EXZA REMOTE / PROMPT
Full prompt
EXZA Remote — guided, permission-aware workflow My objective: Search allowed source files for exact [OLD VALUE] and identify candidates for manual migration, without changing every match blindly. Target device: the currently paired Windows, macOS or Linux device, only where EXZA tools support it. 1. Connection and capabilities. Check the paired-device status and operating system. List the specific EXZA tools you can actually use for this task, including read-only versus write/terminal/desktop capabilities. Check actual operating system and available EXZA tools; do not assume that a Mac-only GUI or Mail action works on other systems. If a required capability is not exposed, stop that part and explain the limitation. 2. Scope and evidence. Ask which exact project, file, folder, window or recipient I intend if a placeholder remains ambiguous. Start with the narrowest read-only inspection supported by the authorized scope. Respect search-depth, size and pagination limits. Never probe hidden folders, credentials or unrelated personal data. 3. Plan before action. Summarize the observed facts separately from hypotheses. Show the exact paths, commands, text changes, recipients or windows that would be affected, plus risks, reversibility and prerequisites. Do not treat this prompt as permission for a write, shell job, GUI interaction, deletion or outbound email. 4. Approval and execution. Ask for explicit consent before every sensitive operation and respect the native EXZA/ChatGPT or Claude approval flow. Only after approval, use the smallest supported operation. Never escalate privileges, bypass a refused tool or automatically retry destructive steps. 5. Independent verification. Use a suitable read-only EXZA check to confirm what actually happened. When data is unavailable, a command fails, or the platform does not support the operation, say so instead of inventing success. Return your answer under: Device and permissions; Observations; Proposed steps; Approval needed; Verified results; Open questions. Keep sensitive values masked and make each recommendation actionable.