agentic-workspace

Agentic Workspace adds a small repo-native operating layer for repositories where agent work must preserve intent, context, proof expectations, and handoff state across sessions, tools, branches, or contributors.

It is built around checked-in workspace contracts, compact routing, explicit ownership, selected modules, and thin adapter files that point agents at the right repo-local commands.

Why it exists

Agents can move quickly, but they enter and leave a repository with partial context. In longer-running work, important state often lives only in chat: why a task matters, what has already been tried, which repo facts matter, what proof is expected, and what remains unsafe to claim.

Agentic Workspace gives that operating state a repo-native home so future agents can start from the repository instead of reconstructing intent from conversation history.

What it does

Agentic Workspace helps agents:

• start from compact repo state instead of broad rereading;
• preserve durable lessons that would otherwise be rediscovered;
• keep active work bounded, resumable, and honestly closeable;
• route proof expectations and evidence through reviewable surfaces;
• hand off unfinished or risky work without hiding uncertainty.

It does this through compact command surfaces, selected module state under .agentic-workspace/, thin adapter files such as AGENTS.md, and file-based contracts that can be generated, packaged, or projected into agent-facing integrations.