JTBD Analysis Skill

Run Jobs-to-be-Done analysis on call transcripts using Tim Lockie's structured framework, with UXinator Expectation Mapping and coaching series tracking. This skill powers both the jtbd-analysis command and the standalone jtbd-analysis skill.

Quick Reference

• Analysis Prompt: Jobs To Be Done/Jobs 2B Done - Plugin & Skills/JTBD-Analysis-Prompt.md
• Synthesis Prompt: Jobs To Be Done/Jobs 2B Done - Plugin & Skills/JTBD-Synthesis-Prompt.md
• JTBD Analyses Notion DB: ID 2f218faa725b41828194e8fc0f93453b, data source collection://fbf274fd-5cf0-4afe-9eaf-cb511cae6b94
• Database Schema: Inline in commands/jtbd-analysis.md Step 6 (authoritative source for all valid property values)
• Notion Meeting Transcripts DB: ID 8368d3474cac4e71bf945934fce957f7, collection 669e7e0b-dfe6-43c4-b4c3-d7b734e06ed5
• HubSpot Portal: 22283601
• UXinator Skill: uxinator:expectation-mapper (runs against raw transcript)

Transcript Sources

1. Pasted text — use directly
2. ~~conversation intelligence URL (Fathom, Fireflies, Gong) — fetch via WebFetch
3. ~~knowledge base Meeting Transcript — fetch via notion-fetch or search via notion-search with data_source_url: "collection://669e7e0b-dfe6-43c4-b4c3-d7b734e06ed5"

Folder Structure

All analyses use a single deterministic path -- never ask the user which folder:

/Users/tim/Dev/claude-cowork/Clients/[Organization Name]/
├── Meetings/
│   ├── Transcripts/        ← raw transcripts
│   └── Analysis/           ← JTBD analysis files
└── Synthesis/              ← cross-call synthesis reports

Create the directory structure if it doesn't exist.

Core Workflow

1. Get transcript from one of the three sources
2. Extract company name, participants, call date
3. Determine series context — query JTBD Analyses DB for same Organization to get session number and link to previous session (see command Step 2b)
4. Create directory — path is always /Users/tim/Dev/claude-cowork/Clients/[Org Name]/Meetings/Analysis/ (never ask, fuzzy-match existing org folders)
5. Read JTBD-Analysis-Prompt.md fresh every time (it evolves)
6. Run full 9-dimension analysis per the prompt
7. Build CONNECTIONS section at top of output:
   • Fathom Recording link
   • Notion Meeting Notes link
   • HubSpot Account (search ~~CRM companies, portal 22283601, record type 0-2)
   • HubSpot Contacts (search ~~CRM contacts, record type 0-1)
   • HubSpot Deal (search ~~CRM deals, record type 0-3)
   • Clay URL (if available)
8. Run UXinator expectation-mapper against the raw transcript (NOT the JTBD analysis) and append the full output as the final ## EXPECTATION MAP section
9. Save to /Users/tim/Dev/claude-cowork/Clients/[Org Name]/Meetings/Analysis/[filename].md
10. Populate JTBD Analyses DB — create a page in the Notion database (data source collection://fbf274fd-5cf0-4afe-9eaf-cb511cae6b94) with:
    • Page body: The COMPLETE analysis text, verbatim — all sections including SERIES and EXPECTATION MAP (never summarize or truncate)
    • Properties: Using exact valid values from command file Step 6b
    • Meeting Transcript: Two-way relation to source transcript (auto-creates back-link)

Output Sections (in order)

1. CONTEXT METADATA (includes Plugin Version: 5.0.0 at bottom)
2. SERIES (session number, previous session link, JTBD evolution)
3. PRIMARY JTBD
4. DETAILED JTBD ANALYSIS
5. SWITCH TRIGGER
6. DESIRED OUTCOME
7. OBSTACLES & ANXIETIES
8. MESSAGING GOLD
9. AUDIENCE SEGMENT
10. IP & FRAMEWORK APPLICATION
11. PRODUCT/OFFERING IMPLICATIONS
12. PATTERN RECOGNITION
13. CONNECTIONS
14. EXPECTATION MAP (from UXinator, run against raw transcript)

