Define the automated daily, weekly, and monthly task schedule. These jobs run on the times specified below, every day including weekends, unless the agent is in Crisis Mode (all jobs paused until owner clears).

PRE-FLIGHT CHECK (runs before EVERY job)

Before executing any scheduled or on-demand job, check:

1. Scan SOUL.md → plugins.git-plugin-amc section for any {{PLACEHOLDER}} string
2. If {{PLACEHOLDER}} found:
   → ABORT this job immediately — do not execute
   → Bootstrap Mode is incomplete
   → Send Bootstrap Opening Message via mcp.lark.message (if not sent in last 2 hours)
     "配置还没完成，我们先把设置做好吧！
      Your setup isn't complete yet — let's finish the configuration first!"
   → Resume Bootstrap interview from the last answered question (check ownerreview Lark Doc)
   → Log: "[YYYY-MM-DD HH:MM] Job '{job_name}' skipped — Bootstrap incomplete"
3. If no {{PLACEHOLDER}} found:
   → Proceed with job normally
   → **Kanban Logging**: You MUST create a task in the AMC Kanban system (status: `in_progress`) for this specific cron job. (e.g., "11:00 Lunch Publish Window"). See `kanban-integration.md`.

This check is the single source of truth. If SOUL.md is clean, all jobs run. If any placeholder remains, no operational work executes — period.

Daily Automation Schedule

Time	Job	Reads	Writes
07:00	Daily Kanban Sync	Content Schedule Bitable	AMC Kanban (status=todo)
08:00	Topic discovery	Trending radar cache + Media Index Bitable	Content Schedule Bitable (status=draft)
10:00	Google Maps actions	Content Schedule Bitable + GMB API	Post records + reply log
11:00	Lunch publish window	Content Schedule Bitable (status=ready)	Post records (status=published)
13:00	Lunch window close + snapshot	Platform Insights APIs	Post records + daily-metrics log
17:00	Dinner publish window	Content Schedule Bitable (status=ready)	Post records (status=published)
19:00	Dinner window close + snapshot	Platform Insights APIs	Post records
20:00	Comment & DM batch reply	Platform Insights APIs + post records	Post records + ownerreview Lark Doc (escalations)
23:30	Daily metrics snapshot (PAUSED)	Platform Insights APIs	report/analytics/daily-YYYY-MM-DD.md

Job Details

Important Kanban Requirement: For every single job listed below, you must log your execution progress to the AMC Kanban. Create the task at the start, update the description as you progress through the steps, and mark it done when the job completes.

06:30 · Trending Radar Refresh (Mon/Thu)

1. Fetch Common Vault Lark Doc (trending-radar URL from SOUL.md config)
2. Cache locally for the day's topic discovery
3. Flag any food-adjacent trends that could become Trending-Radar Riff content
   (see [[core/content-types]] Type 7)
4. If Lark Doc is unreachable: log error, use yesterday's cache, alert owner (Tier 3)

07:00 · Daily Kanban Sync

1. Read Content Schedule Bitable for all items scheduled for today.
2. For each item, create a task in the AMC Kanban system assigned to your agentId.
3. Set the task title to the post topic/slug and status to `todo`.
4. This ensures the human team has full visibility of your planned daily output.

08:00 · Topic Discovery & Content Creation

1. Read cached trending radar
2. Read Media Index Bitable for any new raw media uploaded by owner
3. [content-pillars] Read vault/brand/content-pillars.md:
   - Calculate this week's actual pillar distribution (from published + scheduled content)
   - Identify gap pillars (actual% < target% by >5%)
   - Prioritise idea generation for gap pillars first
4. Cross-reference with Content Schedule Bitable for gaps in the coming 3 days
5. Propose 1-3 new content ideas, each labelled with its pillar:
   "[Pillar: Brand Story] 幕后故事: 师傅手拉面的 30 年..."
6. Create full content drafts (not just ideas) in Content Schedule Bitable (status=ready)
7. Run Compliance Gate → Bilingual Gate on each draft
8. Schedule for next available publishing window
   → No approval needed. Content publishes automatically per cron schedule.
   → If Compliance Gate RED or allergen unconfirmed: hold + Lark alert (see owner-approval.md)
9. Update vault/brand/content-pillars.md weekly health summary

10:00 · Google Maps Actions

1. If Google Maps is in pending_platforms: SKIP. Weekly reminder only.
2. Check GMB API for new reviews since last check
3. For each new review:
   - Positive / Neutral → draft response, auto-publish
   - Negative           → draft empathetic response, auto-publish after 30-min delay
   - Crisis keywords (illness / poisoning / 食物中毒 / 异物)
                        → trigger full Crisis Mode, do NOT auto-reply
4. Check for new Q&A questions → draft responses, auto-publish
5. Log all actions in post records + ownerreview Lark Doc

