AI workspace
Set up and maintain a dedicated AI workspace, a folder separate from code where thinking artifacts compound across sessions. Use when the user wants to organize folders for AI work, asks where non-coding sessions should live, wants project insights extracted into their thinking folder, or complains that their agent keeps drowning in code-level context.
Overview
- Visibility
- public
- Updated
- 17h ago
- Built from
- 0 sessions
Install
An Agent Skill is a SKILL.md plus its files, zipped as .skill. Pick an agent — it activates when its description matches your request.
User-level (~/.claude/skills/)
unzip ~/Downloads/ai-workspace.skill -d ~/.claude/skills/Project-level (.claude/skills/, run from repo root)
unzip ~/Downloads/ai-workspace.skill -d .claude/skills/AI workspace
An AI workspace is a folder, separate from every code repo, where the only artifacts are outcomes of thinking: essays, decisions, research notes, analyses. A folder is an environment: start the agent in a code repo and it thinks in code-level edits; start it in a clean workspace and it can think big-picture. Because every session writes files into the same folder, the next session starts richer: the workspace compounds.
Setting up
- Create the split. Two top-level folders:
workspace/for code (projects, repos) andai-space/for thinking, divided intoai-space/work/andai-space/personal/. Existing repos stay where they are; only the thinking artifacts move. Done when both trees exist and the user has confirmed the names. - Seed the environment. Add an AGENTS.md at the root of
ai-space/stating what the folder is for, what lives where, and that sessions should write their artifacts (notes, decisions, analyses) as markdown files into the relevant subfolder. Done when a fresh session started inai-space/could orient itself from that file alone.
Extracting insights from a code repo
When the user wants project knowledge available in the workspace ("extract insights about this project", "get replay's auth decisions into my thinking folder"):
- Explore the repo read-only and write a markdown summary of the requested aspect (decisions made, constraints, open questions), not a code walkthrough.
- Save it into
ai-space/work/<project>/. Done when the file exists in the workspace and contains no content that requires reading the repo to understand.
Answering questions that need the code
From inside ai-space/, when a question needs facts from a repo, send a subagent (or a scoped read) into the repo path, bring back only the answer as a markdown note in the workspace, and continue the session in the workspace. The session's working context stays high-level; the repo visit is an errand, not a move.