메뉴
Graduate a common-mistakes.md item to docs/learnings/ as a permanent record. Use when the user says "graduate", "graduate mistake", "common-mistakes doldu", or the list reaches 10 items.
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
| name | graduate |
| description | Graduate a common-mistakes.md item to docs/learnings/ as a permanent record. Use when the user says "graduate", "graduate mistake", "common-mistakes doldu", or the list reaches 10 items. |
| argument-hint | <item-number or keyword> |
| allowed-tools | Read, Edit, Write, Glob |
| disable-model-invocation | true |
Target item: $ARGUMENTS
Read .claude/rules/common-mistakes.md. Show the current list.
From $ARGUMENTS:
If the list has fewer than 8 items, ask: "The list isn't full yet — are you sure you want to graduate this now, or is it no longer relevant?"
Read docs/learnings/README.md if it exists to understand the expected format.
Before I graduate "<item>", a few things:
1. Why is this being graduated?
a) It's fully resolved — we fixed the root cause
b) It happened enough times that it needs a proper write-up
c) It's no longer relevant to this project
2. Is there a specific incident or PR where this came up?
(optional — helps future readers understand the severity)
3. Any additional detail beyond what's in the mistake entry?
Wait for answers.
Use today's date. Slug = kebab-case version of the mistake title.
# <Mistake title>
**Graduated**: <today's date>
**Source**: common-mistakes.md
## The mistake
<Original text from common-mistakes.md>
## Why it matters
<User's answer from Step 4 — what went wrong, what was the impact>
## Root cause
<What causes this mistake to happen — be specific>
## How to avoid it
<Concrete steps or pattern to prevent recurrence>
## Reference
<Incident, PR, or session where this was observed — if provided>
Edit .claude/rules/common-mistakes.md:
Graduated: "<mistake title>"
Written to: docs/learnings/YYYY-MM-DD-<slug>.md
Removed from: .claude/rules/common-mistakes.md
Remaining items: <N>/10
The learning is now in Tier 2 (docs/learnings/) — loaded when relevant,
not every session. This frees up a slot in common-mistakes.md.
Analyze accumulated feedback memories and turn recurring patterns into permanent controls (rules, permissions, hooks) or common-mistakes entries. Use when the user says "improve", "iyileştir", "apply feedback", "kurallaştır", or "turn feedback into rules".
Project onboarding and template setup. Use when the user says onboard, setup, initialize, init, configure this project, or first-time setup. Guides through detecting project mode and stack, configuring CLAUDE.md, permissions, hooks, rules, and architecture docs.
Save session learnings, patterns, and corrections to persistent memory. Use when the user says "save learnings", "save memory", "remember this session", or at the end of a productive session.
Add a new Claude Code control following project conventions. Use when the user says "new rule", "add rule", "create rule", "yeni rule ekle", or wants to enforce behavior consistently. Guides through choosing enforcement surfaces, writing effective rule files, and keeping controls consistent.
Create a new Claude Code skill following project conventions. Use when the user says "new skill", "add skill", "create skill", or "yeni skill ekle". Guides through gathering requirements, writing effective SKILL.md files, and structuring skill directories.
Add a new Claude Code hook — shell script wired into settings.json. Use when the user says "new hook", "add hook", "yeni hook ekle", or wants to automate something that runs on every tool use, session start, or stop.