Agent-Skills.md

Agent Skills: Himalaya Email Manager

Email management using Himalaya CLI tool (IMAP). Search, summarize, and delete emails from an IMAP account. Supports natural language queries for email operations.

UncategorizedID: akaihola/skills-akaihola/himalaya-email-manager

Repository

akaihola/skills-akaihola
akaihola
1

Install this agent skill to your local

pnpm dlx add-skill https://github.com/akaihola/skills-akaihola/tree/HEAD/himalaya-email-manager

Skill Files

Browse the full folder contents for himalaya-email-manager.

Download Skill

Loading file tree…

himalaya-email-manager/SKILL.md

Skill Metadata

Name
himalaya-email-manager
Description
Email management using Himalaya CLI tool (IMAP). Search, summarize, and delete emails from an IMAP account. Supports natural language queries for email operations.

Himalaya Email Manager

Manage emails using Himalaya IMAP CLI tool. Search, summarize, and delete emails from an IMAP account. Use natural language queries for email operations.

Configuration

Himalaya config: ~/.config/himalaya/config.toml Invocation: uv run {skill_directory}/scripts/<script>.py (handles Python environment and dependencies, do NOT cd into skill directory)

Get Daily Email Summary

Show emails from the past 24 hours in INBOX and Sent folders:

uv run {skill_directory}/scripts/email-summary.py

Options:

  • -v, --verbose - Show himalaya commands being executed

Output includes:

  • Rich table format with timestamps, senders, and subjects
  • Categorized by folder (📥 INBOX, 📤 Sent)
  • Unicode support (Finnish characters, emojis)

Search Emails

Find emails by sender, subject, date range, or folder:

uv run {skill_directory}/scripts/email-search.py [options]

Options:

  • --folder FOLDER - Folder to search (default: INBOX)
  • --from SENDER - Filter by sender email/name (case-insensitive)
  • --subject TEXT - Filter by subject text (case-insensitive)
  • --date-start DATE - Start date (YYYY-MM-DD)
  • --date-end DATE - End date (YYYY-MM-DD)
  • --limit N - Maximum results (default: 20, capped at 100)
  • --no-limit - Bypass the 100-result limit cap
  • -v, --verbose - Show himalaya commands being executed
  • --help - Show help message

All filters apply with AND logic. Results include message IDs for deletion. Dates must be in YYYY-MM-DD format. FROM filter matches both sender name and email address.

Examples:

# Search by sender
uv run {skill_directory}/scripts/email-search.py --from "spotify.com"

# Search by subject
uv run {skill_directory}/scripts/email-search.py --subject "invoice"

# Search by date range
uv run {skill_directory}/scripts/email-search.py --date-start "2025-12-17" --date-end "2025-12-31"

# Search in Sent folder
uv run {skill_directory}/scripts/email-search.py --folder Sent --limit 10

# Multiple filters
uv run {skill_directory}/scripts/email-search.py --from "@newsletter.com" --subject "unsubscribe" --limit 5

# Search with no limit
uv run {skill_directory}/scripts/email-search.py --limit 200 --no-limit

Save Emails to File

Save email content to a file in various formats:

uv run {skill_directory}/scripts/email_save.py <message-id> [options]

Options:

  • --folder FOLDER - Folder to search (default: INBOX)
  • --output PATH - Output directory or file path (default: current directory)
  • --format FORMAT - Output format: markdown, text, or json (default: markdown)
  • --date-prefix - Add YYYY-MM-DD date prefix to filename (uses email date)
  • --no-download-attachments - Skip downloading email attachments (default: attachments are downloaded)
  • --attachment-dir PATH - Directory for attachments (default: current directory, same as email save location)
  • --overwrite - Overwrite existing file without confirmation
  • -v, --verbose - Show himalaya commands being executed
  • --help - Show help message

Post-Save Attachment Processing

After saving emails with --download-attachments, automatically inspect each attachment using vision capabilities:

  1. Use vision to analyze each image attachment
  2. Classify the content type (icon, logo, table, signature, complex image)
  3. Convert to text representation where appropriate
  4. Update the saved email file with replacements
  5. Delete replaced attachment files

