Unreal Suit
# Iron Breach Claude Skills: Summary Bundle Eight skills for working on Iron Breach with Claude. Source: `X:\IronBreach\Docs\ClaudeSkills\`. Each skill is a folder with `SKILL.md` (instructions) and `references/project-map.md` (shared project map: where things live, how to run the game without a keyboard, rules every change respects). Packaged `.skill` files are in `dist/`. ## At a glance | Skill | One-line purpose | Typical trigger | |---|---|---| | game-mechanic-inventor | Designs new mechanics that fit existing systems and canon | "What should we add?", "Design X" | | unreal-performance-surgeon | Finds what is expensive in UE5 and proposes measured fixes | FPS, stutter, "runs slow" | | multiplayer-desync-detective | Finds why server and clients disagree | Desync, rubber-banding, late join | | ai-playtest-director | Runs automated tests, hunts exploits, writes balance reports | "Is X overpowered?", run the checks | | art-style-guardian | Reviews art against the visual standard | "Does this fit?", asset or lighting review | | feature-debt-hunter | Finds unfinished, dead or duplicated features | "What did we forget?", pre-milestone audit | | game-economy-simulator | Simulates XP, loot and currency from real tuning data | Drop rates, level curves, grind | | developer-handoff-agent | Reconciles git/LFS state and plans conflict-free syncs | "What changed?", push, pull, handoff | --- ## 1. game-mechanic-inventor **Purpose:** Invents new mechanics (abilities, kaiju behaviour, mech systems, modes, progression, loot) that fit the game's design philosophy. **How it works:** Spends most of its effort reading first: existing mechanics, canon and tuning, with sources noted. Every proposal names the existing systems it plugs into and comes with the hooks and tuning knobs needed to build it. **Use when:** You want ideas or a design before any code or Blueprint is written. **Built for:** Iron Breach (Kaiju tiers, dual-pilot mechs, Sync, Clearance), but works for any game. ## 2. unreal-performance-surgeon **Purpose:** Finds what is costly in a UE5 project: assets, materials and shaders, lighting (Lumen, MegaLights), Nanite, Niagara, Blueprints and ticks, replication, streaming, memory. **How it works:** Measures first (starts from `stat unit`), classifies the kind of slowness, finds the single biggest cost and cuts only that. Each finding carries a number, a cause, a precise fix and an expected gain. It does not just turn the whole game down. **Use when:** FPS, frame time, hitches, load times, shader compiles, or "is this asset or map too expensive?" **Inputs:** Stat captures, logs, Unreal Insights traces, or reading the project. ## 3. multiplayer-desync-detective **Purpose:** Explains why machines disagree: rubber-banding, one-sided hits, missing abilities or animations on other screens, late-join differences, host-only or client-only behaviour. **How it works:** Collects the symptom as seen from each machine, then finds the first moment they diverged. The cause is one of four: a change without authority, a change that never replicated, a change that arrived too late or out of order, or different code run for the same event. It points at the line of code or Blueprint node. **Use when:** "Works in single player but not multiplayer", replication, RPCs, RepNotify, listen server, or host and client logs. ## 4. ai-playtest-director **Purpose:** Runs automated gameplay tests (dev-check fixtures, scripted command sequences, net and PvP harnesses, engine automation tests) and produces balancing reports. **How it works:** Starts from one question, runs the smallest set of tests that answers it, turns logs into metrics, and ends with tuning changes as data-asset edits with expected effect. Exploits get a regression check. **Use when:** Balance questions, exploit hunting, "is this weapon, ability, kaiju or mech overpowered?", soak-testing a map. ## 5. art-style-guardian **Purpose:** Reviews meshes, textures, materials, UI icons and screens, VFX, lighting and map looks against the project's visual standards. **How it works:** Establishes the standard from evidence, inventories what changed, renders side-by-side contact sheets, and judges each asset on a Pass/Fix/Reject checklist (palette, silhouette, readability, naming, import settings, performance, licence). It names the exact property, node or import setting to fix. It never deletes anything. **Standard:** Grounded, cinematic, sleek; gunmetal gray, deep navy, matte black, cyan accent `#3DE1FF`. **Use when:** Reviewing imported packs (Fab etc.), icons, a lighting pass, or "something looks off". ## 6. feature-debt-hunter **Purpose:** Finds partially implemented, abandoned, disconnected or duplicated features. **How it works:** Cross-checks three views: what the docs promise, what the code does, and what the content references. It looks for uncalled code, unheard events, unused cvars and enums, unreferenced assets, unbound Blueprints, TODOs, failing checks and stale untracked work. It returns a ranked list where every item cites a file or asset and ends in a decision: finish, cut, keep or ask. **Use