Daily Skill v3.0 - 9-Phase Pipeline

3-source (Gmail/Slack/GitHub) incremental collection + AI cross-source analysis + action recommendation engine.

Paradigm: "collect+display" -> "learn+recommend actions"

Design Reference: C:\claude\docs\02-design\daily-redesign.design.md

Execution Rules (CRITICAL)

When this skill activates, you MUST run the 9-Phase Pipeline below in order.

code

Phase 0 -> 1 -> 2 -> 3 -> 4 -> 5 -> 6 -> 7 -> 8
Config   Expert  Collect  Attach  AI     Action  Project  Gmail    State
Bootstrap Context  (incr)  Analyze Analyze Recom   Ops    Housekp  Update

Phase 0: Config Bootstrap

Search for .project-sync.yaml in CWD; auto-generate if missing.

Step 0.1: File lookup

code

Read: .project-sync.yaml
# If missing, go to Step 0.2
# If present, check version:
#   v1.0 -> compat mode (read-only, treat meta.auto_generated=false)
#   v2.0 + auto_generated=true -> auto-update allowed
#   v2.0 + auto_generated=false -> read-only

Step 0.2: Auto-generate (when file is absent)

bash

# Project identification
# Priority 1: Read CLAUDE.md -> extract project name, tech stack, goals
# Priority 2: Read README.md -> fallback
# Priority 3: Use directory name

# Source discovery
python -m lib.gmail status --json          # Gmail auth + email address
python -m lib.slack status --json          # Slack auth + workspace
gh auth status                             # GitHub auth + repo

# Project type auto-classification
# vendor_management: keywords "업체", "vendor", "RFP", "견적"
# development: presence of src/, package.json, setup.py
# infrastructure: presence of Dockerfile, terraform
# research: high docs/ ratio
# content: keywords "영상", "media", "upload"

Write .project-sync.yaml v2.0:

yaml

version: "2.0"
project_name: "{auto-detected}"
meta:
  auto_generated: true
  confidence: 0.0-1.0
  generated_at: "ISO timestamp"
  pending_additions: []
daily:
  sources:
    gmail: { enabled: true/false }
    slack: { enabled: true/false, channel_id: "{fuzzy-matched}" }
    github: { enabled: true/false, repo: "{from git remote}" }
  project_type: "{auto-classified}"

Error: If both CLAUDE.md and README.md are missing, create a minimal config based on the directory name.

Phase 1: Expert Context Loading

Build project expert context in 3 tiers.

code

# Tier 1: Identity Context (500t)
Read: CLAUDE.md + .project-sync.yaml
-> project name, goals, tech stack, key terminology, data sources, communication_style

# Tier 2: Operational Context (2000t)
Read: .omc/daily-state/<project>.json -> learned_context section
-> entities (vendors, people, statuses), patterns (recurring patterns)

# Tier 3: Deep Context (3000t, only when present)
Read: docs/ core files (README, PRD, architecture)
Read: .omc/daily-state/<project>.learned-context.json
-> domain knowledge, accumulated prior analysis

Output: expert_context JSON (injected into Phase 4, 5 prompts)

Error: Without CLAUDE.md, build minimal Tier 1. Without daily-state, skip Tier 2 (first run).

Phase 2: Incremental Data Collection

Incremental collection across 3 sources. Cursors stored in .omc/daily-state/<project>.json so subsequent runs collect only new data.

Step 0: Auth check

bash

python -m lib.gmail status --json    # check authenticated, valid
python -m lib.slack status --json    # check authenticated, valid
gh auth status                       # check exit code

Skip sources that fail auth for this run. Abort with error if zero sources are active.

Step 1: Gmail collection

python

from lib.gmail import GmailClient
client = GmailClient()

# First run: seed historyId + 7-day lookback
profile = client.get_profile()          # obtain historyId
emails = client.list_emails(query="newer_than:7d", max_results=50)

# Incremental: History API
result = client.list_history(
    start_history_id=state["cursors"]["gmail"]["history_id"],
    history_types=["messageAdded"],
    max_results=100
)
# historyId 404 -> fallback to list_emails(query="after:...")

