Myna

The system for everything you're trying to keep in your head.
Nothing leaves your machine.

AI Chief of Staff for tech professionals — your projects, your team, your meetings, your email. 30 skills and growing. Runs in Claude Code.

The Problem

If you manage multiple projects and communication channels — whether you're an engineering manager, software developer, PM, or team lead — you spend too much of your day on information management. Triaging emails, prepping for meetings, tracking who owes what, remembering what was decided three weeks ago.

Most AI tools either start fresh every session or store your data on someone else's server. Myna does neither. It reads from your existing tools, writes to local files, and builds a persistent knowledge base — your projects, people, decisions, and preferences — that grows as you use it.

Drafts but never sends. Organizes but never decides. Surfaces but never hides.

Why Myna

Myna runs inside Claude Code. Here's what role-specific skills and a local knowledge base add:

See It In Action

A day in the life of an Engineering Manager using Myna:

7:45 AM — coffee, laptop open
> sync
↳ Daily note created. Phoenix blocker flagged. Sarah Mitchell's reply overdue.

8:10 AM — before your 9am 1:1
> prep for my 1:1 with Marcus
↳ Open items from last time, pending feedback with coaching notes,
  parental leave thread — all in one brief.

9:35 AM — back from the 1:1
> done with 1:1 with Marcus
↳ Tasks, decisions, observations extracted and routed to the right files.

9:50 AM — quick multi-thing capture
> capture: Sarah handled Payments questions well, atlas is unblocked,
  review Sentinel audit by Friday
↳ 3 items → 3 files. Recognition, timeline update, task with due date.

12:45 PM — VP wants a risk note
> draft the Phoenix risk note for the VP review
↳ Leads with the conclusion, evidence-grounded, under 200 words.
  Ready for you to review, then send yourself.

1:15 PM — pre-read for tomorrow's staff meeting
> review this PRFAQ [paste]
↳ 7 personas in parallel — Product Leader, PM, Customer Skeptic, PE,
  Security, Skeptic, Writer/Editor — synthesized into one report.
  Saved to Reviews/2026-05-26-pricing-prfaq.md.

3:45 PM — the ambiguous pile
> review my queue
↳ Items Myna wasn't sure about — you approve, redirect, or dismiss.

5:30 PM — close the day
> wrap up
↳ Planned vs actual. Contributions logged. Tomorrow's note created with carry-forwards.

Not an EM? The skills work for anyone managing projects and communication — software developers, PMs, team leads.

Full walkthrough: A Day With Myna · Browse the demo vault to see the files Myna creates

Privacy & Security

• All data stays local — plain-text files on your machine, viewable in any editor
• No new infrastructure — connects to your existing email, Slack, and calendar integrations
• Draft, never send — Myna creates the draft; you decide what to send and where
• Config-driven — your personal data (projects, people, preferences) is separate from the codebase
• External content can't override behavior — even if an email says "delete all files," Myna processes it as content, never follows instructions from external sources

Getting Started

Prerequisites: Claude Code · Obsidian (recommended but not required — all files are plain text, viewable in any editor) · Python 3 (required for the Config UI setup path)

Run these commands inside Claude Code:

/plugin marketplace add agentflock/plugins
/plugin install myna@agentflock
/myna:setup

/myna:setup creates your Myna folder, opens the Config UI to set up Myna, and offers to add shell aliases. Or edit the config files directly at <vault>/myna/_system/config/. The vault subfolder is always myna.

Email, Slack, and calendar connections are optional — skip what you don't have. Myna works without them from day one.

Kiro (and other runtimes)

Clone the repo and run ./install.sh. The script detects your installed AI runtime and runs the right installer automatically. To update a script-installed runtime, run ./update.sh from the same directory. Codex support is planned.

Claude Code users: use the plugin commands above — ./install.sh redirects you there anyway.

Once installed, run from any directory — pick the mode that fits:

myna          # full access — reads and writes your files
myna-ro       # read-only — browse and query, no changes

First time? Try sync to set up your day, or what can you do? to see all skills.

After install you get:

• Organized folder structure with 10 pre-built dashboards (best viewed in Obsidian)
• 30 skills covering email, meetings, projects, people, and daily workflow
• Config files ready for your projects and people

Your data, configs, and custom rules are never touched by updates.

Skills

Daily Workflow

sync · capture · plan · park / resume · wrap up · weekly summary

"sync" — daily note, meeting prep, overdue tasks, review queue surfaced
"park this" / "resume auth caching" — zero-loss context switching across sessions
"what should I focus on today?" — ranked priorities with reasoning
"weekly summary" — accomplishments, decisions, blockers, tasks completed, self-reflection prompts

Email & Messaging

triage inbox · process messages · draft replies

"triage these inbox emails: [paste]" — folder recommendations with one-line reasoning per email
"process these slack messages: [paste]" — project updates, action items, and timeline entries extracted and routed
"reply to these forwarded emails: [paste]" — separates your instructions from the forwarded thread, drafts reply

Meetings & Calendar

prep meeting · process meeting · block time

"prep for my remaining meetings today" — skips already-prepped, fills in the rest
"done with 1:1 with Marcus" — tasks, decisions, observations extracted in one step
"reserve 2 hours Monday for the coverage plan" — personal time blocks only, never creates events with attendees