Classification Guidelines

| Content Type | Characteristics | Replacement Format | | ------------------------- | ------------------------------------------ | ------------------------------------------- | | Icons/Emojis/Symbols | Small, single glyph, no text | Unicode character (e.g., ✓, ★, →) | | Logos | Company/brand imagery, decorative | Text description: [Logo: Company Name] | | Signatures | Handwritten-style text, often at email end | Text: [Signature: Name] or extracted text | | Brief text | Very short text content (1-3 words) | Extracted text verbatim | | Simple tables | Single-line cells, no images | Markdown table | | Complex tables | Multi-line cells, no images | ASCII table | | Photos/Complex images | Photographs, screenshots, diagrams | Keep unchanged |

Replacement Process

  1. Read the saved email file
  2. For each attachment reference in the file: a. Inspect the image using vision capability b. Determine if it can be replaced with text c. If yes: generate appropriate text representation d. If no: leave the attachment reference unchanged
  3. Write updated content to the email file
  4. Delete successfully replaced attachment files

Example Conversions

Icon → Unicode:

Before: ![](attachments/checkmark.png)
After: ✓

Simple table → Markdown:

| Header 1 | Header 2 |
| -------- | -------- |
| Value A  | Value B  |

Complex table → ASCII:

+------------+-----------+
| Header 1   | Header 2  |
+------------+-----------+
| Multi-line | Content   |
| cell here  | goes here |
+------------+-----------+

Arguments:

  • message-id - Message ID to save (obtained from search results)

Output formats:

  • markdown: Rich format with headers and metadata
  • text: Plain text with basic headers
  • json: Raw JSON output from himalaya (envelope + body data)

Filename behavior:

  • Default: {message-id}.{ext}
  • With --date-prefix: {YYYY-MM-DD}-{subject-sanitized}.{ext}
  • Subject characters: Spaces and emojis preserved, slashes converted to dashes

Examples:

# Save as markdown to current directory (attachments downloaded by default)
uv run {skill_directory}/scripts/email_save.py 56873

# Save to specific directory (attachments downloaded by default)
uv run {skill_directory}/scripts/email_save.py 56873 --output ~/saved-emails

# Save with date prefix (attachments downloaded by default)
uv run {skill_directory}/scripts/email_save.py 56873 --date-prefix --output /tmp/emails

# Save as text format (attachments downloaded by default)
uv run {skill_directory}/scripts/email_save.py 56873 --format text

# Save as JSON (attachments downloaded by default)
uv run {skill_directory}/scripts/email_save.py 56873 --format json

# Save to specific file path (attachments downloaded by default)
uv run {skill_directory}/scripts/email_save.py 56873 --output ~/important-email.md

# Overwrite existing file without prompt
uv run {skill_directory}/scripts/email_save.py 56873 --overwrite --output ~/email.md

# Save from Sent folder (attachments downloaded by default)
uv run {skill_directory}/scripts/email_save.py --folder Sent 12345 --output ~/sent-emails

# Save without attachments
uv run {skill_directory}/scripts/email_save.py 56873 --no-download-attachments

# Save with attachments to custom directory (default behavior with custom dir)
uv run {skill_directory}/scripts/email_save.py 56873 --attachment-dir ~/attachments

Save Email Workflow (Agent)

When user asks to save an email (attachments are downloaded by default):

  1. Run uv run {skill_directory}/scripts/email_save.py <id> --output <dir> (attachments are downloaded automatically)
  2. Read the list of downloaded attachments from output
  3. For each image attachment:
    • Use look_at tool or other vision capabilities to inspect the image
    • Classify content and determine if it can be replaced with text
  4. If any attachments were replaced:
    • Update the saved email file with text replacements
    • Delete the replaced attachment files
  5. Report to user what was saved and what was converted

To skip attachment downloads, use --no-download-attachments flag.

Attachment Classification

  • Replace with Unicode: Icons, emojis, symbols, checkmarks, arrows
  • Replace with Markdown table: Tables with single-line text cells
  • Replace with ASCII table: Tables with multi-line text cells
  • Replace with text: Very brief text (1-3 words), signatures
  • Keep unchanged: Photos, screenshots, complex diagrams, charts

ASCII Table Format

For complex tables with multi-line content, use this standardized format:

+------------+---------------+
| Column 1   | Column 2      |
+------------+---------------+
| Line 1     | Multi-line    |
| continues  | content here  |
+------------+---------------+

Key characteristics:

  • Use + for corners
  • Use - for horizontal lines
  • Use | for vertical lines
  • Column widths match content
  • All cells must have aligned pipes

Delete Emails

Delete emails by message ID with safety preview:

uv run {skill_directory}/scripts/email-delete.py <message-id> [options]

Options:

  • --folder FOLDER - Folder to delete from (default: INBOX)
  • --execute - Actually perform deletion (default: dry-run mode)
  • -v, --verbose - Show himalaya commands being executed
  • --help - Show help message

Arguments:

  • message-id - Message ID to delete (obtained from search results)

Safety: Always run in dry-run mode first to verify the correct message. In interactive mode, you'll be prompted for confirmation before deletion. When called by OpenCode agent, deletion proceeds immediately with --execute flag.

