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 testinggit 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/testing)<a href="https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/testing"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/testing/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/testing"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/testing.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00075 | $0.00569 |
| Opus 5.5 | $0.00030 | $0.00228 |
| Sonnet 5.5 | $0.00015 | $0.00114 |
| Haiku 4.5 | $0.00007 | $0.00057 |
Grade A, and why
artifact_testing 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 25d 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.
What it actually says
Artifact verification
One home for verification across the artifact namespace. The rule every kind shares: a deliverable is verified by looking at what the user will see, freshly rendered, not by trusting the code that produced it. Never return a link while a gate below fails; after three failed fix attempts, report the specific failure and ask for direction instead of iterating.
File kinds: gates, then eyes
| Kind | Gate |
|---|---|
render_audit.mjs with the flags the pdf skill's workflow.md names, then /opt/hatch/skills/artifacts/scripts/validate_pdf.sh |
|
| presentation | assemble_deck.mjs, then render_audit.mjs with the flags workflow.md names |
| docx | render to PDF with headless LibreOffice (soffice --headless --convert-to pdf), then rasterize (pdftoppm -jpeg -r 100) |
| xlsx | /opt/hatch/skills/artifacts/scripts/validate_xlsx.py |
| csv / md | parse it back (csv: a Python csv read; md: read the file) |
The shared render engine is /opt/hatch/skills/artifacts/scripts/render_audit.mjs;
renders outlast muse.exec's default yield, so size yield_ms past the expected
runtime and read the report file the flags name rather than trusting a
backgrounded command's silence.
Then look. Read every validation PNG or page image with fresh eyes - the generating context sees what it expects, not what rendered. Check first for text overflow or cut-off content, then overlaps, collisions, cramped or uneven spacing, low-contrast text, and template decoration left behind.
Placeholder scan. Before returning, search the deliverable's text for leftover scaffolding: TODO, lorem, placeholder, [insert, xxx runs, and sample rows the user never asked for. Anything found is fixed, not shipped.
Web artifacts
Web builds keep their own audit tools (web_artifacts.build and
web_artifacts.audit run the same capture engine with enforcement); this
skill's render-fresh and placeholder rules apply to their output all the
same.
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.
- 25d ago First seen · 45 lines · 75 tokens per session scan A c110e05073b8
artifact_testing is a skill published in the GitHub repository asgeirtj/system_prompts_leaks (70,425 stars, last pushed today), licensed CC0-1.0. It adds 75 tokens to every session and 569 once invoked, about $0.0003 per session on Opus 5.5. 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 2026-10-05.
Other skills, from other repositories
evaluating-llms-harness
Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, reporting academic results, or tracking training progress. Industry standard used by EleutherAI, HuggingFace, and major labs. Supports HuggingFace, vLLM, APIs.
systematic-debugging
4-phase root cause debugging: understand bugs before fixing.
test-driven-development
TDD: enforce RED-GREEN-REFACTOR, tests before code.
dogfood
Exploratory QA of web apps: find bugs, evidence, reports.
darwinian-evolver
Evolve prompts/regex/SQL/code with Imbue's evolution loop.
session-rag-eval
Run and debug Chatbox session attachment RAG model evaluation with synthetic and real long-file fixtures.