ai-agent-system

Blog writing operations: concrete drafts, diagnosis, comment strategy, external sources, and recurring gotchas.

사용자 요청을 Blog, Luna, Darwin, Sigma, Ska, Hub/Orchestrator 등 적절한 팀과 런타임으로 라우팅할 때 사용.

Use for non-trivial production code changes that require expert challenge and consensus, test-first implementation, soft and read-only hard verification, runtime and DB consistency checks, refactoring discipline, and an independent code-review gate.

Use when diagnosing or improving Luna investment-team autonomous operation bottlenecks, especially the 30-minute Codex automation loop, live-fire watchdog issues, LLM hot-path overuse, discovery funnel stalls, marketdata connectivity, reconcile blockers, or safe fix planning.

Hub routing, resilience, alarm lifecycle, and Sigma feedback operations.

Darwin research cycle operations: cycle overview, predicate authoring, adopt review, and recurring gotchas.

Archived n8n workflow operations guidance kept for historical reference; do not use for new active routing.

작업 완료 전 Karpathy 원칙 기반으로 가정, 단순성, 변경 범위, 목표 충족 여부를 자체 검증할 때 사용.

API, MCP, 웹훅, secret, 인증/권한 변경을 OWASP 관점으로 검토할 때 사용.

브라우저 기반 확인, 게시물 레이아웃 검증, 대시보드 실시간 상태 확인을 Playwright/MCP 패턴으로 진행할 때 사용.

신규 전략, 게시 자동화, 운영 변경을 live 전환 전에 shadow/dry-run으로 검증할 때 사용.

복잡한 개발 작업을 구현자, 리뷰어, 검증자로 분리해 진행할 때 사용. 큰 변경, 다팀 영향, 회귀 위험이 있는 작업에 적용.

병렬 개발, dirty main 보호, auto-dev 격리 작업이 필요할 때 git worktree를 안전하게 사용하는 절차.

Use when diagnosing, operating, or standardizing Hub LLM OAuth refresh/reimport incidents for Claude Code OAuth, OpenAI Codex OAuth, Gemini OAuth, Gemini CLI OAuth, Gemini Code Assist service, and Groq pool readiness.

Use when creating, extending, reviewing, or debugging team health reports in this repository, especially when a team should use the shared health engine, expose --json health scripts, integrate with unified ops health, or align launchd, HTTP, file, and DB checks.

Use when extending, reviewing, or debugging the shared intent engine in this repository, especially when working on learned patterns, unrecognized phrase capture, auto-promotion, rollback, team intent reports, or shared intent-core and intent-store behavior.

Use when operating, debugging, or improving the orchestrator and Hub control stack in this repository, especially when working on alert publisher/webhook fanout, legacy mainbot queue flow, orchestrator runtime behavior, unified ops health, critical n8n webhook delivery, or briefing and batch formatting.

TRIGGER when: a sigma analyst proposes a feedback change and causal validity needs verification before Tier 2 auto-application; distinguishing correlation from causation in team metric changes. SKIP: Tier 0/1 observation-only runs, data ingestion, non-sigma contexts.

TRIGGER when: analyzing team data before sigma feedback application; validating data completeness, type correctness, or outlier detection for any sigma team. SKIP: non-sigma teams, raw data ingestion pipelines, schema migration tasks.

TRIGGER when: sigma proposes a Tier 2 feedback change that needs A/B experiment validation; designing statistical tests for team configuration changes. SKIP: Tier 0/1 runs, pure observation, data quality checks, causal analysis.

TRIGGER when: sigma needs to identify which team metrics are most analytically valuable; planning feature engineering for a new team onboarding to sigma; detecting feature drift in existing team analytics. SKIP: raw data validation, causal analysis, experiment design.

TRIGGER when: planning OTel spans for a new sigma directive type; defining metric collection points for sigma v2 agents; setting up alert thresholds for commander or pod operations. SKIP: non-sigma observability, infrastructure monitoring, Dexter/Archer scope.

Use when adding, migrating, reviewing, or debugging alerts and reports in this repository, especially when a producer should move onto the shared reporting-hub pipeline, use the standardized payload schema, or align with reporting health and briefing flows.

Use when operating, debugging, or improving the autonomous marketing system — feedback loop, channel orchestration, revenue correlation, strategy evolution, autonomy gate, and master feedback learning.

Use when operating, debugging, or improving the Ska reservation stack in this repository, especially when working on naver-monitor, Pickko registration flow, Ska commander commands, n8n bridge integration, reservation RAG, or Ska health and launchd behavior.