File Naming

• Single participant: [YYYY-MM-DD] - [First Last, Company] - JTBD Analysis.md
• Series session (Session 2+): [YYYY-MM-DD] - [First Last, Company] - Session [N] - JTBD Analysis.md
• Multiple from same org: [YYYY-MM-DD] - [Company] - JTBD Analysis.md

Organization Folder Matching

• Check existing folders under /Users/tim/Dev/claude-cowork/Clients/
• Fuzzy match at >=90% similarity (case-insensitive)
• If match exists, use that folder name exactly
• If no match, create new org folder with full directory structure

Analysis Rules

• Specific over generic
• Quote exact language from transcript
• Mark inferences with [INFERRED] or [IMPLIED]
• Flag contradictions (stated vs. observed)
• Intensity = importance
• Context matters (prospect vs. 90-day client)

Notion DB Population

After saving the .md file, always create a JTBD Analyses record. The database serves as the queryable, filterable index across all analyses. The .md file remains the authoritative full-text source.

Use notion-create-pages with:

parent: { data_source_id: "fbf274fd-5cf0-4afe-9eaf-cb511cae6b94" }

CRITICAL -- Page body content comes first. The complete, unmodified analysis text (the exact .md file content) MUST be pasted as the page body. Do NOT summarize, truncate, or paraphrase. Every section, every quote, every line -- verbatim. This includes the SERIES section and the EXPECTATION MAP section.

Full-text fallback: If MCP tools truncate content, use the Notion API directly via curl with $NOTION_API_KEY to append missing blocks. See command Step 6a for details.

Property mapping: See commands/jtbd-analysis.md Step 6b for the complete field-by-field mapping with exact valid values for every select/multi-select field.

Meeting Transcript is a two-way relation to the Meeting Transcripts DB (collection://669e7e0b-dfe6-43c4-b4c3-d7b734e06ed5). Setting this relation automatically creates a back-link visible on the transcript page under "JTBD Analyses".

Multi-select values must match exactly — only use values listed in Step 6b. Omit rather than guess or create unknown options.

Dispatch Architecture

This skill dispatches as a background agent. The main conversation handles only transcript acquisition (Step 1) and company/folder identification (Step 2) in the foreground -- these may require user input. Everything else (Steps 2b-6) runs as a single background agent:

1. Foreground: Acquire transcript, extract company/participants/date (no user input needed for path -- it's deterministic)
2. Background agent: Series context, JTBD analysis, CRM lookups, UXinator expectation map, file save, Notion push

See commands/jtbd-analysis.md Dispatch Architecture section for the full agent prompt template.

Background rules:

• Do not stop for Bash tool errors -- retry once, then log and continue.
• Do not prompt about file permissions or directory creation.
• If MCP Notion tools fail, fall back to curl-based API calls silently.
• Skip the user review step and save directly.

Backwards Compatibility & Batch Upgrade

Every analysis file carries an implicit version based on which sections are present. When the user says "upgrade all", "update all", or "backfill":

1. Scan all files in Meetings/Analysis/ under Clients/ (and legacy Jobs To Be Done/ paths)
2. Detect each file's version by checking which sections exist (SERIES, EXPECTATION MAP, CONNECTIONS, etc.)
3. Report grouped counts of what's missing
4. Launch background agents to backfill: SERIES (from DB query), EXPECTATION MAP (from raw transcript via expectation-mapper), CONNECTIONS (from CRM lookups), Notion DB records
5. Update Notion page bodies with full text after upgrade
6. Log results to Clients/_upgrade-log.md

See commands/jtbd-analysis.md Step 7 for the full version detection table and upgrade procedures.