The Autonomous Agent Workspace Kit

Everything needed to build a Claude Code agent that runs itself: a Telegram bridge you talk to from your phone, an Obsidian-vault memory it reads and writes, scheduled headless workers that finish queued work unattended, subagents with persistent memory, a self-maintaining skill library, and a local mission-control dashboard.

This is not a course about agents. It is the extracted, sanitised working system — the same files, with the secrets removed and the paths turned into placeholders.

Overview page: https://outlier1804.github.io/agent-workspace-kit/

Start here

Read SETUP-GUIDE.md top to bottom. It is the product. Part 1 gives you a single command that copies everything into place and fills in every placeholder; Part 4 is where the agent first replies on your phone (~25 minutes in).

If something breaks, TROUBLESHOOTING.md is organised by the symptom you'll actually see, not by the cause you'd have to already know.

What's in the box

`install/` — the running system

File	Lines	What it is
`bridge/bridge.py`	~395	Telegram ↔ `claude -p`. Stdlib only. Session rotation, auto-heal, attachment handling, per-message budget cap, allowlist.
`bridge/.env.example`	—	Every setting, documented inline. Copy to `.env`.
`dashboard/server.py`	~638	Localhost mission control. Parses the vault into `/api/state`, plus POST controls for chat, queue and worker runs. Stdlib only.
`dashboard/static/index.html`	~625	The entire UI. Inline CSS + JS, no build step, 7 tabs.
`agents/worker-prompt.md`	~90	What the autonomous worker does each run. The engine of the whole thing.
`agents/run-worker.sh`	—	Runner: lock file, idle exit, budget cap, result parsing into `runs.log`.
`agents/curator-prompt.md`	~45	The weekly maintenance procedure, with its three safety invariants.
`agents/run-curator.sh`	—	Curator runner, including the guard that skips while the worker holds its lock.
`agents/notify.sh`	—	Agent → human message, mirrored locally so a report never lives only in chat history.
`launchd/*.plist`	4 files	Bridge (always on), worker (every 2h), curator (weekly), dashboard (always on). All with the PATH block that stops silent failures.

`claude/` — the agent configuration

CLAUDE.md — the constitution template: memory layout, queue contract, autonomy boundaries, hard lines. Includes a written-out guide to the full-autonomy carve-out, which is the single highest-leverage thing in the setup.
SOUL.md — personality and voice, deliberately a separate file from operating rules.
agents/ — six subagent definitions with persistent memory: researcher, dev, marketing, publisher, qa, curator.
skills/ — three real Claude Code skills, each one a procedure that was paid for in wasted hours:
- debug-claude-subprocess — why headless claude -p exits 1 with an empty stderr, and the resumed-session death spiral that bricks a chat channel permanently.
- schedule-agent — the complete pattern for a new scheduled agent, plus the PATH trap, the lock/dedupe guards, and how to verify a schedule without burning an agent run.
- extend-dashboard — adding a panel without breaking it, including the 5-second-poll trap that freezes the UI with no console error.

`vault/` — the starter memory

A working Obsidian vault skeleton: Home, Queue, Inbox, Ideas MOC + an idea template, Skills MOC, Earnings ledger, Content Log, Agents/Team.md, a memory note per subagent, a daily-log format, and optional Opportunities/ + Platforms/ folders — all in the shapes the dashboard already parses.

vault/Lessons.md ships pre-seeded with eight real operational lessons from running this system — the session cost spiral, verify-delegated-work-before-reporting, "an API returning success is not proof the action happened", and others. They are the most immediately useful thing in the kit.

Requirements

macOS (scheduling uses launchd; a systemd equivalent for Linux is in Part 5 of the guide)
Claude Code, installed and logged in
Python 3.9+ — no third-party packages, ever
A Telegram account
Obsidian (optional, recommended)

Two honest notes

Running an agent costs money. Every scheduled run is a Claude invocation. Part 11 of the guide is entirely about cost control — per-invocation budget caps, session rotation, idle exits, and the biggest lever: moving deterministic work out of agents and into plain scripts.

The four hard lines stay human. Creating accounts, entering passwords, entering payment or payout credentials, and spending money are kept on your side of the boundary by design, and the shipped prompts enforce it. That's not a limitation of the kit — it's the reason it's safe to let the rest run unattended.

License

Personal and commercial use in your own projects, including client work. Don't resell or redistribute the kit itself as a product.

agent-workspace-kit