Step 2: Slack collection

python

from lib.slack import SlackClient
client = SlackClient()

# First run: 7-day lookback
messages = client.get_history(channel=channel_id, limit=100, oldest=str(time.time()-604800))

# Incremental: since last_ts
messages = client.get_history(channel=channel_id, limit=100, oldest=state["cursors"]["slack"]["last_ts"])

Step 3: GitHub collection

bash

gh issue list --since "{last_check}" --json number,title,state,author,updatedAt,labels
gh pr list --json number,title,state,author,updatedAt,reviewDecision
gh api repos/{owner}/{repo}/commits --jq '.[0:10] | .[] | {sha: .sha[0:7], message: .commit.message[0:80], date: .commit.author.date}'

Phase 3: Attachment Analysis

AI analysis of attachments from emails collected in Phase 2.

Target filter: PDF, Excel, images (PNG/JPG)

SHA256 cache: Previous analysis results stored in state.cache.attachments[sha256]. Prevents re-analysis of identical attachments.

Download:

python

data = client.download_attachment(message_id, attachment_id)  # lib/gmail/client.py

Analysis methods:

Type	Condition	Method
PDF (<= 20p)	`page_count <= 20`	Direct analysis via Claude Read tool
PDF (> 20p)	`page_count > 20`	Chunk-split analysis via `lib/pdf_utils` PDFExtractor
Excel/CSV	-	Structure summary (rows/cols, headers, 5-row sample)
Images	PNG/JPG	Claude Vision analysis

Analysis perspective: Apply expert_context.analysis_perspective

•vendor_management: "Is this a quote? Amount, validity, terms?"
•development: "Is this an API spec? Changes, breaking changes?"

Error: Download failure -> skip; encrypted PDF -> skip and report

Phase 4: AI Cross-Source Analysis

Inject expert_context and analyze as a project expert.

Step 1: Independent per-source analysis

Claude reads Phase 2 raw data + Phase 3 attachment analysis and analyzes each source:

•Gmail: sender, key content, urgency (URGENT/HIGH/MEDIUM/LOW), required actions, quote info, status inference
•Slack: decisions, action items, unresolved questions, entity matching
•GitHub: PR status, issue status, CI/CD status

Step 2: Cross-source connection analysis

Detect links across sources:

•Same-topic detection (simultaneous mention in Gmail + Slack)
•Action linkage (email request -> GitHub issue/PR)
•Status inconsistency (Gmail "done" vs GitHub issue open)
•Timeline construction (chronological linking of events on the same topic across sources)

Prompt details: See docs/02-design/daily-redesign.design.md Sections 3.2, 3.3

Error: If only a single source is active, skip cross-source analysis and perform independent analysis only.

Phase 5: Action Recommendation

Generate concrete action drafts from analysis results.

Action Type	Generation Condition
Email reply draft	No response for 48h+, quote received, explicit request
Slack message draft	Unanswered question, follow-up needed
GitHub action	PR review pending 3+ days, unanswered issue

Tone calibration: Reference communication_style

•email_tone: professional / casual / formal
•slack_tone: casual / professional
•language: ko / en / mixed

Limits: Max 10 items, sorted URGENT -> HIGH -> MEDIUM; include estimated time per action

Prompt details: See docs/02-design/daily-redesign.design.md Section 3.4

Error: If no analysis results, display "No additional actions required."

Phase 6: Project-Specific Operations

Conditionally run based on project_type in .project-sync.yaml.

vendor_management type:

•Update Slack Lists (ListsSyncManager)
•Auto vendor status transitions (StatusInferencer)
•Quote comparison table (QuoteFormatter)

bash

cd {project_path} && python -c "
import sys; sys.path.insert(0, 'scripts/sync')
from lists_sync import ListsSyncManager
manager = ListsSyncManager()
manager.update_item('{vendor}', status='{status}', quote='{quote}', last_contact='{date}')
manager.generate_summary_message()
manager.post_summary()
"

development type:

bash

gh run list --limit 5 --json databaseId,status,conclusion,name,createdAt
gh pr list --state open --json number,title,headRefName,updatedAt
gh api repos/{owner}/{repo}/milestones --jq '.[] | {title, open_issues, closed_issues}'

Error: Skip Phase 6 if config_file is missing. On Slack Lists API failure -> print warning and continue.

