mrkeyoor.comAsk

dominikmartn/harptested 16 Nov 2026commit 31eed50

harp review

Harp is worth considering for macOS developers whose main problem is finding which coding-agent session needs input.

  • Safety failed
  • Build timed out
83 sto install
  • Tests failed
  • 1,409 packages
Dev Toolsevaluationupdated 16 Nov 2026
+5stars / 7d

Harp is dominikmartn's unofficial fork of the Warp terminal. It replaces the tab bar with a sidebar for Claude Code and Codex sessions, with agent states, named groups and provider usage rings.

Verdict

Harp is worth considering for macOS developers whose main problem is finding which coding-agent session needs input. Issue #9 supplies a concrete positive account of that use. Choose another option if you require a released Windows port, formatted agent replies or an entirely local service model. Our source build timed out at 900 seconds and the test command failed after 818 seconds; those outcomes do not supply a successful desktop result.

Harp infographic: find which coding-agent session needs input, plus its uses, our test result, pitfalls and who it suits
Tested 16 November 2026: the source build timed out. Before that: source install 83 s.

We ran it

Lab card: what happened when we ran harp
Lab card from our sandbox run
Screenshot of harp (github.com/dominikmartn/harp)
github.com/dominikmartn/harp
Install✓ · 83s1409 packages
Build✗ timed out · 900s
Tests✗ · 818sran, no count parsed
Repo6496 files~1,835,159 lines of source · 317.1 MB · 21 CI workflows

Answers from our run

Does harp build from source?

Dependencies installed in 83 seconds (1409 packages), and the build failed. We cloned commit 31eed50 into a clean Debian container with 3 CPUs and no project-specific setup.

Do harp's tests pass?

No. The project's own test command exited unsuccessfully after 818 seconds, and its output did not expose an individual test count.

Who should not use harp?

Windows users requiring a released port rather than the contributor-local implementation described in issue #6.

What are the alternatives to harp?

Warp, Claude desktop, Codex desktop. Harp is worth considering for macOS developers whose main problem is finding which coding-agent session needs input.

Setup2/5Our source checkout installed but the build timed out.Editor’s note: Supported macOS installation was not reproduced; dependency setup completed, but the source build timed out and the test command failed.
Docs4/5The README has an install section (“Install”) and usage examples in code; we did not run its quickstart ourselves.Editor’s note: The README describes installation and separate status and usage inputs; their execution on supported macOS was not assessed.
Community2/5Last push on 2026-11-10, 715 stars and maintainers answered or closed 0 of the 5 most-discussed issues we sampled.Editor’s note: Sampled issues provide concrete reports, but the evidence does not measure support responsiveness or adoption beyond those authors.
Maturity1/5Latest release v0.1.8 on 2026-10-29 (pre-1.0), repository created 2026-10-20 and its test suite did not pass in our sandbox.Editor’s note: Release and push records show activity; the supplied runs do not assess desktop stability or production reliability.

How we score

Who it’s for

macOS developers running several Claude Code or Codex sessions who want attention states in a sidebar.
Existing Warp users interested in a different session layout.
Contributors interested in the Windows-port work described in issue #6.

Who it’s NOT for

Windows users requiring a released port rather than the contributor-local implementation described in issue #6.
Developers choosing a terminal primarily for formatted agent replies.
Teams requiring entirely local operation, because inherited AI and cloud features connect to Warp servers.

Setup reality

Our 2026-11-16 sandbox run at commit 31eed50 installed dependencies in 83 seconds and recorded 1,409 dependencies. The build timed out at 900 seconds; the test command failed with exit 101 after 818 seconds. The supplied log tail contains compiler arguments and no explanatory diagnostic. The recorded configuration was 3 CPUs, 12 GB RAM and image label lab-rust:1.

The README documents a macOS DMG and macOS source-build scripts. Claude Code session status uses a bundled plugin; Claude plan-limit rings use status-line input. Existing status-line commands are left alone, and the documented Claude rings apply to Claude.ai plans rather than API keys.

A separate application attempt stopped at a safety step requiring a Compose file. Its summary nevertheless claimed deployment. The failed step is the concrete outcome, and teardown reported "teardown: ok".

Our Harp source build timed out after 900 seconds, while one user reports that its sidebar solved a problem when running several Claude Code sessions. Harp is worth considering for macOS developers who want to see which agent needs input. Choose another option if released Windows support, formatted replies or entirely local operation are requirements.

This review covers dominikmartn/harp, an unofficial fork of the Warp terminal. The Harp README describes a replacement for the tab bar, with session states and named groups. The decision is whether that layout addresses your daily work. The evidence combines the project's documentation, attributed GitHub reports and our source-harness results at commit 31eed50 on 2026-11-16.

What you can do with Harp

Harp replaces Warp's tab bar with a sidebar showing Claude Code and Codex session states. The README describes working, waiting and done states, unread marks, a background-work dot and a moon for a tab set aside. These indicators give you a place to look when several agents are running and you need to decide which session deserves attention next.

Named groups add another way to organize the same sessions. With tmux integration enabled, the README says group membership and order are stored on tmux sessions and survive a restart. Tabs attached to tmux also follow session names and read or inactive flags. That makes the feature relevant to developers who use tmux across clients, including sessions started over SSH from another device.

The footer answers a different question: account usage. The README describes two rings per provider, covering the 5 hour window and weekly limit, with numbers and reset times on hover. Codex readings come from session logs; Claude readings come from Claude Code's status line. Session attention and provider usage therefore have separate inputs, even though both appear in the same sidebar.

What people use harp for

The author of issue #9 reports that Harp's sidebar helped while running several Claude Code sessions at once. The other concrete use account in the supplied synthesis comes from a contributor developing a Windows port. These accounts serve different readers: one concerns everyday agent management, while the other concerns implementation work on window behavior, sidebar controls and platform integration.

Who What they did Reported outcome
The user who opened issue #9 Ran several Claude Code sessions together Said the sidebar and "needs you" states were exactly what they had been missing; requested clearer answer formatting.
The contributor who opened issue #6 Developed a local Windows port with sidebar behavior and psmux integration Reported a working local implementation and planned small pull requests.

The issue #9 account separates two tasks that can otherwise blur together. Finding an agent that needs attention made the terminal more useful to that author, while reading the resulting answer remained frustrating. For a prospective user, that is a useful way to assess the sidebar: consider how often you lose track of sessions and how often the actual problem is scanning their output.

What happened when we ran it

Our sandbox installed Harp's dependencies in 83 seconds, the build timed out at 900 seconds, and the test command failed with exit 101 after 818 seconds. Our run used commit 31eed50 on 2026-11-16, with a recorded configuration of 3 CPUs, 12 GB RAM and image label lab-rust:1. These are source-harness results rather than a completed desktop session.

Source-harness step Recorded result What that tells you
Install Succeeded in 83 seconds; 1,409 dependencies recorded Dependency setup completed in this sandbox.
Build Timed out in 900 seconds The build reached the run's time limit.
Tests Failed, exit 101, after 818 seconds The command failed; the supplied tail contains compiler arguments without an explanatory diagnostic.

The checkout scan recorded 6,496 files, approximately 1,835,159 source lines and 317.1 MB checked out. It found 21 CI workflow files, no Dockerfile and no tests directory. These are repository-layout observations. A Rust project can contain tests outside a directory named tests, and counting workflow files says how many files the scan found rather than how their checks performed.

The supplied test-log tail names warp_core, warp_editor and warp_errors in compiler arguments, then ends with a summary marker. It contains no named failing assertion or explanation for exit 101. The recorded result is a failed command. Assigning it to a compiler defect, missing package or application bug would require diagnostic evidence beyond that excerpt.

A separate application attempt produced conflicting records. Its application-run summary claims deployment through Docker Compose on 2026-11-16, with every port bound to 127.0.0.1. The safety step failed because no Compose file existed. The teardown record says "teardown: ok". The failed step determines the application-attempt outcome; the runner's "unsafe repo" wording refers to its Compose requirement.

Where people get stuck: status, themes and readable replies

Issue #8 reports missing limit rings and per-agent status after plugin updates and repeated settings changes. The author also reports opacity and sidebar-theme behavior that differs from their expectations of Warp. These are that user's observations. The README's separate session-status and usage-feed instructions provide a more focused starting point than changing appearance settings while troubleshooting both displays together.

Report Evidence-backed next step Resolution in the supplied record
Status and rings absent, #8 Follow the README's separate status and usage-feed paths; record which display fails Open report.
Custom theme mismatch, #10 Compare sidebar and hover-card colors with the selected theme Theme-derived colors and overrides requested.
Rings difficult to read, #10 The reporter uses whole-window zoom and a smaller font Whole-interface workaround; ring-size control requested.
Replies hard to scan, #9 Compare the value of session visibility with answer readability Formatted replies requested.
CLI installation wanted, #14 Use the documented macOS DMG route Homebrew/curl installation requested.

Issue #10 identifies Harp 0.1.3, macOS 27.0 and arm64 as the reporter's environment. With Rosé Pine Dawn, the author reports a fixed grey sidebar, limit rings and hover card. They describe 20 px rings with 1.6/1.8 px strokes and say a low 5-hour share, such as 5%, becomes a dot-sized arc. Their workaround sets whole-window zoom to 110 and reduces font size, enlarging the whole interface.

