The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes output from external MikroTik routers, which could potentially contain malicious instructions if a device is compromised or its responses are manipulated.
Ingestion points: Commands such as centrs retrieve, centrs execute, and centrs api pull data from remote devices into the agent context.
Boundary markers: Although data is returned in a structured JSON envelope (containing data, warnings, tips, and meta fields), the skill lacks explicit instructions to the agent to treat this data as untrusted or to ignore embedded commands.
Capability inventory: The skill provides access to local credential files, the ability to execute shell commands, and the capability to transfer files to and from remote devices.
Sanitization: The tool features an explain command for offline syntax analysis and utilizes the device's own :parse and /console/inspect machinery to validate commands before execution.
[DATA_EXFILTRATION]: The skill accesses sensitive information required for router management. It reads the local WinBox address book file at ~/.config/tikoci/winbox.cdb to resolve router addresses and credentials. It also utilizes environment variables like CENTRS_USERNAME and CENTRS_PASSWORD. The documentation warns that command output (e.g., from /export) may contain unredacted secrets like private keys and API tokens.
[EXTERNAL_DOWNLOADS]: The skill encourages downloading and running the @tikoci/centrs package using bunx from the npm registry, which executes code from a remote source.
[COMMAND_EXECUTION]: The skill's primary functionality relies on executing the centrs CLI tool to perform network operations, file transfers, and device configuration, which involves spawning subprocesses to interact with network infrastructure.

routeros-centrs