working-loop

name	working-loop
description	Audit Socratink idea-to-PR work that moved the learner-facing product needle. Mines Socratink chat snapshots and product git/PRs for real wins and failures, then names skills, tools, and harnesses to reuse. Use when the user is tired of back-and-forth, asks what actually shipped in Socratink, or wants a repeatable product loop—not tooling, Praxist, or chat archaeology.

Socratink working loop

Audit Socratink the product (product/socratink, remote jon-devlapaz/socratink). Ignore work that did not change what a learner can do in Chat or on the learner surface.

Follow PROMPT.md. Read sources.md. First:

python3 .agents/skills/working-loop/scripts/inventory.py

From another cwd:

python3 /Users/jondev/dev/active/socratink/product/socratink/.agents/skills/working-loop/scripts/inventory.py

Do not dump raw chat, learner wording, secrets, tokens, or private endpoints.

Needle (include)

A change moved the needle if a learner would notice it, or if it unblocked that path: Chat boots, replies, questionnaires, instruction/tutor behavior, learner UI, or a live synthetic Chat run in Braintrust when that was the ask.

Not the needle (exclude)

Praxist runs, task init, baselines, DIG, operator skills, chat-history exports, prototypes, archive, Flue framework rewrites, new eval products, and the reverted R1 learner-evidence expansion. Those may explain a failure; they are not success.

Done when

One playbook: default Socratink loop, 3–7 needle wins, 3–7 failures, what to reuse, what to stop. No menu of follow-ups.