Project instructions
Declare engine and plugin versions, code style, module ownership, build commands, source-control policy, forbidden generated paths, and product claims.
Kimi Code × Unreal Engine workflow
Build a safer Kimi Code task pack for Unreal Engine work
Use Kimi Code for bounded Unreal C++ and Blueprint planning with repository instructions, terminal permissions, tests, rollback, and a tracked playable prototype.
Direct answer
Moonshot AI says Kimi K3 is available in Kimi Code and can navigate large repositories and orchestrate terminal tools. For Unreal work, the right unit is a bounded task pack: starting commit, relevant files, engine version, allowed commands, acceptance checks, prohibited paths, and a rollback point. Terminal access increases both usefulness and risk.
A Kimi Code prompt should be an executable contract
A useful coding-agent brief tells the agent what outcome matters, what evidence is authoritative, which tools it may use, what it must not touch, how to validate, and when to stop. It should not demand a large implementation before the project boundary is understood.
Least-privilege tools
Start read-only. Grant file writes, builds, tests, network access, package installation, editor automation, and destructive actions separately and only when needed.
Native validation
Require the correct Unreal build target, automation or reproduction, logs, performance evidence, packaging checks, and a human review of the diff.
Recovery contract
Checkpoint before risky stages, preserve the starting commit, cap retries, record failed approaches, and return an unresolved-question list instead of guessing.
A four-stage Unreal evaluation workflow
Write the bounded task pack
State the player-visible outcome, reproduction, relevant subsystem, engine version, target platform, acceptance checks, non-goals, and forbidden changes.
Inspect before editing
Map files, symbols, Blueprints, assets, logs, tests, dependencies, and project instructions; ask for named missing evidence.
Change one reversible stage
Produce a small diff, explain assumptions, run the authorized checks, and stop if the result contradicts the frozen task boundary.
Hand off with evidence
Return changed files, commands, outputs, captures, unresolved risks, rollback point, and the exact human verification still required.
Four bounded task prompts
Use these as task contracts, not as capability claims. Each one asks for observable evidence and a stopping condition.
Unreal C++ bug task
Investigate one reproducible gameplay bug in the named module, identify the smallest responsible code path, propose a minimal fix, and list compile, automation, runtime, and rollback checks.
Blueprint ownership task
Map which behavior belongs in Blueprint and C++ for one system, identify authoritative assets and defaults, and return a migration or interface plan without inventing unavailable graph details.
Build failure task
Analyze one bounded Unreal build or packaging log, rank evidence-backed causes, request missing inputs, propose the least risky next checks, and stop before unrelated dependency changes.
Review task
Audit a proposed Unreal diff for lifecycle, ownership, threading, replication, asset references, error handling, tests, performance, packaging, and source-control scope.
Concrete outputs to retain
Task contract
Starting commit, scope, engine version, target, relevant evidence, allowed tools, acceptance checks, non-goals, and stop conditions.
File and asset map
Source files, Blueprint owners, assets, configuration, logs, tests, dependencies, and authoritative documentation connected to the task.
Validation receipt
Commands run, exit status, runtime reproduction, captures, performance evidence, packaging result, failures, and environment details.
Playable intent reference
A browser-playable slice that clarifies the mechanic and feedback while the native Unreal implementation remains separately reviewed.
Best fit and human-review boundary
Best for
- Narrow C++ investigations, reviews, and implementation plans
- Blueprint-to-C++ ownership and interface planning with real project evidence
- Build-log triage and repeatable validation tasks
Still needs human review
- A human must authorize write scope, terminal commands, network access, dependency installation, source-control actions, and publishing
- Unreal developers must inspect Blueprint assets and native code, compile, run, profile, package, and test target platforms
- Security, license, privacy, plugin compatibility, and production release decisions need accountable human owners
Official evidence and adjacent K3 Unreal routes
Capability, availability, architecture, and pricing claims on this page are bounded to Moonshot AI's July 2026 launch post. Social comparisons are treated as demand signals, not verified results.
Kimi Code × Unreal Engine workflow FAQ
Does Kimi Code support Kimi K3?
Moonshot AI's official launch says users can run Kimi Code in the terminal and select Kimi K3 with the model command. Availability, installation, authentication, supported platforms, tool behavior, and pricing can change, so verify the current official Kimi Code documentation before connecting a production repository or granting terminal capabilities.
Can Kimi Code directly edit an Unreal Engine project?
A coding agent can work with files and authorized terminal tools, but native Unreal projects also contain Blueprints, assets, editor state, generated data, platform settings, plugins, and build systems. Editing files is not proof of a valid project. Use explicit permissions, source-control checkpoints, compilation, automation, runtime tests, packaging, and human review.
What should an Unreal AGENTS.md or project instruction file include?
Include engine and plugin versions, module and asset ownership, build targets, approved commands, code style, Blueprint and C++ boundaries, generated directories, forbidden paths, source-control rules, test and packaging commands, target platforms, security and license constraints, product claims, escalation rules, and the required evidence for declaring a task complete.
How should Kimi Code handle Blueprint work?
The task pack should identify authoritative Blueprint assets, parent classes, interfaces, defaults, linked data assets, runtime states, and expected behavior. Screenshots can support discussion but may omit crucial graph information. A human Unreal developer must open the actual project, inspect and compile graphs, run the reproduction, and verify serialization and packaging.
Which terminal permissions are safest to start with?
Begin with read-only repository inspection and explicitly allow each command category. Separate file writes, builds, tests, network access, package installation, editor automation, source-control mutation, and destructive actions. Use timeouts, working-directory restrictions, secret filtering, audit logs, and checkpoints. Broader access should follow demonstrated need, not convenience.
Does the generation button run Kimi Code?
No. The button opens SEELE AI's generation page with a complete browser-playable game brief and full attribution parameters. The prompt does not name or select a model, execute Kimi Code, upload a repository, or promise a native Unreal project. It creates a separate prototype direction for review.
What is a good first Kimi Code task for Unreal?
Start with a read-heavy, bounded task: map one gameplay system, review a small diff, explain one build error, or design a minimal reproduction. Provide the starting commit and acceptance checks, restrict tools, and require an evidence-based handoff. Avoid project-wide refactors until repository navigation, permissions, validation, and recovery behavior are proven.
Test the playable direction before native Unreal production
The prompt describes the complete game slice and does not select a model. This final route keeps the paid-download reminder and full attribution chain attached.