Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/mupengi-bot/mupengism/learning-enginenpx skills add mupengi-bot/mupengism --skill learning-enginegit clone --depth 1 https://github.com/mupengi-bot/mupengismWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/mupengi-bot/mupengism/learning-engine)<a href="https://agentmods.dev/skills/mupengi-bot/mupengism/learning-engine"><img src="https://agentmods.dev/badge/skills/mupengi-bot/mupengism/learning-engine.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00018 | $0.01722 |
| Opus 5 | $0.00009 | $0.00861 |
| Sonnet 5 | $0.00004 | $0.00344 |
| Haiku 4.5 | $0.00002 | $0.00172 |
Grade A, and why
learning-engine scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 249 lines — stays where its author put it; the contents beside it link to each section on GitHub.
learning-engine
시스템이 실수와 성공을 기록하고, 자동으로 패턴을 학습하여 스킬을 개선합니다. "같은 실수 반복 안 함" 원칙을 자동화.
학습 소스
1. memory/errors/
에러 로그에서 실패 패턴 추출
# memory/errors/2026-02-14.md
## 10:30 - insta-post 실패
- 원인: PNG 파일 업로드 → "문제가 발생했습니다" 에러
- 해결: JPG 변환 후 재시도 → 성공
- 교훈: 인스타 게시는 항상 JPG로 변환 후 업로드
2. self-eval 결과
주간 자기평가에서 개선점 추출
# memory/self-eval/2026-W07.md
## 이번 주 실수
- 브라우저 스냅샷 너무 많이 찍음 (토큰 낭비)
- → 개선: exec로 API 직접 호출
## 이번 주 성공
- insta-cli v2로 DM 체크 토큰 95% 절약
3. performance 데이터
성과 추적에서 잘된/안 된 패턴 학습
{
"insight": "오후 7-9시 게시물이 좋아요 +30%",
"rule": "인스타 게시는 19:00-21:00 권장"
}
자동 규칙 생성
학습된 패턴을 규칙으로 변환:
위치: memory/learned-rules/
memory/
learned-rules/
instagram-posting.md
browser-automation.md
api-usage.md
error-recovery.md
규칙 포맷
# Instagram 게시 규칙
## 규칙 #1: 항상 JPG 변환
- **상황**: 인스타에 이미지 업로드
- **실패 패턴**: PNG 파일 → "문제가 발생했습니다"
- **해결**: `convert input.png -quality 92 output.jpg`
- **근거**: 2026-02-10, 2026-02-14 에러 로그
- **적용 스킬**: insta-post, cardnews, social-publisher
## 규칙 #2: 1:1 비율 필수
- **상황**: 인스타 카드뉴스
- **실패 패턴**: 16:9 가로형 → 피드에서 잘림
- **해결**: 1024x1024 정사각형으로 생성
- **근거**: 2026-02-13 형님 피드백
- **적용 스킬**: cardnews, nano-banana-pro
스킬에 규칙 주입
학습된 규칙을 해당 스킬 SKILL.md에 자동 추가:
위치: skills/{skill-name}/SKILL.md
# insta-post
...
## 학습된 교훈
### 이미지 처리
- ✅ 항상 JPG로 변환 (PNG는 에러 발생)
- ✅ 1:1 비율 필수 (1024x1024 권장)
- ✅ 파일 크기 < 8MB
### 타이밍
- ✅ 19:00-21:00 게시 시 참여율 +30%
- ❌ 새벽 시간대 게시 지양
### 자동화
- ✅ exec로 API 호출 (스냅샷 0회)
- ❌ 브라우저 자동화는 최소화
주간 학습 리포트
매주 월요일 자동 생성:
위치: memory/learning/weekly-YYYY-Www.md
# 2026-W07 학습 리포트
## 새로 배운 것 (5개)
1. **인스타 PNG 금지**
- 실수 3회 → 규칙 생성
- 적용: insta-post, cardnews
2. **토큰 절약: exec > 브라우저**
- v1: 스냅샷 5회 → v2: exec 1회
- 95% 절약
3. **게시 최적 시간대**
- 19:00-21:00 좋아요 +30%
4. **브랜드 톤 효과**
- 무펭이 톤 참여율 +40%
5. **에러 복구 자동화**
- browser-dependent 실패 시 → 브라우저 재시작
## 적용된 스킬
- insta-post (규칙 2개)
- cardnews (규칙 1개)
- performance-tracker (인사이트 1개)
## 다음 주 목표
- [ ] A/B 테스트 시스템 구축
- [ ] 자동 복구 패턴 3개 추가
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 249 lines · 18 tokens per session scan A 6d1f2419b2b3
learning-engine is a skill published in the GitHub repository mupengi-bot/mupengism (10 stars, last pushed 2mo ago), licensed MIT. It adds 18 tokens to every session and 1,722 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
hive.note-taking
Maintain a free-form scratchpad of decisions, extracted values, and open questions so context pruning doesn't lose anything you still need.
hive.context-preservation
Proactively extract critical values from tool results into working notes before automatic context pruning destroys them.
Vizra ADK Memory System
Implement persistent memory, session context, and vector memory (RAG) for AI agents.
aspirations-consolidate
Runs session-end consolidation — the hippocampal sleep-replay pass that compresses session observations into long-term memory. Handles micro-hypothesis sweep, encoding queue processing, knowledge debt sweep, tree rebalancing, experience-to-skill mining, skill health check, aspiration archive, user recap, handoff…
felt-sense-checkin
Fires whenever the aspirations-precheck cadence hits 75 completed goals — an autonomous structured 7-lane self-audit that converts the user's proven-high-yield diagnostic question into a routine. Sweeps memory hygiene (including insight curation from agents/{agent}/insights.jsonl — distills unprocessed entries into…
replay
Performs hippocampal replay on resolved hypotheses: compressed sharp-wave review for reconsolidation, reverse-order recency scan, selective encoding-queue replay, category-scoped replay, or domain-transfer bundling. Use whenever the aspirations loop hits the replay cadence, /aspirations-consolidate schedules a replay…