The Agent Skills Directory

[COMMAND_EXECUTION]: The agent uses the uip CLI tool and standard utilities like grep to perform investigative tasks such as fetching job logs and searching for relevant playbooks. These commands are constrained by strict internal guides and the skill's instructions, which explicitly forbid raw REST/curl workarounds and undocumented command discovery.
[PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data, specifically error messages and stack traces provided by the user. However, the agent's behavior is guided by a local playbook corpus, and it includes a mandatory verification checklist and an approval gate for any corrective actions, minimizing the risk of accidental or adversarial obedience.
[DATA_EXFILTRATION]: No unauthorized data exfiltration patterns were identified. Network communication is restricted to authorized UiPath vendor endpoints for fetching job information and documentation. The skill maintains investigation state in a local directory (.local/investigations/) which is handled as verbatim evidence.
[SAFE]: The skill demonstrates secure architecture by implementing clear 'Invariants' that prevent fabrication and self-editing. It delegates all write operations to specific owning skills only after explicit user approval via AskUserQuestion, ensuring human-in-the-loop oversight for all modifications.

uipath-troubleshoot