Browse / General-purpose / autonomous-workspace-template

autonomous-workspace-template

levelupself

Self-driving AI development workspace.

Open levelupself/autonomous-workspace-template → Does not match the four rules.

Built for Claude Code

ships .claude CLAUDE.md and nothing for other agents.

In its own words

Aboutlevelupself/autonomous-workspace-template

Agents plan, build, and document software overnight. You review in the morning.

README.md

A self-driving project workspace where AI agents plan, build, and document software overnight. You review in the morning and make a small number of decisions.

What this is

A git repository that is simultaneously:

  • A Claude Code agent workspace — agents run here autonomously
  • An Obsidian vault — every doc, decision, and metric is human-readable
  • A project management system — stages, burndown, token tracking built in
  • A self-improving tool library — agents create new skills when they get stuck

Open home.md in Obsidian to see everything.

Quick start

Windows (zero prereqs — opens PowerShell):

After bootstrap completes:

Four defense layers (safety)

Run task hooks:test at any time to verify all 37 safety checks pass.

Planning pipeline (new projects)

Nothing executes until you approve the plan.

Supervision and self-improvement

When an agent gets stuck:

  • Writes structured blocker to docs/blockers.md
  • post-blocker-write.sh hook wakes the supervisor agent
  • Supervisor classifies: capability gap → skill-writer, knowledge gap → research, scope → you
  • Skill-writer creates a new SKILL.md and commits it
  • Next session auto-discovers the skill and continues

The .claude/skills/ library grows with every overnight run.

Token tracking

Every session the stop-write-back.sh hook records:

  • Tokens used (input + output)
  • Cost in USD
  • Stop reason (complete / usage_limit / error)

The repository's own description, then 52% of its README. Read the rest on GitHub ↗

Read from the repository's git tree on 2026-08-17.