Start with the two separate Claude Code integrations

The README supplies Claude Code session status through a bundled plugin and Claude plan usage through status-line input. Each Claude Code session started in a Harp shell loads the bundled plugin through CLAUDE_CODE_PLUGIN_DIRS. The documentation says Harp installs nothing into ~/.claude for this plugin, ignores duplicate reports from Warp's plugin and uses jq, which macOS 15 and later include.

For Claude usage rings, Harp checks ~/.claude/settings.json on first start. If no statusLine exists, it adds one that saves input to ~/.harp/claude-statusline.json and prints nothing. An existing status line stays in place. The README's example preserves its command by saving the input first, and says rings appear after Claude Code's next answer for Claude.ai plans rather than API keys.

A useful problem report should identify the app version, where the agent started and whether the missing information belongs to a session row or the footer. For tmux sessions, the README also describes a card message asking for an agent restart when the agent reports nothing. Those details connect a report to the documented input path and give maintainers something more specific than a general statement that the sidebar fails.

Harp keeps Warp's cloud connections

Harp's README says upstream AI and cloud features still talk to Warp's servers. It lists agents, Warp Drive, session sharing and login among those inherited features, while saying Harp adds no servers of its own. Its local limit-ring inputs describe the usage display's data path; they do not change the documented destination of requests made by inherited cloud features.

The documented macOS installation uses a DMG dragged into Applications. Harp has its own bundle ID, URL scheme and ~/.harp settings folder, allowing it to run alongside Warp. The README also describes an updater that checks GitHub releases every 10 minutes, downloads a newer DMG, checks its Developer ID team signature and offers to restart into it.

The README identifies GNU AGPL v3 licensing, with the warpui and warpui_core crates under MIT. It also explains the source-offer obligation when running a modified version for others over a network. For someone considering distribution or modification, those named terms are the relevant starting point. The terminal's licence and the account used by an agent concern separate parts of the workflow.

Recent pushes show activity, not adoption

GitHub metadata records 715 stars and a last push on 2026-11-10 in the snapshot fetched on 2026-11-16. The API's open count is 9 and combines issues and pull requests. These figures describe repository attention and activity at that observation date. The issue accounts supply more specific evidence about how individual people use Harp and what they want changed.

The latest-release record identifies v0.1.8, published on 2026-10-29. Its Harp v0.1.8 release notes say Codex approval or answer waits show "needs you" by reading the terminal title. The notes also say updates are automatic from 0.1.2 or later, while 0.1.0 and 0.1.1 require one manual download. The source harness used commit 31eed50, rather than the release's 54c5286 commit.

Harp vs Warp, Claude desktop and Codex desktop

The author of issue #9 finds Claude and Codex desktop replies easier to read while praising Harp's session sidebar. Their comparison gives a concrete selection criterion: do you need help locating the next session, or help reading its answer? The table preserves that author's preferences and the README's description of Harp's relationship to Warp.

Choice What the supplied reports say Pick it when
Harp Issue #9 praises sidebar visibility and "needs you" states Finding the next agent that needs input is your repeated frustration.
Warp Issue #9 says Warp's own agent displays formatted replies You prefer the upstream terminal and its own agent presentation.
Claude desktop Issue #9's author finds its formatted replies easier to read Answer readability is your priority.
Codex desktop The same author also prefers its formatted reply display Desktop answer presentation matters more than Harp's sidebar.

The comparison comes from the reply-formatting request in issue #9. That author likes Harp enough to say they are happy with it either way, which gives the request useful context. A person can benefit from the terminal's session organization and still prefer another application's presentation. Decide which of those tasks is causing the greater interruption in your own work.

Avoid confusing Harp with namesakes

The HARP use-case page describes an API proxy, while dominikmartn/harp is a Warp terminal fork. The supplied HARP proxy use-case page places that product between applications and remote APIs. Its proxy capabilities belong to a different project and cannot support this terminal review. Keep the repository name attached to searches and comparisons; the supplied namesake autocomplete results do not establish demand for this terminal.

FAQ

Harp documents separate inputs for Claude Code session status and plan-limit rings. That distinction answers the setup question raised by issue #8 and helps organize the other questions in the supplied issues. The answers below distinguish documented behavior, a reporter's workaround and requested changes, so readers can connect each answer to the kind of evidence behind it.

How do I get Harp's limit rings and per-agent status working with Claude Code?

