Workflow

End-to-end flow, from picking an issue to merging the PR. Everything runs inside the terminal multiplexer; the dashboard orchestrates and the AI agent does the heavy lifting at two distinct stages.

  1. Diagram
  2. Stages
  3. Supported tools per stage

Diagram 

flowchart TB
    subgraph Term["Terminal · tmux / cmux"]
        direction TB
        Dashboard(["① Dashboard<br/>pick issue · add context"]):::action
        Editor["Editor<br/><sub>zed · cursor · vscode · nvim · jetbrains</sub>"]:::tool
        AI(("② / ⑤ AI agent<br/>Claude Code")):::agent
        Pager["③ Diff pager<br/><sub>delta · diff-so-fancy · built-in</sub>"]:::tool
        Decision{good?}:::decision
        Ship(["④ git commit + PR"]):::action
        GitHub[("GitHub<br/>PRs · CI")]:::external

        Dashboard ==>|implement| AI
        Dashboard -. write context .-> Editor
        Editor -. context · hand-edit .-> AI
        AI ==> Pager
        Pager ==> Decision
        Decision -->|hand-edit| Editor
        Decision -->|resume| AI
        Decision ==>|ship| Ship
        Ship ==> GitHub
        GitHub -->|reviews · CI| AI
        AI -->|push fix| Ship
    end

    classDef action fill:#dbeafe,stroke:#3b82f6,stroke-width:2px,color:#1e3a8a
    classDef tool fill:#f3f4f6,stroke:#9ca3af,color:#374151
    classDef agent fill:#ede9fe,stroke:#8b5cf6,stroke-width:2px,color:#4c1d95
    classDef external fill:#fef3c7,stroke:#f59e0b,stroke-width:2px,color:#78350f
    classDef decision fill:#fee2e2,stroke:#ef4444,stroke-width:2px,color:#7f1d1d

Bold edges (==>) follow the happy path; thin edges (-->) are branches/loops; dotted edges (-.->) are optional editor side-trips. Colors group nodes by role: blue actions, purple agent, gray tools, yellow external service, red decision.

Stages

  1. Pick + add context — open the dashboard, browse assigned issues, pick one. The inline context box lets you add custom instructions to the prompt; Ctrl+O drops into your editor for longer prose.
  2. AI agent (Claude) — runs in the worktree’s mux window with the rendered ticket + your context. Implements or plans.
  3. Diff pager (review) — [v] opens the diff overlay. From here you iterate: hand-edit in your editor, or resume the Claude session to keep going.
  4. Ship — once the diff looks right, [C] for inline commit + push, then [c] to create the PR (--fill runs Claude non-interactively against the PR template).
  5. PR feedback loop — [f] (fix) and [r] (self-review) re-launch the AI with PR comments + CI failures as input. Patches are pushed; CI re-runs; loop until merged.

Supported tools per stage

Stage Supported today Planned
Issue tracker (① Dashboard) Linear, GitHub Issues Jira, Shortcut, etc.
AI agent (②, ⑤) Claude Code (claude CLI) OpenAI Codex, OpenCode, Cursor agent
Diff pager (③) delta, diff-so-fancy, any unified-diff pager — built-in colorizer when none set —
Editor (side branch) Anything taking a path: zed, nvim, jetbrains, cursor, vscode. cursor + vscode also handle .code-workspace files via [E] workspace shortcut —
Forge (④) GitHub via gh CLI GitLab, Bitbucket, Gitea
Terminal multiplexer (outer frame) tmux, cmux (macOS-only — see #1472), none zellij

The AI agent and issue tracker are already behind interfaces (lib/trackers/, plus the claude CLI shim in lib/ai.ts). Adding a new option in either category means writing one new module and wiring it into the factory — no changes to UI/dashboard code.