System Prompts Leaks is a collection of captured system instructions used to guide AI chatbots and coding agents before they receive user messages. It serves researchers and developers studying how different AI assistants are directed.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add asgeirtj/system_prompts_leaks --skill 3d-objectgit clone --depth 1 https://github.com/asgeirtj/system_prompts_leaksWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/3d-object)<a href="https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/3d-object"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/3d-object/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/3d-object"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/3d-object.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- Snyk pass
- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00014 | $0.01212 |
| Opus 5 | $0.00007 | $0.00606 |
| Sonnet 5 | $0.00003 | $0.00242 |
| Haiku 4.5 | $0.00001 | $0.00121 |
Grade A, and why
3d-object scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
Copies of this mod
1 near-identical copy found in the catalogue:
- 3d-object — 97% identical, 3 lines differ
How it starts
The opening of the file, as written. The whole thing — 80 lines — stays where its author put it; the contents beside it link to each section on GitHub.
3D object
Model a 3D object the user can inspect from every angle and download, built with three.js and presented in the three_d_stage starter.
START by calling copy_starter_component with kind: "three_d_stage.js" — it is the whole viewer: studio lighting, ground shadow, orbit controls, an auto-framed camera, and a toolbar that downloads the object as OBJ + MTL or GLB. Read the copied file's usage block and follow its page skeleton exactly. You only write the model-building module script.
Build the page as plain HTML — a .html file with ordinary tags — even when the project's other designs are .dc.html Design Components: DC restricts scripts to , which races the stage's mount and can't host this skeleton.
Load three.js ONLY through this exact pinned import map, in , before any module script. Do not change versions, URLs, or hashes, do not add other copies of three.js, and do not import addons beyond the three listed — the map is deliberately a closed set, so anything else fails to resolve rather than loading unverified:
Build the model programmatically, as a THREE.Group composed of named parts:
- Compose primitives (BoxGeometry, CylinderGeometry, SphereGeometry, TorusGeometry, LatheGeometry, ExtrudeGeometry with Shape) before reaching for raw BufferGeometry; real objects decompose into far more primitives than you'd guess.
- NAME every mesh and every material ("hull", "walnut", "brass") — the names become the o / usemtl entries in the exported OBJ and the node names in the GLB, which is what makes the download usable in Blender.
- Use MeshStandardMaterial with a small curated palette (3-5 materials, shared across parts); set roughness/metalness deliberately. Textures don't survive the OBJ export — prefer geometry and material color over texture detail.
- Model in real-world meters, y-up, centered on the origin, base resting at the lowest y. Offset deliberately coplanar faces by ~0.001 so nothing z-fights.
- Curved surfaces need enough segments to read as smooth at full screen (32+ radial segments on feature surfaces), but don't tessellate what no one will see.
The stage's toolbar gives the user OBJ + MTL (universal, geometry + per-material colors) and GLB (the modern interchange format — keeps the part hierarchy and PBR materials; imports cleanly into Blender, Maya, Cinema 4D, Unity, Unreal). Those two are the export formats on offer — when the user asks for something else (FBX, USDZ, STEP), say plainly that the stage exports OBJ + MTL and GLB.
Iterate with screenshots: the stage keeps its last frame readable, so the ordinary screenshot tools capture the live canvas — no extra steps. After editing a module file, reload with show_html before screenshotting (the iframe caches already-loaded modules). Look at the object from the default framing and refine silhouette, proportion, and material separation — the silhouette carries the object.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 11d ago First seen · 80 lines · 14 tokens per session scan A b2fca2c7f0ca
3d-object is a skill published in the GitHub repository asgeirtj/system_prompts_leaks (64,614 stars, last pushed yesterday), licensed CC0-1.0. It adds 14 tokens to every session and 1,212 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
claude-design
Design one-off HTML artifacts (landing, deck, prototype).
sketch
Throwaway HTML mockups: 2-3 design variants to compare.
impeccable
Frontend design guidance, upstream-maintained (impeccable).
genui
Render structured, glanceable, interactive UI in the chat instead of plain text — dashboards, stat cards, filterable tables, charts, mermaid diagrams, choice panels, forms, quizzes. Panels can carry an id so later turns update them in place, and most interaction resolves in the browser with no round-trip. Read this…
motion-system
Define the app's motion language before animating anything -- a small set of durations, easings, and named moves that every animation draws from, so the app feels designed instead of decorated. Use before writing the first animation in a project, or when existing motion feels scattered.
polish-pass
The de-AI pass -- audit a vibe-coded app for template smell in design and copy, then fix it with editorial layout, purposeful motion, and human-sounding words. Use after any build session, before anyone else sees it.