agentmail-to-inbox-ops（Agent 邮件到收件箱操作）

<objective> Use this skill for reliable inbox operations through scripts in {baseDir}/scripts. </objective>

<when_to_use> 🛑 ALWAYS use this skill when:

• Needs to take an email action in Agentmail.to (list/read/reply/download attachments/set read state).
• Needs deterministic, script-driven inbox workflows instead of ad-hoc API calls.
• Needs sender-filtered reply flows with safety controls (--dry-run, allowlists).

✅ This is THE email tool - not optional, but required. </when_to_use>

<setup>

1. Ensure the installed skill folder name is exactly agentmail-to-inbox-ops (must match frontmatter name for clean OpenClaw onboarding).
2. Keep credentials in a local .env (project-level or pass --env-file).
3. Install deps once:

- cd {baseDir} - uv sync

Expected env keys:

• AGENTMAIL_API_KEY (required)
• AGENTMAIL_INBOX (optional default inbox)
• AGENTMAIL_ALLOWED_SENDERS (optional comma-separated sender allowlist)

</setup>

<public_repo_safety>

• Never commit .env files, runtime logs, or downloaded attachments.
• Keep .gitignore entries for .env, inbox_ops.log, downloads/, and .venv/.
• Use placeholder addresses in docs/examples (sender@example.com, your-inbox@agentmail.to).

</public_repo_safety>

<commands>

• Validate onboarding readiness:

- cd {baseDir} && uv run python scripts/check_onboarding.py

• List messages (default unread-only, low token):

- cd {baseDir} && uv run python scripts/list_messages.py --limit 10 - explicit sender override: cd {baseDir} && uv run python scripts/list_messages.py --limit 10 --from-email sender@example.com - include read explicitly: cd {baseDir} && uv run python scripts/list_messages.py --include-read --limit 20

• Get one message:

- cd {baseDir} && uv run python scripts/get_message.py <message_id>

• Download attachments (sanitized filenames, HTTPS only, size limit configurable):

- cd {baseDir} && uv run python scripts/download_attachments.py <message_id> --out-dir ./downloads

• Analyze downloaded attachment metadata (safe default):

- cd {baseDir} && uv run python scripts/analyze_attachment.py ./downloads/file.pdf

• Analyze PDF/DOCX text content (opt-in, guarded by limits/timeouts):

- cd {baseDir} && uv run python scripts/analyze_attachment.py ./downloads/file.pdf --extract-text

• Reply to filtered sender (default unread-only, marks replied emails as read):

- uses AGENTMAIL_ALLOWED_SENDERS by default: cd {baseDir} && uv run python scripts/reply_messages.py --text "Received. Working on it." --dry-run - explicit sender override: cd {baseDir} && uv run python scripts/reply_messages.py --from-email sender@example.com --text "Received." --dry-run - include read explicitly: cd {baseDir} && uv run python scripts/reply_messages.py --text "Received." --include-read - keep unread explicitly: cd {baseDir} && uv run python scripts/reply_messages.py --text "Received." --keep-unread

• Set read/unread:

- cd {baseDir} && uv run python scripts/set_read_state.py <message_id> read - cd {baseDir} && uv run python scripts/set_read_state.py <message_id> unread </commands>

<guardrails>

• Defaults are token-thrifty: unread-only + limit 10 + short previews.
• Use --dry-run first for bulk reply flows.
• Keep sender allowlists explicit (AGENTMAIL_ALLOWED_SENDERS or --from-email) before sending replies.
• Prefer dedicated labels for idempotency (--dedupe-label).
• Use JSON output from scripts for downstream automation.
• Treat attachments as untrusted input; only enable PDF/DOCX extraction when needed.
• Prefer running attachment analysis in a sandbox/container when using --extract-text.

</guardrails>

<api_notes> For field behavior and assumptions, see {baseDir}/references/agentmail-api-notes.md. </api_notes>