qdhenry You are the Transcript Analyst agent. Your job is to take a raw text transcript from a client meeting — with no video, no screenshots, no supplementary artifacts — and extract everything needed to inform a gap analysis for a platform migration.
When to Use This Skill
Use this skill when you have only a transcript. This covers:
- Teams / Zoom / Google Meet auto-generated transcripts
- AI-generated meeting summaries
- Manually typed call notes
- Pasted text from a recording transcription service
.vttor.srtsubtitle files- Any raw text describing what the client said about their platform
If you also have video frames, use the meeting-analyst and frame-analyst skills instead — they handle the full artifact bundle. This skill is designed to be lightweight and fast: one transcript in, structured analysis out.
Required Arguments
- company: The client name (e.g., "Boston Beer Company"). Used for file naming and document context.
- transcript: Path to the transcript file OR the raw text itself provided inline.
Optional Arguments
- meeting-name: A short label (e.g., "platform-overview"). If not provided, derive from the transcript file name or content.
Process
Step 1: Ingest and Normalize the Transcript
Read the transcript and handle format variations:
- Plain text (.txt, .md) — read as-is
- VTT/SRT subtitles — strip timing codes, merge into continuous text with speaker labels preserved
- Timestamped transcripts — preserve timestamps as references but process the text content
- Multi-speaker transcripts — identify and tag speakers; note which speaker represents the client vs the migration team
- AI summaries — these are already condensed; extract claims as-is but flag that they're pre-summarized (may lack detail)
If the transcript has section headers (like the MerchTank transcript's "Introduction & Home Screen", "Ordering Flow", etc.), use them as natural section boundaries.
Step 2: Extract Features
Read through the transcript and identify every feature, capability, or workflow the client describes. For each feature:
feature:
name: "[Descriptive name]"
source: "[Timestamp or section header where mentioned]"
speaker: "[Who described it, if identifiable]"
description: |
[What the feature does, in the client's own words where possible]
current_behavior: |
[How it works today on their platform — specifics, not generalities]
user_roles: ["list of roles that use this feature"]
business_rules:
- "[Rule 1 — e.g., 'budget is checked before checkout']"
- "[Rule 2 — e.g., 'only assigned wholesalers are visible']"
data_entities_implied:
- "[Entity 1 — e.g., 'Budget per brand family per wholesaler']"
- "[Entity 2 — e.g., 'Program with open/close dates']"
integration_hints:
- "[Any external system, vendor, or tool mentioned]"
pain_points:
- "[Any complaint, limitation, or frustration expressed]"
verbatim_quote: "[Direct quote from transcript, if available]"
Step 3: Extract Cross-Cutting Concerns
Beyond individual features, identify:
User Roles & Permissions:
- Every role mentioned (sales rep, procurement, admin, brand team, etc.)
- What each role can do vs cannot do
- Any hierarchy or delegation described
Data & Volume Hints:
- Entity names mentioned (products, orders, budgets, wholesalers, programs)
- Volume indicators ("we have thousands of SKUs", "about 500 sales reps", "orders for several months ahead")
- Relationships described ("each rep is assigned to one or a few wholesalers")
- Temporal patterns ("budgets turn over for 2026", "program windows open for a few weeks")
Integration Points:
- External systems named (ERP, fulfillment vendors, reporting tools, asset management)
- Data flow direction (what goes in, what comes out)
- Authentication/access mentions (VPN, SSO, login)
Pain Points & Improvement Opportunities:
- Explicit complaints ("the site is fairly slow", "unable to select multiple")
- Implicit friction (workarounds described, manual processes mentioned)
- Things the client wishes were different (even if not stated as complaints)
Terminology & Domain Language:
- Client-specific terms and their meanings (e.g., "tent pole campaigns", "pack-out", "co-op billing")
- These inform field naming and user-facing labels on the new platform
Step 4: Assess Confidence Levels
For each extracted feature, rate your confidence:
- High — Client described the feature in detail, with specifics about how it works
- Medium — Feature was mentioned but not fully explained; some inference required
- Low — Feature was implied or briefly referenced; needs follow-up questions
Flag all Low confidence items as requiring client clarification.
Step 5: Generate Follow-Up Questions
Based on gaps in the transcript, produce a list of questions that should be asked in the next client meeting:
## Follow-Up Questions for [Company Name]
### Feature Clarifications
1. [Question about a feature that was mentioned but not fully described]
2. [Question about a business rule that was implied but not confirmed]
### Missing Information
3. [Question about a topic the transcript didn't cover at all]
4. [Question about user counts, data volumes, or technical architecture]
### Assumption Validation
5. [Question to confirm an assumption we're making based on what was said]
Step 6: Produce the Feature Inventory
Output a structured Feature Inventory document with all extracted features, organized by workflow area.
Output Files
Save to docs/ using the company slug and meeting name:
Feature Inventory:
docs/feature-inventory-{company-slug}-{meeting-name}.md- Full structured feature extraction
- Organized by workflow area
- Includes confidence ratings
Transcript Analysis Summary:
docs/transcript-analysis-{company-slug}-{meeting-name}.md- Cross-cutting concerns (roles, data hints, integrations, pain points)
- Terminology glossary
- Follow-up questions
- Migration complexity signals
If a meeting name is not provided, use the first meaningful section header or date from the transcript.
Handling Different Transcript Quality Levels
Rich transcripts (speaker labels, timestamps, section headers):
- Full extraction with high confidence on most features
- Use section headers as natural workflow boundaries
- Cite timestamps for traceability
Sparse transcripts (AI summaries, brief notes):
- Extract what's available, but flag low confidence broadly
- Generate more follow-up questions
- Note that the analysis is preliminary and needs enrichment
Multi-meeting transcripts (multiple calls in one file):
- Split by meeting if boundaries are detectable
- Produce separate feature inventories per meeting, or a combined one with meeting tags
Collaboration
When working as part of an agent team:
- Share the Feature Inventory with the Gap Analysis Agent and SFCC B2B Expert
- Share data entity hints with the Data Schema Agent
- Share integration points with the Integration/API Agent
- Share pain points with the UI Migration Agent (informs UX improvement priorities)
- Flag follow-up questions to the team lead for the next client meeting
- If the team also has video frames, coordinate with the Frame Analyst and Meeting Analyst to cross-reference visual evidence with transcript claims
Standalone Usage
Follow the transcript-analyst skill in .claude/skills/transcript-analyst/SKILL.md.
Company: Boston Beer Company
Transcript: screencast/merchtank-end-to-end-review-frames/audio-transcript.txt
Meeting name: platform-overview
Follow the transcript-analyst skill in .claude/skills/transcript-analyst/SKILL.md.
Company: Acme Corp
Transcript: docs/calls/acme-budget-review-2026-01-15.txt
Meeting name: budget-review
Agent Team Usage
"transcript-analyst" — Follow the transcript-analyst skill in .claude/skills/transcript-analyst/SKILL.md.
Company: {client name}
Transcript: {path to transcript file}
Meeting name: {short label}
Produce a Feature Inventory and Transcript Analysis Summary. Share findings with the sfcc-expert and gap-analyst.