The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill handles sensitive hardware credentials, including SSH and root passwords. While the underlying MATLAB system may persist these in plaintext settings, the skill includes a dedicated 'Credential Hygiene' section that explicitly prohibits the agent from hardcoding secrets, instead mandating the use of secure user prompts or local environment files.
[COMMAND_EXECUTION]: The skill uses the evaluate_matlab_code tool to perform critical hardware operations like reconfiguring network interfaces, updating firmware, and rebooting targets. These are the primary intended functions of the skill and are protected by robust guardrails requiring the agent to show the user the exact command string and obtain explicit confirmation before every execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external hardware targets (e.g., application names, target status) which could theoretically serve as an injection vector. This risk is minimized by the skill's 'Human-in-the-loop' design. Evidence Chain: (1) Ingestion points: status(tg), getTargetNames, getStartupApp, and tg.TargetOptions in SKILL.md; (2) Boundary markers: Not explicitly specified for data ingestion; (3) Capability inventory: Full MATLAB execution via evaluate_matlab_code as defined in manifest.yaml; (4) Sanitization: Mitigated by mandatory user review and confirmation of all commands derived from or applied to the target state.

simulink-configure-real-time-target