Introduce interactive `research` sandbox mode #1050

New issue

Open

opened 2025-10-14 16:43:22 -06:00 by navan · 0 comments

navan commented

2025-10-14 16:43:22 -06:00

Owner

Originally created by @eyaltoledano on 4/4/2025

"When working on tasks, I often need focused research about specific libraries or technologies. Having a dedicated, interactive research mode would help me efficiently gather and organize important contextual information directly related to my tasks or subtasks."

Motivation

Users frequently need to explore, clarify, and document research related to specific tasks or subtasks. Currently, there is no easy and organized way to conduct deep, iterative research and seamlessly incorporate findings into task descriptions. Providing an interactive sandbox for research will empower users to deeply investigate topics and efficiently store research insights within task contexts.

Proposed Solution

Implement an interactive sandbox research command (research) using Inquirer, integrated with Perplexity for deep contextual research:

task-master research [--id=<task|subtask_ids>]

This command enters an interactive chat environment where users can iteratively research topics:

If --id is provided, the specified task/subtask contexts are included in the Perplexity deep research call.
User inputs an initial prompt, Perplexity returns detailed research results.
Users can continue iterating and refining their research through chat prompts.

Interactive commands within the sandbox (persistent menu):

c - Continue chatting (send another research prompt).
spacebar - Autoplay mode:
- Automatically formulates and sends a logical next research prompt to continue exploration.
- Allows rapid iterative exploration.
q - Quit sandbox mode, prompts user to save the research session.
s - Save the current research session to the research/ directory:
- Creates a standardized research report file named research_<taskIds>_<timestamp>.txt.
- Includes original prompt, context task IDs, full conversation history, and summarized findings.
u - Update one or more tasks/subtasks with research insights:
- User specifies task/subtask IDs.
- Appends summarized research information (temperature 0 for maximum detail retention) in XML tags <research on timestamp> to each task description.
x - Complexity Level:
- Persistent menu with three options: MVP, Balanced, Engineered.
- Adjusts Perplexity system prompts to shape the research complexity accordingly.
q/e - Temperature Control:
- Adjust the research creativity with 10 steps (0–10), overriding the default.
- q decreases temperature (less creative), e increases temperature (more creative).
up/down arrows - Traverse received AI responses:
- Highlights messages for focused review.
- Ability to regenerate highlighted responses.
r - Regenerate response:
- Double-tap confirmation required (prompts user clearly before action).

Enhanced Additional Features

Contextual Prompt Suggestions: Automatically suggests prompts based on conversation flow.
Research Session Snapshots: Periodic snapshots for easy rollback during extensive sessions.
Interactive Session History View:
- Command research-report lists available research reports in research/ folder.
- User selects a report to display its information clearly in the CLI.
- Option provided to re-enter and continue the research session from the report.
Confirmation Prompts: Clear confirmations before critical actions (saving, quitting, updating tasks).

High-Level Workflow

Initiate sandbox with optional task/subtask context.
Collect initial research prompt from user and query Perplexity.
Display detailed response and interactive command menu using Inquirer.
Support iterative chat research sessions with continued queries.
Implement convenient saving, updating, autoplay functionalities, and enhanced interactive controls.
Provide clear, intuitive UX through the interactive Inquirer interface.

Key Elements

Command syntax:

task-master research --id=5,10

Interactive Inquirer-based UX for smooth research workflow.
Flexible research context and prompt management.
Easy integration of research findings directly into task descriptions.

Implementation Considerations

Seamless integration with Perplexity for accurate deep research.
Interactive and intuitive UX using Inquirer for ease of use.
Robust summarization and research documentation in task context.
Efficient handling of multiple task/subtask contexts.
Clear CLI and file management conventions for saving and updating.

Out of Scope (Future Considerations)

GUI-based research sandbox

*Originally created by @eyaltoledano on 4/4/2025* > "When working on tasks, I often need focused research about specific libraries or technologies. Having a dedicated, interactive research mode would help me efficiently gather and organize important contextual information directly related to my tasks or subtasks." --- ### Motivation Users frequently need to explore, clarify, and document research related to specific tasks or subtasks. Currently, there is no easy and organized way to conduct deep, iterative research and seamlessly incorporate findings into task descriptions. Providing an interactive sandbox for research will empower users to deeply investigate topics and efficiently store research insights within task contexts. --- ### Proposed Solution Implement an interactive sandbox research command (`research`) using Inquirer, integrated with Perplexity for deep contextual research: - `task-master research [--id=<task|subtask_ids>]` This command enters an interactive chat environment where users can iteratively research topics: - If `--id` is provided, the specified task/subtask contexts are included in the Perplexity deep research call. - User inputs an initial prompt, Perplexity returns detailed research results. - Users can continue iterating and refining their research through chat prompts. Interactive commands within the sandbox (persistent menu): - `c` - Continue chatting (send another research prompt). - `spacebar` - Autoplay mode: - Automatically formulates and sends a logical next research prompt to continue exploration. - Allows rapid iterative exploration. - `q` - Quit sandbox mode, prompts user to save the research session. - `s` - Save the current research session to the `research/` directory: - Creates a standardized research report file named `research_<taskIds>_<timestamp>.txt`. - Includes original prompt, context task IDs, full conversation history, and summarized findings. - `u` - Update one or more tasks/subtasks with research insights: - User specifies task/subtask IDs. - Appends summarized research information (temperature 0 for maximum detail retention) in XML tags `<research on timestamp>` to each task description. - `x` - Complexity Level: - Persistent menu with three options: MVP, Balanced, Engineered. - Adjusts Perplexity system prompts to shape the research complexity accordingly. - `q/e` - Temperature Control: - Adjust the research creativity with 10 steps (0–10), overriding the default. - `q` decreases temperature (less creative), `e` increases temperature (more creative). - `up/down arrows` - Traverse received AI responses: - Highlights messages for focused review. - Ability to regenerate highlighted responses. - `r` - Regenerate response: - Double-tap confirmation required (prompts user clearly before action). --- ### Enhanced Additional Features - **Contextual Prompt Suggestions**: Automatically suggests prompts based on conversation flow. - **Research Session Snapshots**: Periodic snapshots for easy rollback during extensive sessions. - **Interactive Session History View**: - Command `research-report` lists available research reports in `research/` folder. - User selects a report to display its information clearly in the CLI. - Option provided to re-enter and continue the research session from the report. - **Confirmation Prompts**: Clear confirmations before critical actions (saving, quitting, updating tasks). --- ### High-Level Workflow 1. Initiate sandbox with optional task/subtask context. 2. Collect initial research prompt from user and query Perplexity. 3. Display detailed response and interactive command menu using Inquirer. 4. Support iterative chat research sessions with continued queries. 5. Implement convenient saving, updating, autoplay functionalities, and enhanced interactive controls. 6. Provide clear, intuitive UX through the interactive Inquirer interface. --- ### Key Elements - Command syntax: ```shell task-master research --id=5,10 ``` - Interactive Inquirer-based UX for smooth research workflow. - Flexible research context and prompt management. - Easy integration of research findings directly into task descriptions. --- ### Implementation Considerations - Seamless integration with Perplexity for accurate deep research. - Interactive and intuitive UX using Inquirer for ease of use. - Robust summarization and research documentation in task context. - Efficient handling of multiple task/subtask contexts. - Clear CLI and file management conventions for saving and updating. --- ### Out of Scope (Future Considerations) - GUI-based research sandbox