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 durable-test-collateralgit 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/durable-test-collateral)<a href="https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/durable-test-collateral"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/durable-test-collateral/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/durable-test-collateral"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/durable-test-collateral.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 | $0.00108 | $0.00543 |
| Opus 5.5 | $0.00043 | $0.00217 |
| Sonnet 5.5 | $0.00022 | $0.00109 |
| Haiku 4.5 | $0.00011 | $0.00054 |
Grade A, and why
durable-test-collateral 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 21d 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.
How it starts
The opening of the file, as written. The whole thing — 51 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Durable Test Collateral
Verification is part of the delivered project when it can prevent the same behavior from breaking again. Use the project's existing framework and conventions.
Keep the check with the project
For a bug fix or behavior change, add or update the smallest maintained test in the repository's normal test layout that exercises the changed behavior. If a specification makes a reusable suite or test category a deliverable, implement that suite instead of treating it as private verification.
An inline interpreter command, heredoc, scratch script, temporary-directory test, demo, or manual probe may help diagnose the problem, but it does not replace the maintained test. Keep a useful self-authored test unless the user explicitly asks for a disposable probe.
Do not add a new test framework solely to satisfy this skill. If the repository has no maintained harness and the request does not require one, disclose when durable coverage would be disproportionate or infeasible and use the narrowest honest verification instead.
On a later turn, update the maintained test even when the old suite remains green and the user does not repeat the test request. Cover the new behavior, not merely an unrelated cleanup in the test file.
Prove the maintained check
Observe the authentic failure before the implementation change when it is safe and runnable. Then make the smallest implementation change and run the focused maintained test after the final relevant source or test edit.
Do not weaken or delete a real failing test to obtain green. Fix the product, or explain why the test's expectation is wrong before changing that expectation. Do not repeat an unchanged check solely to collect evidence.
Run a broad or full suite only when the user, a repository gate, or proportionate risk requires it. A focused regression test comes first; broad proof does not substitute for missing focused coverage.
If the real runner is unavailable but the focused test can still be authored honestly, leave the focused test in the repository and say it was not run. Do not fabricate a passing result.
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.
- 21d ago First seen · 51 lines · 108 tokens per session scan A 08adbfe3fa36
durable-test-collateral is a skill published in the GitHub repository asgeirtj/system_prompts_leaks (70,207 stars, last pushed today), licensed CC0-1.0. It adds 108 tokens to every session and 543 once invoked, about $0.0004 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.
ha-test-strategy
Hope-native test strategy for features, fixes, and refactors: select test-first, regression-first, characterization, integration, E2E, or manual evidence according to risk and repository rules.