add-email-archive

name	add-email-archive
description	Set up email archive system — interactive calibration builds sender rules, creates folder taxonomy, and configures batch processing to reach inbox zero. Requires at least one email account registered via /add-email-account. Run this before using /email-archive in your main chat.

Add Email Archive

Interactive setup for the email archive system. Builds sender-based classification rules from your actual email, creates a folder taxonomy, and configures the container skill for batch processing.

Prerequisite: At least one email account must be registered. Run /add-email-account first.

Phase 1: Prerequisites

Check if already configured

MAIN_FOLDER=$(sqlite3 store/nanoclaw.db "SELECT folder FROM registered_groups WHERE is_main = 1 LIMIT 1;" 2>/dev/null)
echo "MAIN_FOLDER=${MAIN_FOLDER:-unknown}"
test -d "groups/${MAIN_FOLDER}/email-archive" && echo "CONFIGURED" || echo "NOT_CONFIGURED"

If CONFIGURED, ask the user whether to recalibrate or skip to Phase 3.

If MAIN_FOLDER is empty, ask which group folder to use.

Check email accounts

test -f "groups/${MAIN_FOLDER}/email-accounts.yaml" && cat "groups/${MAIN_FOLDER}/email-accounts.yaml" || echo "NO_ACCOUNTS"

If NO_ACCOUNTS or no enabled accounts:

No email accounts registered. Run /add-email-account first to connect your email.

Stop here.

If accounts exist, list them:

Found N registered email account(s):

• gmail (gws) — @.com
• outlook (ms365) — @.edu

Which accounts should be included in the archive? (default: all)

Use AskUserQuestion to confirm.

Create directory structure

mkdir -p "groups/${MAIN_FOLDER}/email-archive/calibration"
mkdir -p "groups/${MAIN_FOLDER}/email-archive/state"
touch "groups/${MAIN_FOLDER}/email-archive/overrides.jsonl"

Phase 2: Calibration

The goal is to build sender rules that cover 80%+ of email volume so the LLM only handles the long tail.

Step 2a: Initial taxonomy

Present the default categories:

Here's a starting folder taxonomy:

• Archive/Work — work-related email
• Archive/Personal — personal correspondence
• Archive/Accounts — service notifications, password resets, confirmations
• Archive/Newsletters — subscriptions, digests, mailing lists worth keeping
• To Delete — spam, marketing, expired notifications

Want to customize? We can add subcategories later after seeing your actual email patterns.

Use AskUserQuestion to confirm or collect changes.

Step 2b: Sample and scan

For each selected account, pull up to 200 inbox messages using the provider operations from /add-email-account's reference table. Look up the account's type in email-accounts.yaml and use the corresponding list inbox command.

Parse results. For each email, extract from address and domain. Group by domain, then by address within domain. Sort by frequency.

Save raw samples to groups/${MAIN_FOLDER}/email-archive/calibration/samples-{account_id}.json.

Step 2c: Interactive rule building

Merge sender frequencies across all selected accounts. Present top senders grouped by domain:

Top email senders (all accounts):

1. [e-mail] (23 emails)
   Subjects: "PR #123 merged", "Issue #456 opened"
   → Which folder? [1-5, or new name]

2. *@clemson.edu (15 emails from 8 addresses)
   Subjects: "Budget meeting", "Student worker hours"
   → Which folder? [1-5, or new name]

Categories:
  1. Archive/Work
  2. Archive/Personal
  3. Archive/Accounts
  4. Archive/Newsletters
  5. To Delete
  6. [Type a new category name]
  7. Skip (let LLM decide per email)

Use AskUserQuestion for each sender. Let the user batch answers (e.g., "1:5, 2:1, 3:3").

Each answer becomes a rule in rules.yaml:

rules:
  - match:
      from_domain: "github.com"
    folder: "To Delete"
    source: "calibration"
    created: "2026-04-11"
    applied_count: 0

  - match:
      from_address: "[e-mail]"
    folder: "Archive/Accounts"
    source: "calibration"
    created: "2026-04-11"
    applied_count: 0

Continue until the user says stop or all senders with 3+ emails are covered.

Step 2d: Coverage check

Re-scan the sample against the rules:

Coverage report:
  Rules match: 164/200 emails (82%)
  Unmatched: 36 emails from 12 unique senders

  Unmatched senders (by frequency):
  - [e-mail] (3 emails)
  - [e-mail] (2 emails)

If coverage < 70%, suggest adding rules for remaining high-frequency senders. If coverage >= 80%, declare calibration successful.

Step 2e: Model selection and cost estimate

The unmatched emails need LLM classification. Estimate costs:

Estimated total inbox: ~N emails
Rule-matched (N%): ~N emails — FREE
LLM-classified (N%): ~N emails

The group's configured model handles LLM classification.
Current group model: [read from DB or .env]

Recommendation: Start with the current model. If accuracy is poor
in interactive review, switch to a stronger model via /model.

Use AskUserQuestion to confirm.

Step 2f: Create folders

For each selected account, create the taxonomy folders/labels using the account's provider operations. Look up the account type and use the corresponding create label/folder command.

Store the resulting folder/label IDs in the archive config (see Phase 3).

Phase 3: Deploy

Write config.yaml

Write groups/${MAIN_FOLDER}/email-archive/config.yaml:

# Email Archive Configuration
# Accounts are registered in email-accounts.yaml (via /add-email-account).
# This file stores archive-specific settings per account.

archive_accounts:
  - id: gmail               # must match an id in email-accounts.yaml
    folder_ids:              # provider-specific folder/label IDs
      "Archive/Work": "Label_abc123"
      "Archive/Personal": "Label_def456"
      "Archive/Accounts": "Label_ghi789"
      "Archive/Newsletters": "Label_jkl012"
      "To Delete": "Label_mno345"
  - id: outlook
    folder_ids:
      "Archive/Work": "AAMkAD..."
      "Archive/Personal": "AAMkAE..."
      "Archive/Accounts": "AAMkAF..."
      "Archive/Newsletters": "AAMkAG..."
      "To Delete": "AAMkAH..."

batch_size: 25
mode: interactive  # interactive | supervised | autonomous

taxonomy:
  - Archive/Work
  - Archive/Personal
  - Archive/Accounts
  - Archive/Newsletters
  - To Delete

thresholds:
  llm_confidence: 0.8
  auto_promote_after: 5

Write initial rules.yaml

Write the rules built during calibration.

Initialize state

Write groups/${MAIN_FOLDER}/email-archive/state/progress.yaml with one entry per selected account:

accounts:
  gmail:
    last_processed_date: null
    total_processed: 0
    status: not_started
  outlook:
    last_processed_date: null
    total_processed: 0
    status: not_started
stats:
  rule_matched: 0
  llm_classified: 0
  corrections: 0
  rules_auto_promoted: 0

Rebuild container

./container/build.sh

Done

Email archive is configured with N account(s). Next steps:

1. Send /email-archive in your main chat to process the first batch
2. Review the classifications — correct any mistakes
3. After a few batches, switch mode from interactive to supervised
4. The system learns from your corrections and auto-generates rules

Commands: /email-archive, /email-archive status, /email-archive review, /email-archive recalibrate