when:** Before a milestone, demo, push or handoff, or "what is half-done?" ## 7. game-economy-simulator **Purpose:** Answers economy and progression questions with numbers: XP and level curves, Clearance, drop rates and rarity by enemy tier, currency sources and sinks, inflation, crafting and upgrade costs, mech ammo regen, bad-luck streaks, exploits. **How it works:** Builds a model from the project's real tables (never from memory, every input cited), runs thousands of simulated player sessions in a script, and proposes data-asset and curve-table edits. The model can be rerun after each change. **Use when:** "Is it too grindy or too generous?", time to reach a level or cap, loot feel. ## 8. developer-handoff-agent **Purpose:** Keeps one repo with several contributors coherent: Connor (C++), Shane (Blueprints, widgets, asset manager rows, content), Codex/Astra (bounded windows) and Claude (passes with result docs). **How it works:** Reads state from all sources first (git status, incoming and unpushed commits, stashes, index lock, accidentally staged files, Git LFS). Then it plans the moves with every path listed, collision notes, and the order to pull, build, verify, commit and push. Any surprise is a stop. **Use when:** "What changed?", "who touched what?", preparing a push or pull, writing a handoff, taking or releasing a work window. --- ## Installing - **claude.ai / Claude desktop app:** open a `.skill` file from `dist/` and choose Save skill (one file per skill). - **Claude Code:** copy the skill folders to `.claude/skills/` in the repo (shared via git) or `~/.claude/skills/` (personal). Update `references/project-map.md` when the project layout changes; the skills read it first.
Works with any agent that signs in to Agensi, such as Claude, ChatGPT, Claude Code, Cursor and Codex CLI.
One payment, lifetime access. 8 skills unlock instantly in your library.
What's included
8 skillsAutomate Unreal Engine playtests and balance checks to generate data-driven tuning reports and exploit audits.
Audit game assets for visual style consistency, naming conventions, and technical import settings.
Automate repository handoffs and safe integrations for multi-developer teams using Git LFS and Unreal Engine.
Finds abandoned code, orphaned assets, and implementation gaps between docs and source code.
Simulate game progression and loot tables using Python to find economy dead zones, dry streaks, and exploits.
Bypass creative tropes by forcing lateral synthesis between unrelated domains for original fiction and system design.
Diagnose Unreal Engine multiplayer desyncs by correlating server/client logs and auditing replication code.
Diagnostic tool for Unreal Engine 5 that identifies thread bottlenecks and proposes surgical performance optimizations.
See it in action
ai playtest director
Example session with this skill installed
Run a balance check on the IronBreach jump kit. Is the jump height allowing players to skip geometry in the tutorial map? Compare it against the standard leap.
- Read your context and instructions
- Compiled the ai playtest director
I have executed the jump kit balance checks using the `IB.JumpPresentation.Check` fixture across 50 simulated runs on the Tutorial_Alpha map. The logs confirm that the Jump Kit currently allows for a vertical displacement that bypasses the primary gate trigger. # Playtest: Jump Kit Geometry Skip (2023-10-27) ## Answer The Jump Kit is overpowered for traversal. It allows a vertical reach of 850 units, which is 150 units above the tutorial gate collision height, enabling a sequence break. ## Matrix and runs | Setup | Runs | Latency | Build | Log Path | | :--- | :--- | :--- | :--- | :--- | | Standard Leap | 25 | 0ms | Dev_0.4.2 | /Logs/Check_Std_001.log | | Jump Kit | 25 | 0ms | Dev_0.4.2 | /Logs/Check_JK_002.log | ## Metrics | Ability | Avg Height | Clearance Rate | Success Rate | | :--- | :--- | :--- | :--- | | Standard Leap | 600u | 0% | 100% | | Jump Kit | 850u | 92% | 100% | ## Exploits and degenerate strategies - **Gate Skip:** By jumping at the apex of the collision box on the North Tutorial Gate, players can bypass the "Collect Battery" objective entirely. - **Severity:** High (Sequence Break). - **Guard:** Increase collision volume height or clamp the `JumpZVelocity` for the tutorial kit. ## Tuning changes | Asset | Field | Old | New | Why | | :--- | :--- | :--- | :--- | :--- | | `DA_JumpKit_Basic` | `JumpZVelocity` | 900.0 | 750.0 | Prevent gate skipping while maintaining feel. | | `CT_MovementCurves` | `AirControl` | 0.45 | 0.35 | Reduce mid-air correction to prevent exploit. | ## Regression guards Added `IB.JumpPresentation.Check --MaxHeight 750` to the `Tutorial_Alpha` automation suite. **Next steps** - Apply the data-asset edits to `DA_JumpKit_Basic`. - Run the `IB.ClassKitsCheck` to ensure other abilities are not affected by the `AirControl` change.
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 these skills and uses them on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every skill in this bundle goes through our trust checks.
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.