godot-gdscript-headless-testing Checked
Installation
Copy the prompt below and send it to your AI assistant (e.g. Claude Code) — it will follow the instructions and install automatically.
Install the "godot-gdscript-headless-testing" skill by following the instructions at https://skill123.me/install/godot-gdscript-headless-testing.
Run in your terminal — downloads and installs to ~/.claude/skills/.
curl -fsSL https://skill123.me/install/godot-gdscript-headless-testing.sh | bash
Download the zip and extract it into your skills directory (e.g. ~/.claude/skills/), then restart your session.
⬇ Download v1.0.0 · 4 KBNot live-tested yet — this skill is rated on static analysis only.
About this skill
Run GDScript test suites from the command line with godot --headless, using a SceneTree/MainLoop runner script that exits non-zero on failure so CI can gate merges. Use when a Godot project needs unit tests without opening the editor, when wiring a CI job (GitHub Actions or similar) that must fail the build on a failing .gd test, or when a godot --headless invocation hangs, opens a window, or exits 0 despite failed assertions.
Godot GDScript headless testing (4.x)
Run GDScript tests from the command line, without the editor GUI, and get a real
process exit code CI can act on. Targets **Godot 4.7** headless CLI.
Documentation
Run GDScript test suites from the command line with godot --headless, using a SceneTree/MainLoop runner script that exits non-zero on failure so CI can gate merges. Use when a Godot project needs unit tests without opening the editor, when wiring a CI job (GitHub Actions or similar) that must fail the build on a failing .gd test, or when a godot --headless invocation hangs, opens a window, or exits 0 despite failed assertions.
Godot GDScript headless testing (4.x)
Run GDScript tests from the command line, without the editor GUI, and get a real
process exit code CI can act on. Targets **Godot 4.7** headless CLI.
When to use
- Use when a Godot project has no testing addon installed and needs a fast way to
verify GDScript logic (pure functions, resource loading, autoload state) from a
terminal or CI pipeline.
- Use when wiring a CI job that must fail the build when a
.gdtest fails. - Use when debugging why a
godot --headlessinvocation hangs, opens a window, or
exits 0 despite failing assertions.
**When *not* to use:** GDScript syntax or language features themselves →
godot-gdscript; export/build pipeline and platform templates → godot-export
(its own --headless use case, producing a binary, not running tests).
Workflow
1. **Confirm the binary resolves headless.** Godot 4.x ships --headless built in
(no export template needed); run godot --headless --version and confirm it
prints a version string, not a GUI window.
2. **On a fresh checkout, import before running tests.** .godot/ is normally not
committed, so a clean checkout has no import cache: class_name types fail to
resolve (Identifier "X" not declared in the current scope) and imported
assets fail to load (No loader found for resource: res://...). Run
godot --headless --path <project_dir> --import once first, in CI and locally.
3. **Write the runner as a SceneTree script, not a Node scene.** A SceneTree
script's _initialize() runs once before any frame — enough for pure-logic
tests and no .tscn required to launch.
4. **Track pass/fail counts yourself and call quit(<code>) explicitly. Do not use
bare assert() to fail a test.** Godot does not turn the process exit code
non-zero on push_error() by itself — the runner must count failures and call
quit(1). Worse, a failed assert() inside _initialize() (official/debug
build) prints SCRIPT ERROR: Assertion failed and **stops execution before
quit() runs**, so the process never exits and CI hangs until its own timeout.
Use an assert_eq()-style helper that records the failure and keeps going.
5. **Invoke with godot --headless --path <project_dir> --script res://<runner>.gd**
and read the **process exit code**, not just stdout, from the shell or CI step.
--script accepts both a res://-relative path and an absolute filesystem
path (e.g. a runner outside the project folder); either works.
6. **Redirect stdout and stderr t
View full documentation
o files when scripting the invocation from a
wrapper shell** (PowerShell, some CI runners). push_error() output goes to
stderr and can be dropped or reordered when only stdout is captured live.
7. **Add a step timeout in CI.** Even with the assert() pitfall avoided, an
await that never resolves (Pattern #2) hangs the runner forever; a
timeout-minutes on the CI step is a backstop CI-side, not a substitute for
backing every await with a timeout node.
