task-efficiency-advisor

Plan efficient device-aware, privacy-aware task workflows; compare supplied quotes, organize deadline tasks, draft from dictated notes, and learn locally from successful feedback.

Task efficiency advisor

  1. When asked for the built-in synthetic food-quote demonstration, execute python "{baseDir}/scripts/task_efficiency_advisor.py" demo exactly. For the strict budget/party/allergy demonstration, execute the same helper with demo-strict. Return stdout without reconstructing it; complete only after the command exits successfully.
  2. Otherwise collect task, available_device, privacy_requirement, and user_preferences; preserve the caller's data_classification label and list every missing detail instead of guessing. Complete when the request is valid JSON with no invented values.
  3. Write the supplied request to a temporary JSON file, then execute python "{baseDir}/scripts/task_efficiency_advisor.py" recommend --input <temporary-request.json> --state-dir <local-state-directory>. Return the helper's JSON result; complete when it contains recommended_workflow, ordered steps, required_tools, suitability_rationale, and missing_information.
  4. For food quotes, treat under as strict <; include every supplied fee and tip; flag missing/unconfirmed charges; compare party_size only with explicit serves; and keep missing allergen or cross-contamination facts unresolved. Never infer allergy suitability. If no quote satisfies every requirement, return the helper's exact decision: "no qualifying option" plus its specific clarification_questions.
  5. Present every approval_required_actions entry as requiring explicit human approval. Never send, purchase, book, order, connect an account, or perform another external action; the helper only plans, drafts, sorts, and calculates.
  6. For positive feedback, execute python "{baseDir}/scripts/task_efficiency_advisor.py" feedback --input <feedback.json> --state-dir <same-local-state-directory>. Call persisted reuse workflow learning, never model retraining; complete when stdout reports saved: true.
  7. Verify changes with python -m unittest discover -s "{baseDir}/tests" -v; complete only when tests pass and synthetic inputs remain distinguished from live integrations.