Compare variants

Compare tailtest variants: Claude Code, Cursor, Codex CLI, and Cline

Same rule layer. Same adversarial test mode. Different host integration. Pick the right tailtest for your editor.

Claude Code
3.14.0 · 491 tests
R1-R15 rule layer yes
9 language baselines yes
7+ framework templates yes
R12 failure classification rule-based
Adversarial mode (V13) yes
Auto-fire on every edit yes (PostToolUse hook)
Plan-before-write enforced rule-based
Persistent project profile .tailtest/
Test count (own dogfood suite) 491
Latest version 3.14.0
IDE coverage Claude Code
Cursor
1.6.0 · 181 tests
R1-R15 rule layer yes
9 language baselines yes
7+ framework templates yes
R12 failure classification rule-based
Adversarial mode (V13) yes
Auto-fire on every edit yes (afterFileEdit hook)
Plan-before-write enforced rule-based
Persistent project profile .tailtest/
Test count (own dogfood suite) 181
Latest version 1.6.0
IDE coverage Cursor
Codex CLI
4.7.0 · 380 tests
R1-R15 rule layer yes
9 language baselines yes
7+ framework templates yes
R12 failure classification rule-based
Adversarial mode (V13) yes
Auto-fire on every edit yes (Stop hook)
Plan-before-write enforced rule-based
Persistent project profile .tailtest/
Test count (own dogfood suite) 380
Latest version 4.7.0
IDE coverage Codex CLI
Cline
1.0.0 · 162 tests
R1-R15 rule layer yes
9 language baselines yes
7+ framework templates yes
R12 failure classification structured (MCP)
Adversarial mode (V13) yes
Auto-fire on every edit instruction + auto-approve
Plan-before-write enforced UX-native (Plan mode)
Persistent project profile .tailtest/ + Memory Bank
Test count (own dogfood suite) 162
Latest version 1.0.0
IDE coverage 9+ editors via Cline

Which tailtest should you install?

Pick by the editor you actually use. The rule layer and adversarial test mode are identical across all four; the differences are in how each plugin hooks into your editor.

Use the Claude Code variant if...

3.14.0

Claude Code users; most precise hook-fired auto experience and the largest dogfood test surface.

Trigger: PostToolUse hook
IDE coverage: Claude Code
Marketplace install
$ claude plugin marketplace add avansaber/tailtest
$ claude plugin install tailtest@avansaber-tailtest
# Restart Claude Code. Then just build.

Use the Cursor variant if...

1.6.0

Cursor users wanting the deepest native integration. Lighter weight than running Cline inside Cursor.

Trigger: afterFileEdit + stop hooks
IDE coverage: Cursor
Cursor Marketplace

Search for "Tailtest" in the Cursor plugin marketplace and click Install.

Restart Cursor.

Use the Codex CLI variant if...

4.7.0

Codex CLI users wanting hook-driven Stop event integration.

Trigger: Stop hook + session_start
IDE coverage: Codex CLI
Three-step install
Step 1: git clone https://github.com/avansaber/tailtest-codex ~/.codex/plugins/tailtest
Step 2: enable hooks in ~/.codex/config.toml ([features] codex_hooks = true)
Step 3: cd <your-project> && bash ~/.codex/plugins/tailtest/scripts/init.sh
# Start codex. Then just build.

Use the Cline variant if...

1.0.0

Anyone whose editor is not Claude Code, Cursor, or Codex CLI directly. One plugin reaches 8+ editors. Pairs with Cline's Plan / Act mode for explicit human review gates.

Trigger: .clinerules instruction + auto-approve OR /tailtest-test slash
IDE coverage: 9+ editors
Cline MCP Marketplace

Open Cline's Extensions panel inside your IDE. Search for tailtest. Click install. Cline clones the repo, installs deps, registers the server in cline_mcp_settings.json automatically.

Reload your conversation. Then run: 'set up tailtest in this project'.

Honest delta: Cline auto-fire is instruction-driven

Tailtest-cline is the only variant where auto-fire-on-edit depends on agent compliance with a .clinerules/ instruction plus Cline's auto-approve setting, rather than a host runtime hook. The structured MCP tool layer compensates by making R12 classification, scenario planning, and framework template lookup deterministic regardless of rule recall. Net: high reliability on what matters (the analysis), one explicit configuration step on the trigger.

If you use Claude Code, Cursor, or Codex CLI directly, prefer the matching variant for the tightest hook integration. If your editor is anything else (VS Code, JetBrains, Zed, Neovim, etc.), tailtest-cline is the way -- one plugin, eight-plus editors.

Cursor vs Claude Code: testing edition

The honest answer: tailtest works in both. The Cursor and Claude Code variants of tailtest hook into each editor's native event model -- afterFileEdit for Cursor, PostToolUse for Claude Code -- and use the same R1-R15 rule layer so generated tests look the same regardless of which agent you're running.

You only need to pick a variant if you use one editor exclusively. If you switch between Claude Code and Cursor, install both (they don't conflict). If you use Cursor through Cline, install just the Cline variant.

MCP test server for Cline

Tailtest-cline is an MCP server. It exposes 5 structured tools (tailtest_setup, tailtest_scenario_plan, tailtest_classify_failures, tailtest_pick_template, tailtest_ping) plus a 451-line .clinerules/ rule pack and a Memory Bank integration. The structured I/O is the difference: instead of asking the agent to remember "always include 2 adversarial scenarios at standard depth," the agent calls a tool that always returns the right number.

One Cline install reaches VS Code, Cursor, JetBrains IDEs, Antigravity, Zed, Neovim, VSCodium, Windsurf, and the Cline CLI -- the same MCP server runs in all of them.

Get started

Same MIT license. Same rule layer. Pick by editor.

All four variants are MIT-licensed, no telemetry, no analytics. Switch any time -- they all read the same .tailtest/ config.