Your AI chief of staff for inbox, calendar, and follow-ups.
One installable plugin for Claude Cowork, Claude Code, and Cursor that triages your inbox, drafts replies in your voice, tracks follow-ups, preps your meetings, runs a morning briefing, and manages your tasks and memory.
It drafts everything for your review and never sends, books, or spends on your behalf. Instructions buried in emails, invites, or pasted docs are never obeyed — they're surfaced for you to confirm. You glance, edit, and send. That's the whole promise.
- Browse the examples gallery — pick a persona and see before/after draft demos
- Run
/assistant:startfor a fast tasks + memory bootstrap, or/assistant:setupfor the full profile interview - Paste a demo thread →
/assistant:inbox triage→ compare your draft to the reference
No connectors required. Eval maintainers: see evals/demo/first-run-script.md (Alex Rivera eval persona).
| Job | Command |
|---|---|
| Configure it to you (10-min interview) | /assistant:setup |
| Fast tasks + memory bootstrap | /assistant:start |
| Morning briefing | /assistant:brief |
Bring the assistant up to date (tasks · memory · --all) |
/assistant:update |
| Weekly report | /assistant:report |
| Set up scheduled tasks | /assistant:schedules |
| Job | Command |
|---|---|
| Triage inbox + draft replies | /assistant:inbox triage (default) |
| Lighter inbox sweep | /assistant:inbox sweep |
| Draft one reply | /assistant:email draft |
| Review what's awaiting a response | /assistant:email review |
| Add / review / sync tasks | /assistant:tasks add · review · sync |
| Prep for meetings / follow up on notes | /assistant:meeting prep · follow-up (default) |
Plus skills that fire on their own as you work: calendar scheduling, meeting follow-up from pasted notes, and two-tier memory that decodes your shorthand ("ask todd re oracle").
Claude Cowork or Claude Code: Customize → Plugins → add a marketplace from this repo's GitHub URL (https://github.com/daddia/my-assistant), then install My Assistant.
Cursor: Settings → Plugins → Add plugin → paste https://github.com/daddia/my-assistant, then install My Assistant.
| What you see | Value |
|---|---|
| Display name | My Assistant |
| Repo & marketplace | my-assistant |
Plugin id (manifest name) |
assistant |
| Working folder | ~/MyAssistant (configurable at setup) |
| Profile | ~/MyAssistant/config/profile.md |
| Policies | ~/MyAssistant/policies/*.policy.md |
| Install config | ~/MyAssistant/config/my-assistant.json |
| Commands | /assistant:* |
Then run:
/assistant:setup
A short interview captures who you are, how you write, your VIP tiers, and your email and calendar policy. Setup creates your working folder (default ~/MyAssistant) and writes config/profile.md, policies/*.policy.md, and selective config/my-assistant.json — outside the plugin, so /plugin update never overwrites it. Policy templates ship in the plugin's policies/ directory.
Every skill is standalone-first: paste an email thread or your calendar and it just works. Connect Slack, Notion, GitHub, or your own MCP servers for ~~email, ~~calendar, and ~~drive to work directly against your accounts. See CONNECTORS.md.
- Scheduled tasks (
/assistant:schedules) — guided decision tree for local Cowork schedules, Claude Code cloud schedules, or managed agents. Canonical job list:scheduled/schedules.yaml. - Managed-agent cookbooks (
managed-agents/) — the same jobs on Anthropic's infrastructure, immune to a sleeping laptop. Advanced/optional. - Reliability guide — Always-on reliability: when to stay local, when to escalate, and how missed runs surface in chat via
scheduled/.
Open {assistantPath}/dashboard.html in Chrome or Edge for a board/list view of TASKS.md, a browser for AGENTS.md + memory/, and a Review tab for pending approvals. Copied into your working folder at setup; or open skills/dashboard.html from the plugin and point it at your working folder.
Plain English — skills, rules, and a profile you can read and edit. See AGENTS.md for the trigger→skill map and the docs for the full guide.
| Topic | Guide |
|---|---|
| Privacy in plain language | Protect your privacy |
| Deploy paths, profile, working folder | Admin & deployment |
| Prove "works with X" without OAuth | Connector smoke tests |
| Security deep dive | security/README.md |
To verify triage, drafts, injection defence, and connector categories against fixtures, see the proof harness.
Released under the MIT licence — you're free to use, modify, and distribute this project.
Copyright (c) 2026 Jonathan Daddia