Blog
Inline GitLab discussions, Bitbucket CI checks, and forge-agnostic intelligence: GitWand v2.14
v2.14 closes the ForgeNotImplementedError stubs from v2.10. GitLab gets diff-line comment anchoring via the Discussions API, Bitbucket gets CI status checks and draft-to-ready conversion, all three forges get updateComment and deleteComment, and the conflict preview and hotspot analysis become forge-agnostic. Plus a side-by-side with Linear Diffs, which launched the same week.
AI code review in your PR diff: how GitWand v2.13 works
Per-hunk AI critique with a four-tier verdict (ok / nit / suggestion / risk), an inline suggestion editor that stages GitHub suggestion blocks without leaving the diff, and named AI prompt presets for commit messages. The architecture, UX decisions, and prompt engineering behind all three.
Read article →GitWand now works with GitHub, GitLab & Bitbucket
v2.10 breaks out of the GitHub silo. A ForgeProvider abstraction routes the entire PR panel to the right backend with zero config — GitLab via the glab CLI, Bitbucket via REST v2 and OS keychain App Passwords, and a multi-account registry for personal + work accounts. Plus auto-detection from the remote URL, lazy-loaded provider chunks, and an honest account of what's still GitHub-only.
Launchpad: GitWand's cross-repo dashboard
The v2.9 release ships a single full-screen view aggregating WIP, PRs, Issues, and Team activity across every repo in your workspace. Four tabs, pin and snooze with localStorage, ⌘L from anywhere, lazy Team tab to keep first open snappy, and the design pass that finally aligns the panel with the rest of the app.
Why we made LLM resolution opt-in (and how): GitWand v2.5
The new llm_proposed pattern sits at priority 998, off by default. Why opt-in, why the v2.4 post-merge validator is the gate, why @gitwand/core still ships zero fetch() calls — and the MCP path that needs no API key.
Hooks, workspaces, agent sessions, and automations: what went into GitWand v2.7 and v2.8
Git hooks manager, multi-repo workspaces, worktree quick-create (⌘⇧N), cross-platform AI agent detection with lsof and /proc/cwd, a daemonless automation scheduler, and conflict resolution memory — six features across two releases.
Read article →The state of automatic merge conflict resolution in 2026: a survey, and where GitWand is headed
Five families of techniques, from RCS in 1986 to ConGra in 2024 — textual diff, AST-based structural merge, semantic merge, refactoring-aware tools, and LLMs. What's in GitWand today, and the v2 roadmap that follows the literature.
Read article →Pairing Claude Code with GitWand: letting AI agents ship without the merge nightmare
How GitWand's MCP server closes the gap between an AI agent that can write code and one that can actually merge it — the preview-first pattern, the auto-resolve handoff, and where it still breaks.
Read article →Contributing to open source with a fork: a GitWand walkthrough
The fork → clone → upstream → PR workflow, step by step. How GitWand's triangular-workflow badge surfaces the right information at the right time — something GitHub Desktop still doesn't do.
Read article →How often does GitWand's auto-merge get it wrong? A catalog of known failure modes
Honest per-pattern catalog of where the conflict classifier can be wrong, the structural safeguards that keep the blast radius small, and the design trade-offs behind each one.
Read article →Splitting a commit by hunks: what went into GitWand v1.7.0
How I added a hunk-level commit splitter to GitWand, why merge commits needed a hard block, and the patch-header bug that only surfaced on file creations.
Read article →Worktrees, submodules, and a broken auto-updater: what went into GitWand v1.6.3
How I added Git worktree and submodule management to a Tauri desktop app, and fixed four silent bugs that were keeping auto-update from ever working.
Read article →Why I built another Git client
There are already a dozen Git GUIs. Here's what was missing from all of them — and why I spent a year building GitWand anyway.
Read article →How I built automatic merge conflict resolution: pattern classification and composite confidence scoring
Pattern-based engine that auto-resolves trivial Git merge conflicts using classification, composite confidence scoring, and format-aware resolvers for JSON and Markdown.
Read article →