- Home
- Skills
- Game Development
- Godot agent delivery
Works with the AI Agents you already use
Godot agent delivery
An operator-led workflow for Godot 4 projects with source-traced lessons from real development, false-green-safe headless checks, save/simulation guidance and reusable typed GDScript helpers.
$6.50
Godot agent delivery
Example session with this skill installed
Set up Godot agent delivery in my existing Godot 4 project without rewriting its architecture. First check the Godot MCP connection and headless execution; ask before installing or changing anything. Our active slice is fixing a save/load regression. Prepare the relevant checks without touching my real saves, then identify the first reproducible scenario and the player-visible proof needed before I accept the fix.
- Read your context and instructions
- Compiled the godot agent delivery
Result
I completed the read-only preflight. Nothing has been installed or changed in the project, and no existing saves were opened or written. Setup is waiting for your approval.
Tooling status
Godot MCP server: reachable. I called godot_get_server_info() first, then godot_list_toolsets(). The server reports version 2026.09.23; only core and inspection are enabled.
Editor bridge: disconnected. No project identity is available, so I cannot verify an editor connection to this project. I did not enable toolsets, change configuration, add an addon/runtime probe, or start an editor.
Headless CLI: verified. The configured Godot binary reports 4.7.2.stable.official.ed1daf0bf. The delivered helper ran in a separate disposable probe with isolated user data:
"$GODOT_BIN" --version
python3 skill/scripts/run_suites.py --engine "$GODOT_BIN" \
--project probe --suite test_suite_template --timeout 30 --strict \
--json probe/results.json
The runner exited
0: PASS 1/1 checks, zero script errors and zero other error lines. Its results recorded the expected test_suite_template PASS marker. This verifies CLI/helper execution—not your game's persistence, rendered UI, or MCP bridge.
Active regression
I kept docs/SPEC.md as the active specification; no replacement plan or architecture was introduced.
Read-only inspection found that session_state.gd:27–30 parses and validates loaded_value but never applies it to the authoritative value. It then emits the unchanged value, which main.gd displays. That source finding matches the reported symptom;
I have not applied or runtime-verified a fix. Your report is the ground truth, not something I reran merely to confirm.
First scenario and acceptance proof
The first regression scenario is the existing same-session route:
- Start the normal main scene with fresh, isolated user data: Count: 0.
- Increment three times; Save. Check that the saved snapshot contains value 3.
- Increment twice: Count: 5.
- Load. Require both authoritative state and displayed count to become
3, with successful load status—not merely successful JSON parsing.
5. Increment once more. Require Count: 4, proving the state was restored rather than only the label being patched.
After authorization, the automated check should exercise the real scene/actions and save/load code in a temporary user-data root, enforce a timeout, and require exit 0, its own exact PASS marker, and no script/parse/load errors. It has
not been installed or run yet.
For player-visible proof, review that same
0 → 3 → 5 → 3 → 4 route through the rendered buttons and labels, with the tested source/build identity recorded. A headless PASS or “Loaded” message alone is insufficient. Only you accept the fix.
Your setup choice
Which scope do you approve?
MCP repair proposal first: inspect the existing setup read-only and present the exact addon/configuration/editor changes for approval before applying anything.
Headless-only setup: add the minimal runner and isolated regression check, retain the existing architecture/spec, and leave MCP untouched.
- Not now: leave everything unchanged.
Application changes remain unapproved. This run used an authored disposable example and the actual delivered ZIP; it does not establish a customer installation or marketplace activation.
Connects securely to your tools. The creator never sees your data.
About this skill
Godot agents can report success while a loaded script failed, exercise an API without making it playable, test source resources that disappear from an export, or overwrite real settings during a test run. This skill brings those failure modes into the agent’s workflow before the next change.
It includes the complete operator-led delivery protocol plus a categorized field guide covering typed GDScript, scene ownership and teardown, deterministic simulation and saves, real-input routes, physics and generated geometry, Blender/glTF imports, UI and controller prompts, Linux capture limits, Windows/export packaging, handheld checks and performance measurement.
Historical incidents are distinguished from code-confirmed or locally reproduced behavior; project-specific numbers are not universal recommendations.
Bundled Python helpers check the delivery plan and run Godot suites with timeouts, per-suite isolated user data, expected PASS markers and error-output checks. Reusable GDScript helpers support deterministic tick scenarios and real input injection. Your game supplies its own systems, node references and acceptance criteria; the skill does not install a spaceship simulation or rewrite your architecture.
On first use, the agent checks MCP/editor-bridge availability and isolated headless execution separately. Missing or disconnected tools trigger a setup choice; installation and client/ project configuration changes happen only with explicit user approval. Existing tools and declined setup choices are respected. No MCP implementation is bundled.
The general workflow is already included: you do not need the separate general listing. OpenWolf and Graphify are optional, independently installed companions. No game source, third-party tool implementation, credentials or automatic publishing is included.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Game Development.
Godot 4 only. Requires manual approval for OS-level input/credentials. Headless CLI cannot prove visual or tactile outcomes. Needs specific engine binary paths configured.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 26 days ago
- Software
Python