gbrain-briefing

name

gbrain-briefing

description

Generate entity-aware morning briefing from gbrain — active deals, recent meetings, people in play, and upcoming timelines. Posts a structured Slack thread to briefing", "entity briefing", "who's in play", "active deals", "gbrain morning", "gbrain-briefing", "엔티티 브리핑", "gbrain 아침", "인맥 브리핑", "활성 딜", or wants an entity-centric morning overview. Do NOT use for daily stock analysis (use today). Do NOT use for general web search (use parallel-web-search). Do NOT use for Cognee knowledge graph queries (use cognee).

gbrain Briefing — Entity-Aware Morning Intelligence (v0.10)

Generate a structured entity briefing from gbrain's people, companies, deals, and meetings data. Includes brain health score (0-100), feature availability, autopilot daemon status, and stale compiled truth detection. Designed to run as Phase 8.5 of daily-am-orchestrator.

Prerequisites

• gbrain CLI at ~/.local/bin/gbrain
• PostgreSQL running with gbrain schema initialized
• gbrain MCP server registered in .cursor/mcp.json

Subagent Model Routing

Task Type	Model	Rationale
Exploration / search / file reading	haiku	낮은 비용, read-only 작업
Analysis / implementation / generation	sonnet	균형 잡힌 품질-비용 비율
Architecture / complex reasoning	opus	복잡한 추론이 필요한 경우만

Agent tool 호출 시 반드시 model 파라미터를 지정한다.

Skip Conditions

Set status to "skipped" and exit gracefully when:

• gbrain CLI not found at ~/.local/bin/gbrain
• gbrain doctor --json fails (PostgreSQL unreachable)
• gbrain has < 5 total pages (insufficient data for meaningful briefing)

Procedure

Step 1: Health Check & Brain Status

Run the full v0.10 diagnostic suite. All three commands are independent; run in parallel when possible.

1a. Doctor (connectivity + schema)

~/.local/bin/gbrain doctor --json

Parse output to confirm connectivity and page count. If pages < 5, skip with reason.

1b. Health Dashboard (text)

~/.local/bin/gbrain health

health prints a TEXT dashboard (there is NO --json for health in v0.4.1 — use doctor --json from 1a for structured fields). Parse the text for the brain health score and per-dimension lines (freshness, link density, embedding coverage, citation density). Store what is present; do not fabricate dimensions the dashboard does not emit.

REMOVED: gbrain features and gbrain autopilot status do NOT exist in v0.4.1. There is no feature-flag command and no autopilot daemon. Skip any feature-map or autopilot-status collection — derive capability from doctor/health/stats output only.

1c. Stale Compiled Truths

~/.local/bin/gbrain search "compiled-truth"

For each compiled truth page, compare its updated_at against the updated_at of its source pages (found via gbrain backlinks <slug>). A compiled truth is stale if any source page was updated after it. Collect stale items for the briefing alert.

Step 2: Gather Entity Intelligence

Query gbrain for entity-centric intelligence using the following CLI commands:

2a. Recent Activity (last 7 days)

~/.local/bin/gbrain search "updated recently" --limit 20 --json

2b. People in Play

~/.local/bin/gbrain search "people" --limit 10 --json

Filter results to people/ namespace pages. Extract names, roles, companies, last interaction dates.

2c. Active Deals / Companies

~/.local/bin/gbrain search "companies deals active" --limit 10 --json

Filter results to companies/ and deals/ namespace pages.

2d. Recent Meetings

~/.local/bin/gbrain search "meetings" --limit 5 --json

Filter to meetings/ namespace pages.

2e. Ideas / Decisions

~/.local/bin/gbrain search "ideas decisions" --limit 5 --json

Filter to ideas/ namespace pages.

Step 3: Synthesize Briefing

Compile the gathered data into a structured Korean briefing:

## gbrain 엔티티 브리핑 (YYYY-MM-DD)

### 주요 인물 (People in Play)
| 이름 | 소속 | 최근 컨텍스트 | 마지막 접점 |
|------|------|------------|-----------|
| {name} | {company} | {context} | {date} |

### 활성 회사/딜
| 회사 | 상태 | 최근 변경 |
|------|------|----------|
| {company} | {status} | {date} |

### 최근 미팅
- {meeting title} ({date}) — {key takeaway}

### 주요 아이디어/의사결정
- {idea/decision title} — {summary}

### 🧠 Brain Health
- **종합 점수**: {score}/100 {🟢 ≥80 | 🟡 50-79 | 🔴 <50}
- 신선도: {freshness}/100 | 링크 밀도: {link_density}/100
- 임베딩 커버리지: {embedding_coverage}/100 | 인용 밀도: {citation_density}/100

> `gbrain health`가 텍스트 대시보드로 출력하는 차원만 채운다. 대시보드에 없는
> 차원(파일링 준수 등)은 생략한다 — 없는 값을 만들어내지 않는다.

### ⚠️ Alerts
- 오래된 컴파일 진실: {N}건 (소스 갱신 후 미반영)
  - {stale_truth_1}: 소스 {source_page} ({source_updated}) > 컴파일 ({compiled_updated})

### 통계
- 총 엔티티: {N}페이지
- 최근 7일 업데이트: {N}건
- 임베딩 대기: {N}건

Step 4: Distribute

If invoked as part of daily-am-orchestrator:

• Post as a thread reply to the Phase 8 main message in #효정-할일
• Include the briefing as a formatted Slack mrkdwn message

If invoked standalone:

• Post to #효정-할일 as a new message with thread replies for each section

Step 5: Persist

Write output to outputs/daily-am/{date}/phase-8.5-gbrain-briefing.json:

{
  "status": "success",
  "date": "YYYY-MM-DD",
  "stats": {
    "total_pages": N,
    "recently_updated": N,
    "people_found": N,
    "companies_found": N,
    "meetings_found": N,
    "ideas_found": N,
    "stale_embeddings": N
  },
  "brain_health": {
    "composite_score": 85,
    "freshness": 90,
    "link_density": 80,
    "embedding_coverage": 95,
    "citation_density": 70
  },
  "alerts": {
    "stale_compiled_truths": 2
  },
  "briefing_markdown": "...",
  "slack_posted": true,
  "slack_ts": "..."
}

Error Handling

Error	Action
gbrain CLI not found	Skip with reason "gbrain CLI not installed"
PostgreSQL unreachable	Skip with reason "database unreachable"
gbrain search returns 0 results	Generate minimal briefing noting empty state
Slack post fails	Log error, persist briefing to file only
Timeout (>60s per query)	Use partial results, note incomplete data

Non-Blocking Guarantee

This phase is non-blocking. Failures in gbrain briefing:

• Do NOT affect overall_status of the morning pipeline
• Do NOT prevent Phase 8 consolidated briefing from completing
• Are logged in phase-8.5-gbrain-briefing.json with status: "failed" and error details