The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides detailed recipes for interacting with RouterOS devices via REST and native APIs. This includes using the /execute and /file menus to handle script parsing. These operations are presented as necessary components of the skill's primary function—providing deep syntax inspection for network administrators.
[DYNAMIC_EXECUTION]: Instructions describe using the RouterOS :parse command to dynamically lower scripts into an Intermediate Language (IL) for structural analysis. The skill provides guidelines on how to safely capture this output using temporary files to avoid string-escape collisions.
[INDIRECT_PROMPT_INJECTION]: The skill documentation addresses the risks associated with processing untrusted RouterOS scripts.
Ingestion points: Untrusted data enters the agent context via console input strings and the contents of RouterOS script files read from the device.
Boundary markers: The skill suggests static segmentation of input and uses the :parse engine as a structural validator.
Capability inventory: The documented capabilities include file creation/deletion and script execution on the target RouterOS device.
Sanitization: The skill recommends ASCII-normalization of input to prevent character offset desynchronization and provides a "sentinel decision rule" to distinguish between valid syntax and prospective candidates.
[EXTERNAL_DOWNLOADS]: The skill references external documentation and research artifacts hosted on GitHub repositories owned by the author (tikoci/lsp-routeros-ts, tikoci/restraml, etc.). These links are provided as evidence for the described syntax behaviors and do not involve the installation of unverified software.
[SAFE]: The skill includes "Known hazards" warnings, such as identifying specific keyword paths that can deadlock older versions of the RouterOS REST server, demonstrating a security-conscious approach to automation.

routeros-syntax-inspection