email-assistant — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited email-assistant (Plugin) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
An autonomous email triage and reply system built with Claude Code and Google Workspace CLI (gws).
Email triage screenshot
Every workday at 8:30, Claude Code reads your inbox, drafts replies in your voice, saves them as Gmail drafts, and blocks time in your calendar. You wake up, open Gmail, review the drafts, and hit Send.
The system learns from every edit you make. If you change a draft before sending, next morning it compares what it wrote vs what you actually sent — and adjusts.
8:30 launchd triggers Claude Code
│
▼
┌─────────────────┐
│ 1. Triage inbox │ Classify emails: reply / info / ops
│ 2. Learn │ Compare yesterday's drafts vs what was sent
│ 3. Draft │ Write replies in your voice, save as Gmail Drafts
│ 4. Label │ Tag emails: AI ready / info / ops / high / $
│ 5. Summarize │ Estimate time per email
│ 6. Calendar │ Block time based on actual workload
└─────────────────┘
│
▼
┌─────────────────┐
│ You open Gmail │ Drafts are there, in the right threads
│ ✓ Send as-is │ Most drafts just need one click
│ ✎ Edit + Send │ Fix anything directly in Gmail
│ ✗ Delete │ Skip what doesn't need a reply
└─────────────────┘
│
▼
┌─────────────────┐
│ Next morning │ System diffs drafts vs sent messages
│ Learns from edits│ Every correction improves future drafts
└─────────────────┘AI ready, info, ops, high, $launchd for scheduling)npm install -g @anthropic-ai/claude-code)See INSTALL.md for the full setup guide.
# 1. Clone this repo
git clone https://github.com/marian-kamenistak/claude-email-assistant.git
cd claude-email-assistant
# 2. Run the setup script
./setup.sh
# 3. Test it
./scripts/morning-triage.shclaude-email-assistant/
├── .claude-plugin/
│ └── plugin.json # Plugin manifest for marketplace
├── CLAUDE.md # Claude Code project instructions
├── INSTALL.md # Full installation guide
├── LICENSE # MIT license
├── setup.sh # Interactive setup script
├── screenshot-triage.png # Triage output screenshot
├── scripts/
│ └── morning-triage.sh # The scheduler script (launched by launchd)
├── launchd/
│ └── com.claude.morning-triage.plist # macOS scheduler config
├── skills/
│ └── reply-email/
│ ├── SKILL.md # Reply drafting skill definition
│ ├── email-voice-guide.md # Your email voice examples (you fill this in)
│ ├── learned-replies.md # Auto-populated from corrections
│ └── drafts-log.md # Draft tracking for the learning loop
└── shared/
└── communication-style.md # Your writing style rulesThis is the key differentiator. Most AI email tools are fire-and-forget. This one learns.
After ~30 corrections, the drafts start sounding like you actually wrote them.
The system uses labels to track email state:
| Label | Meaning |
|---|---|
AI ready | Draft created, waiting for you to review and send |
info | No reply needed (newsletters, notifications, FYIs) |
ops | Needs operational work (download invoice, sign doc, update system) |
high | Urgent / time-sensitive |
$ | Invoice or billing context |
Each morning, the system creates a calendar event with:
Edit skills/reply-email/email-voice-guide.md with 10-20 real emails you've sent. The more examples, the better the voice match. Include:
Edit shared/communication-style.md with your writing rules:
Edit the launchd plist to change when the triage runs. Default is weekdays at 8:30.
The morning triage script uses --dangerously-skip-permissions because it runs unattended via launchd. This flag allows Claude Code to execute tools without interactive approval prompts. Review the prompt in scripts/morning-triage.sh to understand exactly what it does autonomously.
gws CLI with Google Workspace API access.MIT
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.