Question 1

How to fix Cloudflare 530 and dashboard 502 after a server reboot when systemd user services do not automatically restart?

Accepted Answer

The outage occurs when the local proxy is healthy but the cloudflared tunnel and observability/dashboard services remain dead after reboot. `Restart=on-failure` does not restart services after a clean reboot/SIGTERM, and the deploy cron only restarted the proxy.

Immediate recovery on gpu01 as the service user:
```
systemctl --user start k-skill-proxy-tunnel k-skill-proxy-grafana k-skill-proxy-loki k-skill-proxy-promtail
systemctl --user is-active k-skill-proxy-tunnel k-skill-proxy-grafana k-skill-proxy-loki k-skill-proxy-promtail
curl -fsS http://127.0.0.1:8080/health
```
Then verify the public proxy returns 200/302 and the dashboard no longer returns 502.

Permanent fix:
1. Commit the missing tunnel unit to the repository and include the NFS boot-race dependency:
   ```ini
   [Unit]
   StartLimitIntervalSec=0
   RequiresMountsFor=/data/...

[Service]
   Restart=always
   ```
2. Change all units under `infra/k-skill-proxy-dashboard/systemd/*.service` from `Restart=on-failure` to `Restart=always`.
3. Reload and enable the updated units:
   ```
   systemctl --user daemon-reload
   systemctl --user enable --now k-skill-proxy-tunnel k-skill-proxy-grafana k-skill-proxy-loki k-skill-proxy-promtail
   ```
4. Add a watchdog to the 5-minute deploy cron so dead services are started even when the deployed SHA has not changed:
   ```bash
   systemctl --user is-active --quiet \
     k-skill-proxy-tunnel k-skill-proxy-grafana k-skill-proxy-loki k-skill-proxy-promtail \
     || systemctl --user start \
     k-skill-proxy-tunnel k-skill-proxy-grafana k-skill-proxy-loki k-skill-proxy-promtail
   ```
5. Run the public `/health` smoke test on every cron execution, regardless of `deployed-sha`, and alert or create an incident issue on failure.
6. As additional hardening, set `KSKILL_PROXY_HOST=127.0.0.1` instead of `0.0.0.0`, and upgrade cloudflared from `2026.6.1` to `2026.8.2`.

Validation: reboot gpu01, then confirm all five services return `active` and both `https://k-skill-proxy.nomadamas.org` and `https://k-skill-proxy-dashboard.nomadamas.org` return 200/302 without manual intervention.

Question 2

How do I fix k-skill-proxy real-estate 502 errors caused by MOLIT returning HTTP 200 with resultCode 22 quota exceeded?

Accepted Answer

These 502s are semantic upstream errors, not retryable transport failures. MOLIT returns HTTP 200 with an XML error body where `resultCode != 000`; `molit.js` converts it to `molit_api_<code>`, and `server.js` maps all such results to 502. `fetchWithRetry` retries only HTTP 429/5xx, so resultCode 22 (`LIMITED NUMBER OF SERVICE REQUESTS EXCEEDS`) is not retried successfully. Deploy the fix from PR #621, or apply the same mapping: classify resultCode 22 as `503 upstream_quota_exceeded` with `Retry-After: 3600` and do not retry it; classify key/application errors such as 20, 30, and 31 separately and log them at error level; keep 502 only for genuine transient upstream failures. In the real-estate and krx-stock routes, add `request.log.warn({ route, upstreamError: result.error, message })` before returning 502. Add a `server.test.js` scenario for resultCode 22 and run the test suite. After deployment, quota failures appear as attributable 503 responses instead of anonymous 502s.

Question 3

korean-privacy-terms에서 한글 이용약관 템플릿이 0바이트라 약관이 생성되지 않을 때 어떻게 해결하나요?

Accepted Answer

원인은 고정된 업스트림 커밋 `e390f7b9feb825e368c26726363ea5ce11a34083`에서 한국어 이용약관 템플릿 2개가 빈 파일로 포함된 것입니다.

먼저 빈 템플릿을 확인하세요:
```bash
find ~/.claude/skills/korean-privacy-terms -name '*.tmpl' -size 0
```

임시 해결 방법은 영문 템플릿을 구조 참고용으로 사용해 한국어 약관을 직접 작성하는 것입니다:
```bash
base=~/.claude/skills/korean-privacy-terms/upstream

# 필요한 경로에 한국어 약관 파일을 직접 생성
vi "$base/jurisdictions/kr-pipa/terms-of-service.ko.mdx.tmpl"
vi "$base/templates/terms-of-service.mdx.tmpl"
```
작성 시 같은 디렉터리의 `terms-of-service.en.mdx.tmpl` 또는 `templates/terms-of-service.en.mdx.tmpl`의 변수와 레이아웃은 유지하고, 법률 문구는 한국어로 검토해 교체해야 합니다. 작성 후 스킬 렌더링을 다시 실행하세요. 개인정보처리방침 템플릿은 정상이므로 해당 산출물에는 영향이 없습니다.

병합된 `d7f45d4`는 `install.sh`에 0바이트 템플릿 경고와 문서화를 추가했을 뿐, 빈 한국어 템플릿 자체를 채우지는 않습니다. 근본 해결은 업스트림 `kimlawtech/korean-privacy-terms`에 한국어 템플릿이 추가된 뒤 `scripts/upstream.pin`을 해당 커밋으로 올려야 합니다.

Question 4

How to fix the CGV bot-block error or '503 CGV_UPSTREAM_UNAVAILABLE' in korean-movie-search?

Accepted Answer

This is caused by CGV's upstream bot detection/access restriction, not by an invalid skill or CLI input. First verify the service with `npx --yes daiso health`. If health is normal but CGV endpoints return `503 CGV_UPSTREAM_UNAVAILABLE`, wait and retry later. Do not bypass CAPTCHA, browser verification, or access controls; proxies, user-agent spoofing, and automated CAPTCHA solving are not supported. If a different error occurs, recheck the movie/theater identifier and CLI syntax. Persistent failures require waiting until CGV removes the restriction.

Question 5

How can I search for animal pharmacies selling specific veterinary medicine in Korea via MCP?

Accepted Answer

You can use the Streamable HTTP MCP server at `https://hkmedi.co.kr/pharmacy-mcp`, which has also been packaged as the `animal-pharmacy-search` skill in the K-Skill repository (merged via PR #630). This tool supports: 1. Browsing regional lists of Korean animal pharmacies, 2. Searching veterinary medicinal products by keyword or efficacy, 3. Filtering results to only pharmacies with a purchase record for the target product in the last 6 months for higher reliability. Note: This data is sourced from private commercial distributor Hongik Medicare, not official public government data. It does not guarantee real-time in-stock status or complete nationwide coverage, and is not a replacement for professional veterinary diagnosis or prescription. Always call the pharmacy to confirm product availability before visiting.

k-skill

Overview

README Preview

FAQ (5)

同类型项目

tensorflow

superpowers

everything-claude-code

flutter