New: 98,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

Harness Test Drive

by RyanAlberts in RyanAlberts/best-of-Agent-Harnesses

Test-drives coding agents (Claude Code, Codex, Gemini CLI) on tasks mined from the user's own git history: each agent gets a past commit message in a fresh copy of the repo, and the repo's own tests…

MITAuto-check passedDevelopment

Install Harness Test Drive

skills CLI
$ npx skills add RyanAlberts/best-of-Agent-Harnesses --skill harness-test-drive -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install RyanAlberts/best-of-Agent-Harnesses harness-test-drive --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/RyanAlberts/best-of-Agent-Harnesses.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/harness-test-drive .claude/skills/harness-test-drive && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
harness-test-drive
GitHub stars
1.1k
Token cost
~2.9k tokens
SKILL.md length
1,568 words
Files
11 (incl. scripts, references)
Repo updated
First seen
Licence
MIT

At a glance

Test-drives coding agents (Claude Code, Codex, Gemini CLI) on tasks mined from the user's own git history: each agent gets a past commit message in a fresh copy of the repo, and the repo's own tests…

  • Works in 6 steps: Find the test command and the candidate… → Check the tasks. For each candidate, in… → Choose the harnesses and show the estimate → …
  • The user asks which coding agent
  • SKILL.md covers When to use, When not to use, Steps and Read the results, plus 2 more sections
  • Runs Python scripts from its folder; calls python3, npm and pnpm

What it does

Harness Test Drive is an agent skill from RyanAlberts/best-of-Agent-Harnesses. Test-drives coding agents (Claude Code, Codex, Gemini CLI) on tasks mined from the user's own git history: each agent gets a past commit message in a fresh copy of the repo, and the repo's own tests score how many tasks it completes, dollars per task, minutes, and change size. Use when the user asks which coding agent or harness works best on their codebase; wants to compare or benchmark agents on their own repository instead of trusting leaderboards such as SWE-bench; wants to trial one agent against another…

Its SKILL.md is about 2.9k tokens, which your agent loads only when the skill is triggered. The skill folder holds 12 other files, including scripts and reference files (for example `README.md`, `references/harness-commands.md` and `references/method.md`). Compatibility notes: Python 3.9+ and git on macOS or Linux, plus the harness CLIs to compare (claude, codex, gemini). The scripts make no network calls themselves; each harness…

It sits in Development, covering Git workflow and Commit messages. The repository describes itself as: 🏆 Ranked list of 167 AI agent harnesses, plus templates, playbooks, MCP, and learning resources. Rescored weekly. The licence is MIT.

When your agent uses it

  • The user asks which coding agent
  • Harness works best on their codebase
  • Wants to compare
  • Benchmark agents on their own repository instead of trusting leaderboards such as SWE-bench

Example prompts

  • “/harness-test-drive”

Requirements

  • Python 3
  • Compatibility (from SKILL.md): Python 3.9+ and git on macOS or Linux, plus the harness CLIs to compare (claude, codex, gemini). The scripts make no network calls themselves; each harness CLI sends the prompt and the code it reads to its model provider over the network, and those runs cost money. The repository's test command must run on this machine.

Workflow steps

6 steps, taken from the first numbered list in SKILL.md.

  1. Find the test command and the candidate tasks. This reads git history and runs nothing
  2. Check the tasks. For each candidate, in a fresh copy of the repository, the test command must
  3. Choose the harnesses and show the estimate
  4. Get an explicit dollar cap. This is the one skill in the set that spends money. Ask: "What is
  5. Run
  6. Report in the shape below. To print the scoreboard again at any time

What it can do on your machine

Read from SKILL.md and the folder at commit 4fa20bc. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Ships 6 files in scripts/ (Python), which the agent can run.

    Shell commands in SKILL.md call:

    • python3
    • npm
    • pnpm
    • pip
    • python

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use npm, pnpm and pip, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

  • Compatibility

    Python 3.9+ and git on macOS or Linux, plus the harness CLIs to compare (claude, codex, gemini). The scripts make no network calls themselves; each harness CLI sends the prompt and the code it reads to its model provider over the network, and those runs cost money. The repository's test command must run on this machine.

    From compatibility in the SKILL.md frontmatter.

Context cost

Harness Test Drive loads about 2.9k tokens when it runs, and up to ~8.7k if it reads all its reference files. Until then it costs about 180 tokens; SKILL.md has 1,568 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~180
When it runs · the whole SKILL.md, loaded when a task matches
~2.9k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~8.7k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.

SKILL.md

The full file from RyanAlberts/best-of-Agent-Harnesses at commit 4fa20bc, republished under its MIT licence (© RyanAlberts). 1,568 words, ~2,880 tokens.

Download SKILL.mdSave it as .claude/skills/harness-test-drive/SKILL.md (or your agent's skills folder). This skill also uses 10 other files; get the full folder from GitHub.
name
harness-test-drive
description
Test-drives coding agents (Claude Code, Codex, Gemini CLI) on tasks mined from the user's own git history: each agent gets a past commit message in a fresh copy of the repo, and the repo's own tests score how many tasks it completes, dollars per task, minutes, and change size. Use when the user asks which coding agent or harness works best on their codebase; wants to compare or benchmark agents on their own repository instead of trusting leaderboards such as SWE-bench; wants to trial one agent against another before switching; or asks what each agent costs per fixed bug. Spends money: the harness CLIs call their model providers, so a dollar cap is required. The scripts make no network calls.
compatibility
Python 3.9+ and git on macOS or Linux, plus the harness CLIs to compare (claude, codex, gemini). The scripts make no network calls themselves; each harness CLI sends the prompt and the code it reads to its model provider over the network, and those runs cost money. The repository's test command must run on this machine.
license
MIT
metadata.author
Ryan Alberts
metadata.version
1.0.0
metadata.source
https://github.com/RyanAlberts/best-of-Agent-Harnesses

Harness test drive

Public leaderboards measure someone else's code, and harness rankings barely carry over from one repository to the next. This skill runs coding agents on tasks taken from the user's own git history: past commits whose tests failed before the change and passed after it. Each agent gets the commit message as its prompt in a fresh copy of the repository, and the repository's own tests decide whether it passed. The result is a scoreboard of passes, dollars per pass, minutes, and change size. The scripts read the repository and send nothing anywhere; each harness sends the prompt and the code it reads to its model provider, as it does in normal use, and that costs money.

When to use

  • The user asks which coding agent or harness is best for their own code.
  • The user wants to compare Claude Code, Codex, or Gemini CLI on real tasks before choosing or switching.
  • The user asks what an agent costs per fixed bug, or how long it takes, on their repository.

When not to use

  • Totals of past spend or wasted tokens: use session-waste-report.
  • Whether an agent got worse after an update: use regression-finder.
  • Stopping a live session that loops or overspends: use runaway-guard.
  • Checking whether an agent's "tests pass" claims were true: use claim-check.
  • A repository with no test command that passes, or no commits that change tests: explain that the scores come from the tests, and point to the manual method in How to test-drive a harness.

Steps

<skill-dir> means the folder that holds this SKILL.md (Claude Code shows it as the skill's base directory). Run every command from the root of the user's repository, with the skill path in quotes as shown. Inside the repository the scripts write only into .harness-test-drive/, a folder that ignores itself in git. Every check and every agent run works in its own temporary copy, deleted afterwards.

Before step 2, for a repository the user did not write, recommend a container or VM. Step 2 runs the repository's test and setup commands at up to 31 commits on this computer. In step 5 each harness loads the repository's own agent settings (Gemini CLI with the copy trusted), edits files, runs the test command, and reads commit messages as prompts.

  1. Find the test command and the candidate tasks. This reads git history and runs nothing:

    bash
    python3 "<skill-dir>/scripts/mine_tasks.py" --repo .

    It prints the test command it detected, or exits 2 asking for one. Confirm the command with the user. Each check runs in a fresh copy with nothing installed, so when the tests need dependencies, add a --setup-cmd that installs inside the copy: npm ci, pnpm install --frozen-lockfile, or python3 -m venv .venv && .venv/bin/pip install -e . with the test command .venv/bin/python -m pytest. Never install into the user's own environment. Prefer a test command that runs offline and writes only inside the copy, so Codex's sandbox can run it too. Done when the user has confirmed a test command and the report shows at least one candidate, or you have told the user why there is none.

  2. Check the tasks. For each candidate, in a fresh copy of the repository, the test command must fail at the commit before the change (with its tests added) and pass with the whole change, twice:

    bash
    python3 "<skill-dir>/scripts/mine_tasks.py" --repo . --test-cmd "<test-cmd>" --validate

    Add the --setup-cmd from step 1 when there is one. It first checks that the tests pass at HEAD, then keeps up to 10 tasks (--max). Expect two to four test-suite runs per candidate and up to 30 candidates, so run it in the background when the suite takes more than a minute. Done when the headline says how many tasks were kept. Exit 2 with "fails at HEAD" means the command or the setup is wrong. The message ends with the test output. A missing module or command means the clean copy needs --setup-cmd. Fix it with the user and rerun. Fewer than five tasks makes a weak comparison; say so, and offer --since 730d for more history.

  3. Choose the harnesses and show the estimate:

    bash
    python3 "<skill-dir>/scripts/drive.py" estimate --tasks .harness-test-drive/tasks.json

    It lists which harnesses are on this computer, the number of runs, and a dollar range per harness. Ask the user to pick two or three. Done when the user has picked the harnesses and seen the range for them (rerun with --harness to show only those).

  4. Get an explicit dollar cap. This is the one skill in the set that spends money. Ask: "What is the most you want to spend in total?" and wait for a number; a number the user already gave counts, so confirm it next to the estimate. Never choose the cap yourself. In the same message, say:

    • The cap limits counted spend. Claude Code stops itself at the budget left. A Codex run has no limit except --timeout, and a run cut off there is counted as an estimate, so the bill can pass the cap by more than one run.
    • On a subscription plan, runs count against the plan's limits; the cap still counts API prices.
    • Gemini CLI's spend cannot be counted (it reports no cost, and the price table has no Gemini prices), so it runs only with --allow-unpriced gemini-cli, after the user agrees to that.
    • Codex runs are priced with the model named in Codex's config, which the estimate shows. When the config names none, pin one with --model codex=<id>; Codex never runs unpriced.
    • For someone else's repository, run inside a container or VM (see above).

    Done when the user has given a dollar amount.

  5. Run:

    bash
    python3 "<skill-dir>/scripts/drive.py" run --tasks .harness-test-drive/tasks.json --harness claude-code,codex --max-usd <cap>

    Add --allow-unpriced gemini-cli only when the user agreed to it in step 4, and --model only for a model the user pinned. Each run can take up to --timeout seconds (900 by default), so start it in the background and relay the progress lines. A harness that cannot run (an expired login, an old version) prints a fix, such as "sign in: run claude once" or "upgrade Codex", and its other runs are skipped at no cost. Every run is appended to .harness-test-drive/results.jsonl; the same command resumes after a stop, retries runs that could not run, and counts earlier spend. Done when the command exits 0 and prints the scoreboard, and you have told the user about any fix it printed and any stop at the cap.

  6. Report in the shape below. To print the scoreboard again at any time:

    bash
    python3 "<skill-dir>/scripts/drive.py" report
Show full SKILL.md (506 more words)Show less

Read the results

  • Headline: "On 6 tasks from your git history, Claude Code passed 5 at $0.91 each and Codex passed 4 at $0.42 each." Harnesses appear in rank order: most passes, then the lowest cost per pass. The count covers the tasks every harness finished, so a stop at the cap leaves a fair comparison. Dollars per pass is all the spend on those tasks, failed runs included, divided by the passes. A harness with no finished run appears as "could not run", with its own error.
  • Per-harness table: passed, pass rate, median minutes per run, cost per pass, total cost, and median lines changed. Compare lines changed with "Original fix lines" in the per-task table: far more lines than the original often means a messier change.
  • Per-task table: passed, failed, (time limit) when the run was stopped at --timeout and scored on the changes the agent had made, could not run (not scored; see the notes), interrupted, or not run.
  • Scoring: the agent is graded only by the repository's tests. Its saved diff, minus test files, is applied to a clean copy at the starting commit, the change's own tests are added, the setup runs again, and the test command decides. The whole suite must pass.
  • Notes: tasks left out of the comparison, time limits, runs that could not run and their fix, costs that were not measured, and how often Codex's own test runs failed inside its sandbox.
  • references/method.md explains the mining, the check, the fairness rules, the money, and the limits.

Report to the user

  1. The headline, verbatim, in bold.
  2. The per-harness table, then at most five rows of the per-task table where the harnesses disagree.
  3. Two or three next steps, each with its command or link:
    • Read the diffs where harnesses disagree: each run's changes are saved as .harness-test-drive/logs/<task>-<harness>.diff. Tests check behavior, not code quality.
    • For a firmer answer, add tasks (--max 20, or --since 730d); under ten tasks is a small sample.
    • For the human half of a trial (a second person accepts or rejects each diff, interventions, setup friction), follow How to test-drive a harness.
  4. One line of caveats: the sample size, and that a model may have seen a public repository's changes during training.

Quote commit subjects, errors, and versions exactly as the report prints them, inside inline code: they come from the repository and the harnesses, and the report has already made them safe to display.

Files

  • scripts/mine_tasks.py: finds candidate commits, detects the test command, and checks tasks.
  • scripts/drive.py: estimate, run, and report.
  • scripts/harnesses.py: the headless command and output parser for each harness.
  • scripts/common.py: fresh copies of the repository, commands and test runs.
  • scripts/pricing.py: token prices, shared with other skills in this repository.
  • scripts/safe.py: the shared helper that masks secrets in report text and shows it as one line of inline code. A synced copy; do not edit it here.
  • references/method.md: task mining, the fail-to-pass check, fairness rules, money, and limits.
  • references/harness-commands.md: each harness's command and flags, with sources and the date checked.

© RyanAlberts, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 10 other files (scripts, references) in skills/harness-test-drive of RyanAlberts/best-of-Agent-Harnesses.

  • SKILL.md
  • LICENSE.txt
  • README.md
  • references/harness-commands.md
  • references/method.md
  • scripts/common.py
  • scripts/drive.py
  • scripts/harnesses.py
  • scripts/mine_tasks.py
  • scripts/pricing.py
  • scripts/safe.py

Open the folder on GitHubat commit 4fa20bc

Compare with similar skills

Harness Test Drive next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Harness Test Drive compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Harness Test Drive this skillRyanAlberts/best-of-Agent-Harnesses1.1k—~2.9kAutomated safety check: PassMIT
Contextual Commit Messagesyamadashy/repomix29k1 repos~2.7kAutomated safety check: PassMIT
ToolJet Multi-Repo CommitToolJet/ToolJet41k—~1.3kAutomated safety check: PassAGPL-3.0
Git Workflow and Versioningaddyosmani/agent-skills105k2 repos~3.5kAutomated safety check: NotesMIT
React Router Pull Request Creatorremix-run/react-router57k—~2.5kAutomated safety check: PassMIT
Saleor Commit Workflowsaleor/saleor23k—~575Automated safety check: PassBSD-3-Clause

Similar skills

  • Contextual Commit Messages

    yamadashy/repomix

    Writes Conventional Commits whose bodies carry action lines recording the intent, decisions and constraints behind a change, not only what changed.

    29k GitHub starsUsed in 1 repo~2.7k tokens
    DevelopmentAuto-check passed

  • ToolJet Multi-Repo Commit

    ToolJet/ToolJet

    Commits changes across ToolJet's root repo and its server/ee and frontend/ee submodules, writing messages from the diffs and updating submodule pointers in order.

    41k GitHub stars~1.3k tokensUpdated yesterday
    DevelopmentAuto-check passed

  • Git Workflow and Versioning

    addyosmani/agent-skills

    Sets git habits for every change: short-lived branches, atomic commits with descriptive messages, clean pull requests, plus versioning, tagging and changelogs for releases.

    105k GitHub starsUsed in 2 repos~3.5k tokens
    DevelopmentAuto-check: notes

  • React Router Pull Request Creator

    remix-run/react-router

    Packages finished React Router work into a draft pull request: branch, commit, push, a written PR body and the right GitHub labels.

    57k GitHub stars~2.5k tokensUpdated yesterday
    DevelopmentAuto-check passed

  • Saleor Commit Workflow

    saleor/saleor

    Commits changes in the Saleor codebase and works through pre-commit hook failures from ruff, mypy, the GraphQL schema check and the migrations check.

    23k GitHub stars~575 tokensUpdated 2 days ago
    DevelopmentAuto-check passed

  • Emoji Commit Conventions

    baptisteArno/typebot.io

    Sets the repository's convention for commit messages and pull request titles: one emoji prefix for the main intent, a concise title and clean follow-up commits.

    11k GitHub stars~424 tokensUpdated 4 days ago
    DevelopmentAuto-check passed

More from RyanAlberts/best-of-Agent-Harnesses

All 9 skills in this repo

  • Agents Md Checker

    RyanAlberts/best-of-Agent-Harnesses

    Checks which instruction files (AGENTS.md, CLAUDE.md, GEMINI.md, Cursor rules, Copilot instructions) each coding agent loads from a repo, what gets cut or skipped, and whether the commands those…

    1.1k GitHub stars~3.1k tokensUpdated 2 days ago
    Auto-check passed

  • Claim Check

    RyanAlberts/best-of-Agent-Harnesses

    Claim checker that audits a coding agent's statements that tests pass or a build is clean against its own session transcripts: whether a matching run happened before the claim, whether it passed…

    1.1k GitHub stars~2.4k tokensUpdated 2 days ago
    Auto-check passed

  • Guardrail Tester

    RyanAlberts/best-of-Agent-Harnesses

    Guardrail tester that checks whether the permission rules and PreToolUse hooks already set up in Claude Code, Codex, Gemini CLI, OpenCode, or Cursor stop a battery of dangerous commands, including…

    1.1k GitHub stars~2.9k tokensUpdated 2 days ago
    Auto-check passed

  • Regression Finder

    RyanAlberts/best-of-Agent-Harnesses

    Regression check for coding agents: shows how the agent behaved before and after each harness update, model switch, or week in the user's own Claude Code or Codex history, and finds the point where…

    1.1k GitHub stars~3k tokensUpdated 2 days ago
    Auto-check passed

  • Rules To Guards

    RyanAlberts/best-of-Agent-Harnesses

    Rule enforcer that finds which written rules in AGENTS.md, CLAUDE.md, and GEMINI.md a coding agent keeps breaking, counts every violation in recent Claude Code, Codex, Gemini CLI, and OpenCode…

    1.1k GitHub stars~2.7k tokensUpdated 2 days ago
    Auto-check: notes

  • Runaway Guard

    RyanAlberts/best-of-Agent-Harnesses

    Runaway guard: a hook that stops a live Claude Code or Codex session when the agent loops on the same tool call, keeps failing, or exceeds a dollar cap.

    1.1k GitHub stars~2.6k tokensUpdated 2 days ago
    Auto-check passed

Categories

Questions about Harness Test Drive

What does Harness Test Drive do?

Test-drives coding agents (Claude Code, Codex, Gemini CLI) on tasks mined from the user's own git history: each agent gets a past commit message in a fresh copy of the repo, and the repo's own tests…. Harness Test Drive is an agent skill from RyanAlberts/best-of-Agent-Harnesses. Test-drives coding agents (Claude Code, Codex, Gemini CLI) on tasks mined from the user's own git history: each agent gets a past commit message in a fresh copy of the repo, and the repo's own tests score how many tasks it completes, dollars per task, minutes, and change size.

When should I use Harness Test Drive?

Harness Test Drive fits situations like: the user asks which coding agent; harness works best on their codebase; wants to compare; benchmark agents on their own repository instead of trusting leaderboards such as SWE-bench.

How do I install Harness Test Drive in Claude Code?

Run `npx skills add RyanAlberts/best-of-Agent-Harnesses --skill harness-test-drive -a claude-code`. Or copy the skill folder (skills/harness-test-drive in RyanAlberts/best-of-Agent-Harnesses) into .claude/skills/harness-test-drive in your project. Claude Code loads it when a task matches its description.

How do I install Harness Test Drive in Codex?

Run `npx skills add RyanAlberts/best-of-Agent-Harnesses --skill harness-test-drive -a codex`. Or copy the skill folder (skills/harness-test-drive in RyanAlberts/best-of-Agent-Harnesses) into .agents/skills/harness-test-drive in your project. Codex loads it when a task matches its description.

Can I use Harness Test Drive in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add RyanAlberts/best-of-Agent-Harnesses --skill harness-test-drive -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/harness-test-drive, .gemini/skills/harness-test-drive, .github/skills/harness-test-drive and .opencode/skills/harness-test-drive in your project.

What does Harness Test Drive need to run?

Going by SKILL.md and its folder, Harness Test Drive needs Python for the scripts in its folder and the command-line tools its instructions call (python3, npm, pnpm, pip and python). Our summary lists: Python 3. Compatibility (from SKILL.md): Python 3.9+ and git on macOS or Linux, plus the harness CLIs to compare (claude, codex, gemini). The scripts make no network calls themselves; each harness CLI sends the prompt and the code it reads to its model provider over the network, and those runs cost money. The repository's test command must run on this machine..

Does Harness Test Drive access the network?

SKILL.md contains no URLs. Its commands use npm and pip, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is Harness Test Drive safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.

What licence does Harness Test Drive use?

Harness Test Drive is published under the MIT licence (declared in SKILL.md). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Harness Test Drive use?

About 2.9k tokens (SKILL.md is roughly 12k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 5.9k tokens, read only when the agent opens those files.

What are the alternatives to Harness Test Drive?

Skills that share tags, products or a category with Harness Test Drive: Contextual Commit Messages (yamadashy/repomix, 29k stars), ToolJet Multi-Repo Commit (ToolJet/ToolJet, 41k stars), Git Workflow and Versioning (addyosmani/agent-skills, 105k stars) and React Router Pull Request Creator (remix-run/react-router, 57k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Harness Test Drive?

RyanAlberts (a GitHub user) maintains it in RyanAlberts/best-of-Agent-Harnesses, which has 1,133 GitHub stars. The repository was last updated on October 9, 2026.

Source: RyanAlberts/best-of-Agent-Harnesses on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.