Two Claude Code skills that turn desk research, chat corpora, interview transcripts and notes into a full job list — segments, jobs in When / Want to / So that form, a link graph, product-fit ratings and a Next Move Theory segment map — built on Ivan Zamesin's Advanced JTBD canon, not generic Christensen JTBD. One renders it to a FigJam board. The other renders one self-contained HTML file that needs no account.
You hand it research. It extracts every customer job it can find, labels each one by its source — a direct quote, the research author's hypothesis, or an AI assumption — and sorts jobs into segments defined by the job itself, not by demographics.
Everything lives in one file, work/jobs.json. The job-list document, the FigJam
board and the HTML page are all views generated from it — never the other way around. Edit the
file, re-render; nothing drifts out of sync.
Honesty is the whole point. A made-up "fact" is worse than a blank. A field with no source becomes a "—", not a guess.
Method: Advanced Jobs To Be Done +
Next Move Theory, the canon by Ivan Zamesin — distinct from Clayton Christensen's generic JTBD.
A job is When (context, trigger, negative emotions, consideration set) /
Want to (one infinitive, success criteria) / So that (one higher-level
job, positive emotions).
Stages run in order; the skill stops at three checkpoints for you to review before building on your edits.
Install one, or both — the right person to ask is whoever's about to use it.
/jtbd-research-to-figjam | /jtbd-research-to-html | |
|---|---|---|
| Output | a FigJam board | one self-contained HTML file |
| Needs | the Figma MCP, a Figma account | nothing but Python 3 and a browser |
| Editing the graph | drag cards by hand, skill reads edits back | edit jobs.json, re-render in under a second |
| Collaboration | several people on one board | share the file, or print to PDF |
| Works offline | no | yes |
jtbd-extractor — reads one research source, pulls out every job it can find, returns a jobs.json fragment with evidence quotes. Run once per source, in parallel.jtbd-auditor — reviews the merged file: checks every field against the evidence, flags anything unsupported, catches Fake Job mislabels and missed Tax Jobs.Read the code before you run it against your own research — it's MIT-licensed, 38 KB of Markdown and Python, nothing hidden.
Installs both skills and both sub-agents. Update later with
/plugin update jtbd-job-mapper@jtbd-job-mapper.
Downloads both skills and both agents straight into ~/.claude/.
Same result, from your own clone.