11:00 · Lunch Publish Window

1. Read Content Schedule Bitable for items with time=11:00 and status=ready
2. For each ready item:
   - Update its corresponding AMC Kanban task status to `in_progress`
   - Verify Compliance Gate + Bilingual Gate still pass (content may have aged)
   - Publish via AMC Kanban (primary path — covers all major platforms)
     or mcp.gbp (Google Business Profile posts)
     or push draft to Lark for manual publish (RedNote + any pending_platforms)
   - If AMC Kanban publish fails: autonomously attempt the next available path
     (platform direct API → browser automation → Lark manual draft)
     and log the reason + method used in the Kanban task description
   - Update post record status to "published"
   - Update the AMC Kanban task status to `done`
   - Log publish timestamp and URL
3. If Compliance Gate RED or allergen unverified at publish time:
   - Hold item, move to status=hold
   - Send Lark alert to team (see owner-approval.md Compliance Hard Stops)

13:00 · Lunch Window Close + Snapshot

1. Capture engagement metrics for all posts published in the 11:00 window
2. Log to post records (likes, comments, shares, views, saves — per platform)
3. Detect anomalies: if engagement is >50% below average for similar posts → flag in ownerreview Lark Doc
4. Update daily-metrics log
5. [performance-learning] For any post now at 24h age:
   → Write metrics_24h fields to vault/analytics/performance-log.md
6. [performance-learning] For any post now at 48h age:
   → Write metrics_48h.engagement_rate
   → Calculate score vs. brand baseline
   → If score ≥ 3 (Winner): trigger winner extraction + Lark notify
   → If score = 0 (Poor): write failure analysis to what_didnt field

17:00 · Dinner Publish Window

Same as 11:00 Lunch Publish Window. Publish items with time=17:00 and status=ready.

19:00 · Dinner Window Close + Snapshot

Same as 13:00 Lunch Window Close. Capture metrics for 17:00 posts. Also includes performance-learning 24h/48h scoring for dinner-window posts.

20:00 · Comment & DM Batch Reply

1. Fetch all new comments and DMs since last check (across all active_platforms only)
   - **Fallback Protocol:** If the AMC Kanban comment/DM API is unavailable or rate-limited, you MUST automatically use your Browser Control tool to log in to the platform and simulate a real human to fetch and reply to comments. Log the reason in the Kanban task.
2. For each comment/DM:
   - Classify: question / compliment / complaint / spam / crisis keyword
   - Draft reply in customer's language ([[localization/bilingual-gate]])
   - Compliment / question → auto-reply
   - Complaint → auto-reply with empathetic tone; if crisis keyword detected → Crisis Mode
   - Spam → flag for deletion, do not reply
3. Log all replies in post records
4. Log any crisis triggers in ownerreview Lark Doc

23:30 · Daily Metrics Snapshot

1. Pull final engagement metrics for ALL posts published today
2. Compare to 7-day rolling average per platform
3. Flag any platform with >30% drop week-over-week → Tier 3 alert to owner
4. Write raw data to report/analytics/daily-YYYY-MM-DD.md

Weekly Triggers

Day/Time	Job	Output
Mon/Thu 06:30	Trending radar refresh	Local cache
Monday 08:00	Self-improvement report	Send weekly feedback summary & self-assessment to team (merges human feedback + performance-learning data insights)
Monday 08:00	7d metrics batch update	Read performance-log.md, fetch 7d data for all posts ≥7 days old, update metrics_7d fields, refresh winner-library.md platform rankings
Monday 09:00	Plugin version check	Check latest version; if update available → send Lark prompt; apply on team reply "更新插件"
Monday 09:00	Pending platform reminder	For each platform in pending_platforms: notify team "[Platform] 尚未连接账号，连接后即可开始自动运营"
Monday 09:00	Allergen pending check	Scan allergen-gate.md for [?PENDING] entries; if any found → Lark alert: "[菜品名] 过敏原信息未确认，涉及该菜品的帖子将暂停发布直到补全"
Monday 10:00	Weekly report generation	report/weekly/YYYY-Www.md + Lark notification to team
Sunday evening	Weekly content batch	Propose 2-3 themes → run repurpose chain → fill next week's postschedule
Friday (any time)	Weekly performance review	Analyze engagement data → update hook-engine.md or scheduling.md if patterns found

Monthly Triggers

Day/Time	Job	Output
1st of month, 10:00	Monthly report generation	report/monthly/YYYY-MM.md + Lark Doc notification
1st of month	Compliance review	Review platform-policies.md for any updates; update if needed
1st of month	Voice drift check	Compare a random sample of 5 posts against brand-voice.md; flag drift to owner
1st of month	KPI reset	Reset 30-day baseline metrics in report-rules.md

Crisis Mode Override

If Crisis Mode is active ([[operations/owner-approval]] Crisis section):

ALL cron jobs are paused except:
- 06:30 Trending Radar (continue reading, do not act)
- 10:00 Google Maps (crisis keyword check only, no auto-replies)
Resume all jobs only after owner sends "CRISIS-CLEAR"