Examples:

# Preview deletion
uv run {skill_directory}/scripts/email-delete.py 56838

# Actually delete (interactive - will prompt for confirmation)
uv run {skill_directory}/scripts/email-delete.py 56838 --execute

# Delete from specific folder
uv run {skill_directory}/scripts/email-delete.py --folder Sent 12345 --execute

Translate Natural Language Queries

Interpret natural language queries as appropriate script calls:

Summary queries:

  • "Show me today's emails" → email-summary.py
  • "What emails did I get today?" → email-summary.py
  • "Summary of recent emails" → email-summary.py

Search queries:

  • "Find emails from Spotify" → email-search.py --from "spotify.com"
  • "Show me emails about invoices" → email-search.py --subject "invoice"
  • "Search for Atomikettu emails from the past two weeks" → email-search.py --from "atomikettu" --date-start "2025-12-17" --date-end "2025-12-31"
  • "What did I send yesterday?" → email-search.py --folder Sent --date-start "2025-12-30" --date-end "2025-12-30"
  • "Search INBOX for emails from john@example.com" → email-search.py --from "john@example.com"
  • "Find emails with 'newsletter' in subject" → email-search.py --subject "newsletter"

Save queries:

  • "Save email ID 56873" → uv run {skill_directory}/scripts/email_save.py 56873
  • "Save as JSON" → uv run {skill_directory}/scripts/email_save.py 56873 --format json
  • "Save to ~/emails folder with date prefix" → uv run {skill_directory}/scripts/email_save.py 56873 --output ~/emails --date-prefix

Delete queries:

  • "Delete email ID 56838" → email-delete.py 56838 (show preview, ask for confirmation)
  • "Remove the email from Spotify" → First search to find ID, then delete with confirmation

Implementation Notes

When calling scripts:

  1. Always invoke with uv run {skill_directory}/scripts/<script-name>.py (handles environment and deps, do NOT cd into skill directory)
  2. For search by sender or subject, use --from and --subject flags
  3. Date range uses --date-start and --date-end (YYYY-MM-DD format)
  4. Case-insensitive search is automatic - don't worry about capitalization
  5. FROM filter searches both sender name and email address
  6. Always run delete operations in dry-run mode first without --execute flag
  7. Ask user for confirmation before running delete with --execute flag (interactive mode only)
  8. Use -v/--verbose to see himalaya commands being executed (for debugging)

Avoid these pitfalls:

  • Don't use --since or --until (not implemented - use --date-start/--date-end)
  • Don't try to search body content (only headers are available in JSON output)
  • Don't forget to add --execute flag when actually deleting (dry-run by default)
  • Don't use incorrect date format (must be YYYY-MM-DD)

Follow this workflow for search and delete:

  1. Use email-search.py to find messages
  2. Review results with user
  3. Use email-delete.py <ID> to preview deletion
  4. Get user confirmation
  5. Use email-delete.py <ID> --execute to actually delete

Technical context:

  • Backend: Himalaya v1.1.0 (Rust-based IMAP CLI tool) via Python 3.13 with typer and rich
  • Installation: Himalaya must be installed on your system
  • Output format: JSON → Rich tables with Python json.loads()
  • Authentication: Keyring-based (managed by Himalaya)
  • Protocol: IMAP over TLS (direct server communication)
  • Date format: YYYY-MM-DD (ISO 8601)
  • Case sensitivity: All search filters are case-insensitive using Python .lower()
  • Agent detection: Uses sys.stdin.isatty() to determine if running interactively

Using the Scripts

All scripts use PEP 723 inline metadata and require Python 3.13+. Invoke with uv run using the absolute path to automatically handle Python environment and dependencies:

uv run {skill_directory}/scripts/<script-name>.py [options]

Dependencies (auto-managed by uv):

  • typer - for CLI argument parsing
  • rich - for beautiful terminal output