twenty-partner-application-triage

Run the rank.py script, which pulls all partners from the Twenty REST API, scores APPLICATION-stage records on net-new geography, languages, scope, skills, and proof of real Twenty work, then prints ranked JSON. Review the top notes and output a tiered chase-list.

Scoring adds +3 per net-new geography and language, +1 per new scope, +1 per new skill capped at 3, and +6 for a proof signal in the notes. Scores of 12 or higher, or any proof signal, land in tier A; 5 or higher is tier B.

The script reads TWENTY_PARTNERS_API_URL and TWENTY_PARTNERS_API_KEY from ~/.twenty/credentials.env. Copy the key from the partner app's .env.prod file into that credentials file on first setup.

No. The skill is strictly read-only: it never mutates records, never moves anyone to REJECTED, and never filters anyone out of the pipeline. It only surfaces a ranked chase-list for human follow-up.

The callBookedAt field is not yet on the partner data model, so the ranker cannot detect booking state and scores every application. Once that field exists, it automatically narrows to un-booked applicants without any skill changes.

Run rank.py with the --selftest flag. It asserts that a gap-filler with proof ranks above a skill-sprayer, which ranks above an empty record, and that skill volume stays capped at the configured limit.