Documentary Research — Documentary Production Full Pipeline

Workflow

Phase 1: Preparation (Performed Directly by the Orchestrator)

1. Extract from user input:
   • Topic: The subject the documentary will cover
   • Format: Investigative/Character-driven/Historical/Observational
   • Length: 15 min/30 min/60 min/90 min
   • Tone (optional): Objective/Emotional/Tense/Reflective
   • Existing Files (optional): Research materials, proposals, etc.
2. Create the _workspace/ directory in the project root
3. Organize the input and save it to _workspace/00_input.md
4. If existing files are present, copy them to _workspace/ and skip the corresponding Phase
5. Determine the execution mode based on the scope of the request

Phase 2: Team Assembly and Execution

Tasks 3a (interview) and 3b (narration) are executed in parallel. Both depend on Tasks 1 (research) and 2 (treatment).

Inter-team communication flow:

• researcher complete -> Deliver timelines and key facts to story-architect, deliver expert list to interviewer
• story-architect complete -> Deliver scene-by-scene interview needs to interviewer, deliver tone and emotional flow to narrator
• interviewer <-> narrator: Share interview topics to prevent overlap with narration
• fact-checker cross-validates all deliverables. When RED Must Fix is found, request revisions from the relevant agent -> rework -> re-verify (up to 2 times)

Phase 3: Integration and Final Deliverables

1. Verify all files in _workspace/
2. Confirm that all RED Must Fix items from the review report have been addressed
3. Report the final summary to the user:
   • Research brief — 01_research_brief.md
   • Treatment — 02_structure.md
   • Interview guide — 03_interview_guide.md
   • Narration script — 04_narration_script.md
   • Fact-check report — 05_review_report.md

Modes by Task Scale

Using Existing Files: When the user provides research materials, treatments, etc., copy the files to the appropriate location in _workspace/ and skip the corresponding agents.

Data Transfer Protocol

File naming convention: {sequence}_{deliverable_name}.{extension}

Error Handling

Test Scenarios

Normal Flow

Prompt: "Plan a 30-minute documentary about the rise in single-person households and social isolation" Expected Result:

• Research: Statistics (single-person household ratio, social isolation indicators), expert list, multiple perspectives
• Treatment: 3-act structure, 7-10 scenes, emotion curve (empathy -> deepening -> hope)
• Interviews: 5-7 person guide including sociology professor, single-person household member, policy official, etc.
• Narration: Approx. 7,500-word script with visual directions
• Fact-check: Statistical source verification, balance verification

Existing File Utilization Flow

Prompt: "Write a documentary treatment and narration script from this research" + research file attached Expected Result:

• Copy existing research to _workspace/01_research_brief.md
• Treatment Mode + Narration Mode combined: Deploy story-architect + narrator + fact-checker
• Skip researcher, interviewer

Error Flow

Prompt: "Just do documentary research, any social issue topic" Expected Result:

• Switch to Research Mode (researcher + fact-checker)
• Topic unclear, so researcher proposes 3 recent social issues then proceeds
• Fact-check report notes "Treatment/Interview/Narration not generated"

Extended Skills per Agent

Each agent leverages the following extended skills' specialized knowledge to enhance deliverable quality: