TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Show HN: Plan-linter – pre-flight safety checker for AI agent plans

4 点作者 mercurialsolo16 天前
just released our plan-linter – a tiny static-analysis tool that catches &quot;obvious-stupid&quot; failures in AI agent plans before they reach runtime.<p>GitHub repo -&gt; <a href="https:&#x2F;&#x2F;github.com&#x2F;cirbuk&#x2F;plan-lint">https:&#x2F;&#x2F;github.com&#x2F;cirbuk&#x2F;plan-lint</a><p>also read on how to deal with safety using a 4-step safety stack (“No Safe Words”) → <a href="https:&#x2F;&#x2F;mercurialsolo.substack.com&#x2F;p&#x2F;no-safe-words" rel="nofollow">https:&#x2F;&#x2F;mercurialsolo.substack.com&#x2F;p&#x2F;no-safe-words</a><p>Why?<p>Agents now emit machine-readable JSON&#x2F;DSL plans. Most prod incidents (loops, privilege spikes, raw secrets) could have been caught by scanning those plans offline, yet everyone focuses on runtime guardrails.<p>What it does<p>* Schema + policy validation (JSONSchema &#x2F; YAML &#x2F; OPA)<p>* Data-flow + taint checks for secrets &amp; PII<p>* Loop detection (graph cycle)<p>* Risk score 0-1, fail threshold configurable<p>* Plugin rules via entry_points<p>Runs in &lt;50 ms for 100-step plans, zero token cost.<p>how are you dealing with safety (budget overruns, token leaks) when deploying agents in prod with tool access?

暂无评论

暂无评论