The Harp integration instructions say to start Claude Code in a Harp shell for the bundled status plugin. Claude plan rings use status-line input and appear after the next answer. Check an existing status-line command separately, because Harp leaves it alone. This separates the two paths implicated in issue #8's setup report.

Can Harp display formatted Claude Code and Codex replies with clearer separation between prompts, tool calls, and final answers?

Issue #9 requests formatted replies or clearer separation of prompts, tool calls and final answers. The author says headings, lists, code and answers look alike in terminal output, while Claude and Codex desktop replies are easier to read. Their positive account concerns the sidebar and attention states, which solve a different part of managing agent work.

Can Harp's sidebar, limit rings, and hover card use my custom theme's colors?

Issue #10 reports fixed grey colors with Rosé Pine Dawn on Harp 0.1.3 and requests theme-derived colors or overrides. The author identifies sidebar palette tokens and literal ring colors in the implementation. Their suggested settings describe a requested interface, so the issue is useful for understanding the complaint rather than as a configuration recipe.

Can I enlarge Harp's limit rings without enlarging the whole interface?

The issue #10 workaround sets whole-window zoom to 110 and reduces font size. The author explicitly says this makes the whole UI bigger, including elements beyond the rings. They request a dedicated ring-size setting with a default of 20, which explains both the desired behavior and why their current workaround changes more of the window.

Does Harp support Windows, and what still needs to be implemented?

The issue #6 contributor reports a working local Windows implementation and plans small pull requests. Their report describes a main-branch compile problem involving tab_grouping.rs and libc::kill, plus window glass, turning rings, header dragging, shortcuts and installer identity work. They propose sending window-layer changes first, followed by sidebar and installer changes, while keeping self-update macOS-only for now.

Can Harp's Windows tmux integration use psmux or tmux inside WSL?

The Windows-port contributor says the local implementation uses psmux and excludes tmux inside WSL. That statement belongs to the port described in their issue. It matters for contributors comparing Windows session options, especially because the same report separates the working local implementation from the small pull requests the author intends to submit.

Can I install and update Harp through Homebrew or curl instead of a DMG?

The README's installation route is a macOS DMG, and it describes an in-app updater that checks GitHub releases. Issue #14 requests installation through a CLI tool instead of a DMG. That request concerns a different installation route from the documented updater, so it should not be read as saying Harp has no update mechanism.

Why doesn't window opacity work in Harp?

Issue #8 reports that window opacity does not work and that sidebar theming differs from the author's Warp expectations. The supplied report gives neither a diagnosis nor a corrective setting. Its useful contribution is identifying the observed appearance problem alongside the same author's missing rings and agent states, rather than supporting a particular explanation for the opacity behavior.

What we have not tested yet

Our evidence contains no successful supported macOS desktop run: dependency installation completed, the source build timed out, the test command failed and the separate application attempt stopped at its Compose safety check. We have not tested the DMG, login, updater, restored tmux groups, state accuracy or usage-ring path. The README also labels login through Harp untested. Complete toolchain logs are needed for runtime versions and diagnosis of exit 101.

Desktop memory use, simultaneous-agent capacity, response latency, long-session stability, accessibility, security and network traffic remain outside this run. The 0.1.3 theme report has no supplied retest on v0.1.8. The Windows port and its psmux behavior are contributor reports rather than our execution, and the supplied record establishes no released Windows port. The requested formatting, theme controls, ring-size setting and CLI installer have no recorded delivered resolution; issue #8 has no confirmed repair.

The supplied comparisons express one issue author's preferences rather than a feature-by-feature trial of the alternatives. Rendered layout, assets, automatic related links, recommendation modules and template content also remain outside this text review. For a macOS developer considering Harp, a useful next step is to start the usual agents, make one wait for input, check whether the sidebar identifies it, and then assess how easily its answer can be read.

Alternatives

ProjectWhat it isPick it when
Warp gh↗Harp's upstream terminal; the author of issue #9 says Warp's own agent displays formatted replies.Choose it when you prefer the upstream terminal and its own agent presentation.
Claude desktopThe author of issue #9 finds its formatted replies easier to read than terminal output.Consider it when reading agent answers is the main problem you want to solve.
Codex desktopThe same issue author also prefers its formatted reply presentation.Consider it when desktop answer presentation matters more to you than Harp's terminal sidebar.

What people are saying

  1. [velocity-scout] dominikmartn/harp

Sources

  1. Harp README
  2. Harp public repository metadata
  3. Harp v0.1.8 release
  4. Sidebar and limit rings: follow the active theme's colors, and a larger ring option
  5. Formatted view for Claude Code/Codex replies
  6. Homebrew/curl install
  7. Windows support
  8. Status, limit rings, opacity and theming report

More dev tools reviews