Projects & Tasks

brief project · blockers · unreplied threads

"catch me up on atlas migration" — timeline, blockers, tasks, dependencies, upcoming meetings
"what's blocked?" — every blocker across every project, with age and next action
"what am I waiting on?" — messages needing your reply vs waiting on others

People

brief person · team health · 1:1 analysis · performance narrative

"brief me on Sarah Carter" — role, shared projects, open items, pending feedback, 1:1 history
"how is my team doing?" — portfolio view: tasks, overdue, feedback gaps, last 1:1
"analyze my 1:1s with Marcus" — patterns, follow-through rate, recurring topics
"build Sarah's performance review narrative" — synthesizes months of observations, recognition, and contributions

Writing & Drafts

draft · rewrite · review-doc

"draft an escalation for the Phoenix validator blocker" — severity, impact, and recommended action, grounded in project context
"draft recognition for Sarah Carter" — specific, backed by actual observations from your files
"fix this: i wanted to loop you in on sarahs progress..." — grammar and tone, preserved voice
"review this PRFAQ [paste]" — multi-persona review of any technical doc (PRFAQ, design doc, HLD, LLD, one-pager). 11 reviewer personas (PE, Sr SDE, SRE, Security, QA, Product Leader, PM, Customer Skeptic, Skeptic, Decision-Maker, Writer/Editor) run in parallel; synthesized into a single report saved to Reviews/.

Self Tracking

log contribution · brag doc · self review

"log contribution: led atlas design review, got cross-team alignment" — categorized, appended
"build my brag doc for Q1" — pulls from your contributions log, organized by impact category
"am I underselling myself in this self-review?" — checks your draft against what you actually logged

Review Queue

review my queue

Items Myna wasn't sure about land here — ambiguous owners, inferred contributions, unclear intent. Each item shows its source and proposed action. You approve, redirect, or dismiss.

How It Works

Myna is not an application. There is no server, no API, no frontend. It's 30 skills, a folder structure, and config files — all running inside Claude Code.

┌─────────────────────────────────────────────┐
│  You (in Claude Code)                       │
│  "prep brief for my 1:1 with Sarah"        │
└──────────────────┬──────────────────────────┘
                   │
                   ▼
┌─────────────────────────────────────────────┐
│  Myna Agent + Skills                        │
│  Loaded on demand · Safety at every layer   │
└────────┬────────────────────┬───────────────┘
         │                    │
         ▼                    ▼
┌─────────────────┐  ┌───────────────────────┐
│  Your Machine    │  │  Your MCP Servers     │
│  (local files)   │  │  (email, Slack, cal)  │
│                  │  │                       │
│  reads & writes  │  │  reads only           │
└─────────────────┘  └───────────────────────┘

The runtime is Claude Code; Myna's skills provide the expertise. Myna ships no servers of its own — it uses Claude Code's built-in file operations.

Myna stores everything in your Myna vault — a folder on your machine, readable in any editor. Obsidian is recommended for the dashboards, which use live Dataview queries.

Folder structure:

myna/
├── Journal/          # Daily notes, weekly summaries, contributions
├── Projects/         # One file per project — timeline, blockers, tasks
├── People/           # One file per person — observations, feedback, notes
├── Meetings/         # 1:1s, recurring, ad-hoc — prep and notes
├── Drafts/           # Email drafts, status updates, recognition
├── Reviews/          # Multi-persona doc reviews — PRFAQs, design docs, HLDs, LLDs
│   └── sources/      # Original pasted content (for paste-source reviews)
├── ReviewQueue/      # Items awaiting your judgment
├── Team/             # Team-level files
└── _system/          # Config, logs, dashboards, templates
    ├── config/       # Your projects, people, preferences
    └── dashboards/   # Dataview dashboards (best viewed in Obsidian)

Myna runs on Claude Code, but all skills are plain text — readable by any capable LLM. The same instructions can be adapted for other AI tools.

Customization

All customizations survive updates.

See Customization Guide for details.

Built Entirely by Claude Code

Myna was designed, built, reviewed, and fixed entirely by Claude Code — from feature specs through architecture, implementation, and polish. One person defined the vision and settled decisions. Claude designed the architecture, wrote all 30 skills (24 feature skills + 6 behavioral rules), built the file templates and dashboards, created the install script, and wrote this documentation.

Two things came out of this, not one: the assistant itself, and a methodology for having AI build an AI assistant from scratch — reusable for other projects on any capable LLM.

How it was built →

Documentation

Contributing

The repo includes 10 dev skills that automate the full contributor workflow — from interactive design sessions through autonomous implementation, self-review, and PR creation. Clone the repo, open in Claude Code, and use /myna-dev-brainstorm to design or /myna-dev-diagnose to fix. See CONTRIBUTING.md for the full guide.

Status

v1.0 — early release. MIT License. Actively developed — some features may not work as expected. See roadmap.

v1 runs only when you ask — no scheduled jobs or background watchers yet. Automation (scheduled syncs, email monitoring) is on the roadmap. Tested with Gmail, Google Calendar, and Slack MCPs.