulnerability surface exists in rules/agents.md where user-supplied or runtime variables (like $offset) are interpolated into the string that will eventually be eval()'d. The skill correctly identifies this risk and advises developers to "cast/escape them" to prevent arbitrary code execution.
[COMMAND_EXECUTION]: The skill documents the use of the Bitrix CLI tool in rules/messenger-config.md for consuming message queues (e.g., php bitrix/bitrix.php messenger:consume). This is standard operational procedure for the Bitrix Messenger system.
[INDIRECT_PROMPT_INJECTION]: The Messenger system described in rules/messenger.md and rules/messenger-config.md processes data serialized as JSON. While this is structured data, the skill notes that messages are rebuilt via constructor parameters, which is a potential surface for injection if message contents are not properly validated by the receiver's process() method.
Ingestion points: Variables interpolated into CAgent strings (rules/agents.md); Public properties of AbstractMessage classes (rules/messenger.md).
Boundary markers: None for agent name strings; JSON structure for Messenger messages.
Capability inventory: PHP eval() execution via CAgent; Callable execution via Application::addBackgroundJob (rules/background-job.md).
Sanitization: Documentation includes explicit instructions to cast or escape arguments in agent names.

bitrix-background-jobs