Agent 01 — Transcript

Processes raw project briefings, meeting notes, or conversation transcripts into structured JSON.

File: agents/sdlc/01-transcript.md · Model: Sonnet · Tools: Bash, Read, Write

Purpose

The transcript agent converts raw, unstructured input — meeting notes, product briefs, conversation logs — into a clean structured JSON that all downstream agents can rely on. It surfaces ambiguities rather than silently resolving them.

What it does

1. Reads raw project input (meeting transcript, brief, or notes)
2. Reads environment_report.json
3. Extracts: project name, goals, stakeholders, constraints, timeline, key decisions
4. Flags open_questions for ambiguous items — never fills gaps with assumptions
5. Produces transcript.json

Core principle

Preserve signal, surface ambiguity. A correctly flagged open question is worth more than a confident wrong answer.

Outputs

// .rstack/runs/<run_id>/transcript.json
{
  "project_name": "Customer Portal v2",
  "goals": ["Migrate legacy portal to React", "Add SSO support"],
  "stakeholders": [
    { "name": "Sarah K.", "role": "Product Owner" }
  ],
  "constraints": ["Launch by Q3", "Must support IE11"],
  "timeline": "12 weeks",
  "key_decisions": ["Use React + TypeScript", "Auth via Okta"],
  "open_questions": [
    "Which database? PostgreSQL vs MySQL not specified.",
    "Is IE11 support mandatory or best-effort?"
  ]
}