auto

name	auto
description	Morning Check / Session Entry Point. Delta pipeline: Stage 1 COLLECT → Stage 2 DETECT (+ Feed parallel). On-the-fly engagement analysis, no stored contacts.
user-invocable	true
allowed-tools	["Bash","Read","Write","Edit","Agent","Glob","Grep"]

/auto — Morning Check (Session Entry Point)

The central entry point for every automated session. Runs the 2-stage pipeline and gives the user a prioritized summary.

Replaces the previous /daily and /auto. Called by cron job (daily ~08:00) or manually.

IMPORTANT: Delegate the work to the specialized agents. Do NOTHING yourself — you are only the orchestrator.

Use the Agent tool to start the following agents sequentially (Stage 1→2) or in parallel (2a+2b):

• Stage 1: Start data-collector agent
• Stage 2a: Start signal-detector agent (with Stage 1 output)
• Stage 2b: Start feed-analyst agent (parallel to 2a)

Usage

/auto              # Full Morning Check (2-stage pipeline)
/auto quick        # Stage 1 only (collect data, no feed)

2-Stage Pipeline

Stage 1: COLLECT              Stage 2: DETECT
data-collector  ────────────► signal-detector
(Data Analyst)         parallel:
                               feed-analyst
                               (Social Media Scout)

Stage 1: COLLECT (data-collector agent)

What happens:

1. Local calculations (NO API call):
   • Post lifecycle transitions (Active → Cooling → Archived)
   • Archive posts at day 14 (full → mini-summary in data/posts/archive/)
   • Signal cleanup (> 7 days → delete)
   • Draft cleanup check (> 30 days → suggestion)
2. Auto-discover new published posts via CLI
3. Fetch notifications (1 API call = 80% of deltas)
4. Analytics for active posts (only lifecycle: Active/Cooling)
5. Snapshot checks (Day 3/7/14)

Output → Stage 2: Engagement data (commenters, reactors, profile viewers, new connections — with name, headline, post URN).

Stage 2: DETECT (parallel)

2a: signal-detector agent

• Input: Engagement data from Stage 1
• On-the-fly ICP matching (headline vs. config.icp — no stored contacts)
• Generate signals: outreach_candidate, comment_reply, new_connection_icp, profile_view, keyword_mention
• Write signal files to data/signals/
• Output: Prioritized signal list

2b: feed-analyst agent (parallel)

• Independent from Stage 1 (own feed call)
• Fetch feed (1 API call)
• Write insight files to data/feed-insights/
• Detect trends, find comment opportunities
• Output: Trending topics + comment opportunities

Summary to User

Morning Check complete (since last session: 18h)

DATA:
  [n] notifications processed
  [n] post metrics updated
  [n] lifecycle transitions (Active→Cooling: 1, Archived: 0)
  [n] new posts auto-discovered

SIGNALS ([n] new):
  HIGH: comment_reply — Sarah K commented on "AI in Practice"
  HIGH: outreach_candidate — Anna Schmidt (CTO @ TechAG, ICP: High)
  MEDIUM: keyword_mention — "AI Agents" in post by @tech-leader

FEED:
  Trending: "AI Agents" (7x, 2.3x avg), "Remote Work" (4x)
  Comment Opportunities:
    1. [HIGH] @sarah-k: "The future of..." (89 Rx in 3h)
    2. [MEDIUM] @tech-leader: "Why we switched..." (45 Rx in 5h)

NEEDS YOUR DECISION:
  - 2 comment replies pending → reply on LinkedIn
  - Last post 5 days ago → /draft for new post?
  - New patterns detected → /evolve for strategy update?
  - 2 comment opportunities → Write draft?

ALL GOOD:
  Content pipeline: 5 Ideas, 2 Drafts
  Competitors: current (8 days ago)
  Strategy: v1.2

Optional Checks (at the end)

After the pipeline — hints only, no actions:

• Content pipeline thin (< 3 Ideas in drafts/)? → "/ideas to refill"
• Published post without analysis? → "/analyze "
• Competitor data > 2 weeks old? → "Competitor update will happen at next /report"
• Weekly report due? → "Create /report"

Rules

• Update session — after completion set session.last_session_date
• Delta-based — only data since last session
• Pipeline order — Stage 1 → 2 (Feed parallel to 2)
• Collect data: yes. External actions: only with confirmation.
• Never post, send, or comment autonomously
• Short and actionable — no long explanations
• Agents in parallel where possible (Stage 2a + 2b)
• Priority — signals before content