Promote confirmed judgment predictions into lessons #1240

New issue

Open

opened 2026-10-22 22:29:50 +00:00 by jared · 0 comments

jared commented

2026-10-22 22:29:50 +00:00

Owner

Problem

Agents classify each ruling by decision stakes tier: mechanical, low-stakes, or judgment. They call /os-me:predict only for judgment rows. Today a confirmed prediction teaches nothing. The next agent meets the same kind of decision and calls predict again.

When a judgment prediction has high confidence and Jared confirms it or does not reverse it, the deciding reason must become a lesson, a lint or detector lesson text, or a skill instruction. Then the next agent classifies that kind of decision as mechanical or low-stakes and decides it without predict.

Jared's goal, 2026-10-22: "Over time the agents should rely less and less on me and similarly less and less on /os-me:predict as well. That's what all of the lint lessons and detector lessons are about. Giving the AI the instructions and clarity it needs to always make the right choice."

Needs: a way to find confirmed high-confidence predictions in the Store prediction log (~/dev/os-me-store), group them by decision kind, and propose the lesson.

Reproduce

n/a: design ticket

Origin

Trigger: Jared stated this goal in session 207dfa34
Improvised this session: none
Chain: repeated predict calls on one decision kind <- no path from a confirmed prediction to lesson text. DESIGN (os-me predict skill, Store prediction log)
Root candidate: none found; searched predict, prediction log, lesson, promote, confirmed, stakes, mechanical, os-me, detector lesson, Relay
Where: plugins/os-me (predict skill, Store prediction log) and plugins/os-sdlc (fast-batch, crc-refactor lessons)
Session: 207dfa34-e3c6-4e15-9823-00c0208b1d38
Transcript: /home/jared/.claude/projects/-home-jared-dev-cc-os/207dfa34-e3c6-4e15-9823-00c0208b1d38.jsonl

VERDICT: CREATE

## Problem Agents classify each ruling by decision stakes tier: mechanical, low-stakes, or judgment. They call /os-me:predict only for judgment rows. Today a confirmed prediction teaches nothing. The next agent meets the same kind of decision and calls predict again. When a judgment prediction has high confidence and Jared confirms it or does not reverse it, the deciding reason must become a lesson, a lint or detector lesson text, or a skill instruction. Then the next agent classifies that kind of decision as mechanical or low-stakes and decides it without predict. Jared's goal, 2026-10-22: "Over time the agents should rely less and less on me and similarly less and less on `/os-me:predict` as well. That's what all of the lint lessons and detector lessons are about. Giving the AI the instructions and clarity it needs to always make the right choice." Needs: a way to find confirmed high-confidence predictions in the Store prediction log (~/dev/os-me-store), group them by decision kind, and propose the lesson. ## Reproduce n/a: design ticket ## Origin - Trigger: Jared stated this goal in session 207dfa34 - Improvised this session: none - Chain: repeated predict calls on one decision kind <- no path from a confirmed prediction to lesson text. DESIGN (os-me predict skill, Store prediction log) - Root candidate: none found; searched predict, prediction log, lesson, promote, confirmed, stakes, mechanical, os-me, detector lesson, Relay - Where: plugins/os-me (predict skill, Store prediction log) and plugins/os-sdlc (fast-batch, crc-refactor lessons) - Session: 207dfa34-e3c6-4e15-9823-00c0208b1d38 - Transcript: /home/jared/.claude/projects/-home-jared-dev-cc-os/207dfa34-e3c6-4e15-9823-00c0208b1d38.jsonl VERDICT: CREATE

jared added the

plugin/os-sdlc

filed-by/agent

category/investigation

labels

2026-10-22 22:29:50 +00:00

jared referenced this issue

2026-10-23 15:17:29 +00:00

decision-system S9: demotion and the One Thing question #1251

jared referenced this issue

2026-10-23 19:37:30 +00:00

decision-system S9b: the One Thing question at SessionStart and its verdict readback #1278