Fan-built agent skills for BTS fans. English-first, with Korean summaries.
팬이 만든 BTS 팬용 에이전트 스킬 모음입니다. 영어가 기본이며 한국어 요약을 함께 제공합니다.
Unofficial fan project. Not affiliated with or endorsed by HYBE, BIGHIT MUSIC, BTS, Weverse, or any ticketing company. Non-commercial: it never sells tickets, takes payments, or holds accounts. Always confirm on official sources before paying anyone. Full notice: DISCLAIMER.md
비공식 팬 프로젝트입니다. HYBE, BIGHIT MUSIC, BTS, Weverse 및 어떤 티켓 판매처와도 관련이 없으며 승인받지 않았습니다. 비상업적 프로젝트로 티켓 판매, 결제, 계정 보관을 하지 않습니다. 돈을 보내기 전에 반드시 공식 채널에서 확인하세요.
Status: v0.2.0 (2026-09-09) is the current release. It ships both
skills - bts-ticket-safety and bts-official-news - plus the executable
tools in tools/ and the CI workflows. v0.1.0 (2026-09-08) contained
bts-ticket-safety only.
One command loads both skills into your agent through the skills CLI (Claude Code, Codex, Cursor, and more):
npx skills add helenanova/sevensignalNew here? The 5-minute quickstart has copyable prompts with expected outputs, and there is a matching static demo for the official-news skill.
A collection of open-source skills that an AI agent can load to help BTS fans with practical, safety-critical tasks. The first skill focuses on the highest-risk moment in fandom life: buying concert tickets without getting scammed.
The name: "seven" is a nod to the seven members. "Signal" is what the skills do - separate trusted signals (official announcements, official sellers) from noise (scams, rumors, resellers). The codename deliberately avoids "BTS" and "ARMY" to keep trademark risk low.
- Official sources only. Every trust decision anchors on a deterministic allowlist of official domains, not on vibes or search rankings.
- Deterministic over clever. Domain matching is exact string logic that anyone can audit. No model judgment calls on what "looks official".
- Advise, never transact. These skills help a fan decide. They never buy tickets, hold credentials, or move money.
- Freshness is a feature. Tour dates, onsale times, and prices go stale. Claims that expire are marked for re-verification, with the official place to check.
- Refusal is a valid answer. Out-of-scope requests get a clear refusal plus a safe redirect, not a half-answer.
sevensignal/
LICENSE MIT
README.md this file
DISCLAIMER.md unofficial-project and no-affiliation notice
SECURITY.md how to report security and trust problems
CONTRIBUTING.md content and evidence rules for contributors
CODE_OF_CONDUCT.md contributor covenant
CHANGELOG.md release history
MAINTAINERS.md the weekly maintenance loop and unmaintained policy
docs/ demo GIF and social preview image (no product code)
.github/ issue forms, pull request template, CI workflows
tools/ executable checks (matcher, validators, eval runner)
skills/
bts-ticket-safety/
SKILL.md the skill definition (English + full Korean guidance)
data/
official-domains.yaml deterministic allowlist of official domains
references/
scam-patterns.md catalog of known ticket-scam patterns
official-channels.md where official announcements actually appear
evals/
README.md eval schema and how to run them
golden.yaml legitimate scenarios -> expected trust verdicts
adversarial.yaml scam scenarios -> expected scam verdicts
refusal.yaml out-of-scope asks -> expected refusals
freshness.yaml claims that expire -> re-verification rules
bts-official-news/
SKILL.md official tour/news tracking, public pages only
evals/ golden, adversarial, refusal, freshness suites
The domain-matching rules are executable, and the trust data carries its own checks:
python3 tools/test_domain_matcher.py # unit tests (no dependencies)
python3 tools/validate_data.py # schema + 30-day staleness check
python3 tools/eval_runner.py # deterministic eval assertions
python3 tools/check_links.py # live source-link checkSee tools/README.md for details. CI runs the first three on every change;
a weekly workflow re-checks data freshness and source links and opens one
stale-data issue when anything needs re-verification.
한국어 요약: 위 명령으로 도메인 일치 규칙과 신뢰 데이터를 직접 검증할 수 있습니다. 데이터가 30일 이상 오래되면 자동으로 경고가 열립니다.
- Fan community safety (Discord/X impersonation detection) - deliberately deferred: impersonation judgment carries real defamation risk and needs more design before it ships as a deterministic skill.
- v0.2.0 after community validation: soft-launch feedback triaged, field
reports turned into fixtures, all
last_verifieddates re-checked.
MIT. Home: https://github.com/helenanova/sevensignal
