메뉴
Promote accepted learning-inbox candidates into the right corpus / guide / sensor layer, then refresh INDEX.json.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Full dual-flywheel audit — Learning (capture from review) + Pruning (remove dead rules). Periodic charter hygiene.
One-time initial charter scaffold — detect stack, seed state, scaffold idioms, classify sensors, build the globs index.
Run the drift sensor on the current diff — fast pre-verify check that loaded charter rules still match what the code is doing.
Triage the code-debt ledger — review the debt sensor's findings and update corpus/state/code-debt.md.
Regenerate .keystone/INDEX.json — the primitive descriptor index every agent reads at session start.
Capture a learning candidate (surprise, incident, review finding) into .charter/learning/inbox/ for later synthesis.
| name | keystone-synthesize |
| description | Promote accepted learning-inbox candidates into the right corpus / guide / sensor layer, then refresh INDEX.json. |
| tools | ["Read","Write","Edit","Grep","Glob","Bash"] |
| model | opus |
The Learning flywheel's promotion step. Walks
.charter/learning/inbox/, promotes status: accepted
candidates to the right layer (guide / corpus / sensor), regenerates
the globs index, and refreshes Cursor projections if present.
Canonical playbook: .charter/actions/synthesize.md.
Open .charter/actions/synthesize.md and execute every
activity in order. Every promotion is a proposed diff — never overwrite
silently. After promotions land, run keystone index so
.keystone/INDEX.json reflects the new shape.
/keystone:learn.No silent overwrites. Propose every diff before applying.