jasonsuhari/gridbash

Problem Terminal compatibility reports are most useful when they capture the same launch, input, resize, mouse, copy, and shutdown evidence. Today contributors have no reusable smoke-test checklist, so reports are difficult to compare across Windows, Linux, macOS, terminals, shells, and multiplexers. Requested behavior Add docs/TERMINAL_SMOKE_TEST.md with a concise manual checklist and a copyable report template. Link it from CONTRIBUTING.md and use it to record one real terminal/OS test result in the pull request description. Likely files docs/TERMINAL_SMOKE_TEST.md (new) CONTRIBUTING.md Acceptance checks The checklist covers version, OS, architecture, host terminal, shell/profile, and exact launch command. It checks first launch, keyboard input, selected-pane routing, resize, scroll, mouse selection/copy, worktrees when available, pane restart, and clean quit. Unsupported environments and the current single-process limitation are not presented as bugs. The report template avoids c

Problem The README shows launch commands, but a new user does not have an end-to-end example of using separate agent panes for implementation, review, and tests. That makes the core GridBash workflow harder to learn and harder to demonstrate accurately. Requested behavior Add a short, platform-neutral tutorial at docs/tutorials/first-multi-agent-workflow.md and link it from the README. The tutorial should take a user from a clean sample repository to a three-pane grid with isolated worktrees and distinct implementation, review, and test roles. Likely files docs/tutorials/first-multi-agent-workflow.md (new) README.md Optionally one small diagram or screenshot under docs/assets/ if it materially clarifies the workflow Acceptance checks The prerequisites and exact launch command are explicit. The tutorial gives each pane a safe, concrete role and example prompt. Focused and selected-pane input are both demonstrated. Worktree behavior, quitting, and the current lack of live-process reat

Problem GridBash has a growing CLI (resume, profile/config flags, agent API/MCP, and worktrees) but ships no PowerShell, bash, zsh, or fish completions and no generated command reference. Requested behavior Generate completions from the Clap definition and document installation without hand-maintaining option lists. Likely files src/cli.rs and the smallest appropriate command entry point Cargo.toml / Cargo.lock if a completion helper is needed docs/REFERENCE.md npm packaging files only if generated completions are shipped in the package Coordinate the output location and packaging decision in the issue before implementation. Acceptance checks PowerShell, bash, zsh, and fish completions are generated from the current CLI. Completion generation has a deterministic test or CI drift check. README or reference docs show install/use paths for packaged and source installs. Hidden and internal options are handled intentionally. Generated files, if committed, have one documented regeneration

Problem Most UI tests exercise helper strings and rectangles, but there is no end-to-end Ratatui buffer coverage for help, settings, recovery, rename, follow-up, image, and grid-resize overlays at representative terminal sizes. Layout regressions can compile and pass unit tests while clipping controls or cursor state. Requested behavior Render deterministic app fixtures into TestBackend buffers at 80x24, 120x40, and a supported narrow size, then assert essential controls remain visible and in bounds. Starting points src/composer.rs contains existing TestBackend examples. src/ui.rs owns most rendering behavior. src/app.rs owns modal and interaction state used by the renderer. src/layout.rs contains shared layout calculations. Keep production refactors out of the first pull request unless a small seam is required to construct deterministic fixtures. Acceptance checks Help, settings, and recovery overlays have buffer-level tests. Modal bounds never exceed the terminal. Essential close/

Problem The packaged README links to repository-relative files such as docs/V1_ACCEPTANCE.md, docs/DAEMON_ARCHITECTURE.md, roadmap/demo assets, and the social preview, but package.json ships only selected docs. npm pack --dry-run confirms several linked targets are absent. Requested behavior Either include every README-linked artifact needed by npm consumers or use canonical GitHub/Pages URLs for repository-only assets. Add a focused package test so the mismatch cannot return silently. Likely files README.md package.json npm/scripts/gridbash-launcher.test.js or a small focused package test beside it Avoid adding large demo-source or video directories solely to satisfy a link. Acceptance checks An extracted root npm tarball has no broken relative README links. Repository-only links use stable canonical GitHub or Pages URLs. Package size does not accidentally grow with large source/demo assets. A launcher/package test compares README-relative targets with packed files. npm and GitHub