A desktop (Electron) GUI that wraps agentic coding-harness CLIs (Claude Code, Codex, Cursor, OpenCode) — workspace-organized chats, reusable static context, multi-harness model selection, and first-class visibility into cost and the files the agent changes. NAC Code is a wrapper, never a harness (no agent loop, no raw model endpoints).
Status: pre-implementation. The engineering plan and core specs are complete, and cross-provider context portability is validated by a spike. Next: scaffold the Electron app (M0-7), then the shell (M1).
Start with docs/DECISIONS.md — the current state, next step, and locked decisions (the source of truth across devices). Then read the plan and specs.
| Path | What |
|---|---|
docs/DECISIONS.md |
Living decision log / current state — read first |
docs/plans/ |
Master engineering plan (scope, architecture, milestones) |
docs/specs/ |
Closed specs (agent runtime + adapter interface; cross-provider context) |
docs/research/ |
Investigations (t3code wrapping patterns; OpenCode/pi carrier evaluation) |
docs/reviews/ |
PRD / handoff reviews |
docs/design/ |
Design reference: PRD, interactive prototypes (*.dc.html), screenshots |
spikes/ |
Throwaway validations (e.g. cross-provider context portability) |
AGENTS.md / CLAUDE.md |
How an agent should work in this repo + keep docs updated |
Full docs index: docs/README.md.