Patterns
### 1. Minimal SceneTree test runner with a real exit code
# res://test_runner.gd — run with:
# godot --headless --path . --script res://test_runner.gd
extends SceneTree
var passed := 0
var failed := 0
func _initialize() -> void:
test_add()
print("Results: %d passed, %d failed" % [passed, failed])
quit(1 if failed > 0 else 0) # non-zero exit fails the CI step
func assert_eq(actual, expected, label: String) -> void:
if actual == expected:
passed += 1
else:
failed += 1
push_error("FAIL %s: expected %s, got %s" % [label, expected, actual])
func test_add() -> void:
assert_eq(2 + 2, 4, "test_add")
Verified against Godot 4.7.2: `godot --headless --path . --script
res://test_runner.gd prints Results: N passed, M failed` to stdout, routes
push_error lines to stderr, and returns process exit code 0 when
failed == 0, 1 otherwise.
### 2. Testing something that needs a frame, a timer, or a signal
extends SceneTree
var passed := 0
var failed := 0
func _initialize() -> void:
await run_tests()
print("Results: %d passed, %d failed" % [passed, failed])
quit(1 if failed > 0 else 0) # track and report failures here too
func assert_eq(actual, expected, label: String) -> void:
if actual == expected:
passed += 1
else:
failed += 1
push_error("FAIL %s: expected %s, got %s" % [label, expected, actual])
func run_tests() -> void:
# `root` is not inside the tree yet during _initialize(): a Timer started now
# errors ("not inside the tree") and its `timeout` never fires. Wait one frame.
await process_frame
var timer_node := Timer.new()
timer_node.one_shot = true # default Timer restarts after timeout
root.add_child(timer_node)
timer_node.start(0.1)
await timer_node.timeout
# assertions here can rely on the node having been in the tree for a frame
assert_eq(timer_node.is_stopped(), true, "timer_fires_once")
timer_node.queue_free()
_initialize() may await, which is what makes this pattern work for anything
that needs a node to actually enter the tree, a timer to fire, or a signal to
emit — none of which happen before the engine has processed at least one frame.
Use the same passed/failed counter and assert_eq() helper as Pattern #1;
a version of this pattern that always calls quit(0) can never fail a build.
### 3. CI step (GitHub Actions) that gates on the exit code
- name: Import project (populates .godot/ on a fresh checkout)
run: godot --headless --path . --import
- name: Run GDScript tests
timeout-minutes: 5
run: godot --headless --path . --script res://test_runner.gd
The import step is required on a clean checkout — without it, class_name types
and imported resources fail to resolve. No extra flag is needed for the test
step itself: the runner already fails the job on a non-zero exit code from
run:; the discipline lives in the runner script's quit() call, not in the CI
configuration. timeout-minutes is a backstop against a hung await (see
Pitfalls), not a substitute for backing every await with a timeout node.
Pitfalls
- **Script "does nothing" or opens the editor window** → missing
--headless, or
the script path is wrong. --script accepts a res://-relative path resolved
against --path <project_dir>, and also an absolute filesystem path — both work.
- **
Identifier "X" not declared in the current scope, or a resource fails to
load, only on a fresh checkout** → .godot/ (the import cache) is normally not
committed, so class_name types and imported assets aren't resolved yet. Run
godot --headless --path <project_dir> --import once before the test step.
- **A failed
assert()hangs instead of failing the test** → in an official/debug
build, a failed assert() inside _initialize() prints `SCRIPT ERROR:
Assertion failed and stops that function before it reaches quit()` — the
process never exits and CI waits until its own timeout. Use an assert_eq()
counter (Pattern #1) instead of bare assert() in test runners.
- **Exit code stays 0 despite failed assertions** → the runner never called
quit(1), or (Pattern #2) it always calls quit(0) regardless of failures.
Track failures yourself and call quit() explicitly with a code that reflects
them; do not rely on assert() or push_error() alone to change the exit code.
- **
_initialize()runs before nodes, timers, or signals exist** → logic that
needs a frame to have processed must await a signal or a timer before
asserting; see Pattern #2. root itself is not inside the tree yet, so a
Timer added and started there errors and its timeout never fires (the
runner hangs) — await process_frame first.
- **Output looks empty or out of order from a wrapper shell** → some shells
(PowerShell in particular) can reorder or drop a native
Score breakdown
Assessment methodology
Version history
Imported from gamedev-skills/awesome-gamedev-agent-skills@main