Phase 7: Gmail Housekeeping

Clean up Gmail based on housekeeping in .project-sync.yaml.

7a. Auto label application (gmail_label_auto: true):

•Filter Phase 2 emails without labels
•Auto-apply labels on vendor_domains match
•Exclude system mail (noreply, notifications, drive-shares)

7b. INBOX cleanup (inbox_cleanup setting):

Mode	Behavior
`"auto"`	Auto archive
`"confirm"`	Show target list, confirm via AskUserQuestion
`"skip"`	Skip (default)

Error: On Gmail API failure -> skip action, print warning

Phase 8: State Update

Phase A (right after collection — record cursors):

python

state["cursors"]["gmail"]["history_id"] = new_history_id
state["cursors"]["gmail"]["last_timestamp"] = datetime.utcnow().isoformat() + "Z"
state["cursors"]["slack"]["last_ts"] = last_message_ts
state["cursors"]["github"]["last_check"] = datetime.utcnow().isoformat() + "Z"
state["last_run"] = datetime.utcnow().isoformat() + "Z"
state["run_count"] += 1

Phase B (after analysis — cache + learning record):

python

state["cache"]["attachments"][sha] = analysis  # attachment cache
state["learned_context"]["entities"].update(new_entities)
state["learned_context"]["patterns"].extend(new_patterns)

Auto config update (only when auto_generated: true):

•On new domain detection, append to pending_additions

Error: On file write failure -> same data is re-collected next run (no data loss).

Output Format

code

================================================================================
                   Daily Dashboard v3.0 (YYYY-MM-DD Day)
                   Project: {project_identity}
================================================================================

[Sources] --------------------------------------------------------
  Gmail: {N} items ({new} new) {auth_status}
  Slack: {N} items ({new} new) {auth_status}
  GitHub: {N} issues, {N} PRs {auth_status}

[Cross-Source Insights] ------------------------------------------------
  1. {topic}: {insight} (sources: Gmail+Slack)

[Action Items] --------------------------------------------------------

  URGENT ({N} items)
  #{id}. [{type}] {target}
     Draft: {draft_preview}
     ETA: {estimated_time} | Reason: {reason}

  HIGH ({N} items)
  #{id}. [{type}] {target}
     Draft: {draft_preview}
     ETA: {estimated_time} | Reason: {reason}

[Per-Source Details] --------------------------------------------------------
  Gmail / Slack / GitHub details

[Attachment Analysis] --------------------------------------------------------
  * {filename}: {summary}

================================================================================
  Total actions: {total} | Estimated: ~{total_time} min
  Next run: incremental collection (cursors saved)
================================================================================

vendor_management extra sections: vendor status table, quote comparison, Slack Lists update result development extra sections: CI/CD status, branch status, milestone progress

Subcommands

Command	Description
`/daily`	Full dashboard (all 9 phases)
`/daily ebs`	EBS daily briefing (existing EBS-specific workflow)

EBS Briefing Mode

/daily ebs preserves the existing EBS-specific workflow.

bash

cd C:\claude\ebs\tools\morning-automation && python main.py --post

Detailed workflow: identical to the prior v2.0.0 EBS section.

Changelog

Version	Changes
3.0.0	Full 9-Phase Pipeline redesign. Removed Secretary dependency. Added incremental collection, attachment AI analysis, cross-source analysis, action recommendation engine, Config Auto-Bootstrap, Expert Context Loading. Absorbed daily-sync functionality.
2.0.0	Integrated EBS briefing mode (`/daily ebs`)
1.1.0	Skill-chain integration (Gmail, Secretary)
1.0.0	Initial release