惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

D
DataBreaches.Net
B
Blog
博客园_首页
C
Check Point Blog
Microsoft Security Blog
Microsoft Security Blog
MyScale Blog
MyScale Blog
P
Proofpoint News Feed
Engineering at Meta
Engineering at Meta
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
B
Blog RSS Feed
M
MIT News - Artificial intelligence
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
WordPress大学
WordPress大学
宝玉的分享
宝玉的分享
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
The Cloudflare Blog
量子位
V
V2EX
Y
Y Combinator Blog
Hugging Face - Blog
Hugging Face - Blog
Martin Fowler
Martin Fowler
Recent Announcements
Recent Announcements
I
InfoQ
博客园 - 【当耐特】

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
GitHub - av29nassh-sketch/PreFlight: The local security g...
Avenassh · 2026-06-23 · via Hacker News - Newest: "AI"

Stop AI Coding Drift before it becomes production technical debt. PreFlight is a local-first safety gate for AI-generated code, built to catch unsafe auth, RLS, SQL, SSRF, command execution, dependency, and secret-handling changes before they get committed.

Website: https://preflight-vibe.vercel.app

Choose Your Remediation Depth

PreFlight runs in two distinct tiers depending on what your codebase needs.

Free Tier: PreFlight Guardian

  • What it does: Unlimited local scanning plus 10 free patch applications across local deterministic fixes and proxy-backed AI fixes.
  • Setup: Zero config for scanning. A Pro key is only required after the 10 free patches are used.
  • Commands:
npm install -g preflight-pro
preflight init
preflight scan . --fix

Installing preflight-pro exposes the universal preflight command in your shell.

Pro Tier: PreFlight Pro

  • What it does: Unlimited scans and unlimited fixes, including deep reasoning remediation for complex multi-file architectural flaws, tenant isolation logic, and parametric SQL injections.
  • Setup: Requires an active PREFLIGHT_PRO_KEY or a saved key from preflight auth.
  • PowerShell:
$env:PREFLIGHT_PRO_KEY="PREFLIGHT-BETA-XXXXX"
preflight scan . --fix
  • Bash / macOS:
export PREFLIGHT_PRO_KEY="PREFLIGHT-BETA-XXXXX"
preflight scan . --fix

Installation Flow

PreFlight supports both a terminal-first workflow and an IDE-first workflow. Both paths end with preflight init, because that wizard connects your editor, MCP clients, and Pro/Beta key in one place.

Path A: CLI

npm install -g preflight-pro
preflight init

Then scan any project from its root:

Path B: VS Code / Cursor

  1. Install the global CLI command. The VSIX gives you the in-editor UI, but the extension still uses the global preflight command to start The Eye daemon and run fixes.
npm install -g preflight-pro
  1. Download and install the PreFlight Companion VSIX extension:
  1. Run the setup wizard once:
  1. Open your project in the IDE. The extension starts The Eye automatically, watches file saves, and surfaces PreFlight alerts in-editor.

The Eye and MCP

  • The Eye: The VS Code/Cursor extension starts PreFlight's local daemon automatically. It watches file saves and raises in-editor alerts when AI-generated code introduces a hard-block issue.
  • MCP bridge: preflight init can also wire preflight mcp into supported AI editors so agents can call PreFlight tools without leaving the coding flow.

Beta / Pro Keys

Free users get unlimited scans and 10 total patches across local fixes and proxy-backed AI fixes. After the 10 free patches are used, unlimited fixes require a Pro/Beta key.

You can add your key during preflight init, or activate it directly:

preflight auth PREFLIGHT-BETA-XXXXX

For one terminal session, you can also set it manually:

$env:PREFLIGHT_PRO_KEY="PREFLIGHT-BETA-XXXXX"
export PREFLIGHT_PRO_KEY="PREFLIGHT-BETA-XXXXX"

Pricing

  • Free Tier: Unlimited scans, 10 Free Patches (Local + Deep-Reasoning AI).
  • Solo Pro: $19/mo for unlimited scans and fixes.
  • Teams: $49/seat/mo for team rollout, shared onboarding, and unlimited scans and fixes.

Engine Upgrades

PreFlight is now powered by deeper local analysis primitives:

  • Micro-Fuzzer: Generates focused security payloads for risky data-flow paths, such as SQL injection, command injection, auth bypass, SSRF, and path traversal.
  • Quantized CPG (Code Property Graph): Builds a compact in-memory graph of syntax, control flow, and data flow so PreFlight can trace untrusted input into dangerous sinks instead of relying on brittle string matching.
  • The Eye daemon: Runs locally through the CLI/extension workflow and watches file saves so issues appear while the AI coding session is still active.

Tri-State Risk Score Engine

This is the core PreFlight signal. Every scan resolves into one of three clear outcomes so you know whether to stop, review, or ship.

Score Meaning What It Catches
🔴 Hard Block Stop immediately. This change is unsafe to ship. Exposed frontend secrets, leaking database service roles, command execution, SQL injection, or missing Supabase Row Level Security (RLS).
🟡 High-Risk Drift Review carefully. The code may be structurally wrong even if it runs. Structural state inconsistencies, un-idempotent webhooks, weak validation, or open CORS contexts.
🟢 Pass Safe to continue. No blocking structural risk was detected. Standard local edits matching your expected stack rules.

2-Phase Pipeline

PreFlight runs fixes in a strict sequence:

  1. Phase 1: Offline Local AST Sweep PreFlight completes an ultra-fast offline structural pass first and applies any deterministic local fixes it can resolve safely.
  2. Phase 2: PreFlight Pro Deep Reasoning Handoff Remaining SQL, fuzzer, and complex architectural flaws are handed off through the secure proxy-backed reasoning path when a patch requires deeper context.

The first 10 patch applications are free across both phases. After that, a PREFLIGHT_PRO_KEY is required.

Editor & MCP Usage

PreFlight can run directly in the terminal, through the VS Code/Cursor extension, or as an MCP server for AI-native editors.

Start the MCP server locally:

Available MCP tools include:

  • scan_project
  • preflight_fix
  • audit_dependencies

scan_project remains free and unlimited. preflight_fix shares the global 10-patch free allowance before a PREFLIGHT_PRO_KEY is required.

Post-Fix Verification Loop

PreFlight is designed to be used as a closed loop, not a one-shot scanner:

  1. Generate or modify code with your AI coding assistant.
  2. Run preflight scan . to classify the change under the Tri-State Risk Score.
  3. If PreFlight returns Hard Block, stop and repair the structural issue before moving forward.
  4. If PreFlight returns High-Risk Drift, run preflight scan . --fix and inspect every proposed fix before applying it.
  5. Re-run preflight scan . after each accepted fix to confirm the repository settles into Pass.
  6. Ship only after the final verification pass is green and the structural receipt matches the architecture boundary you intended.

This verification loop is the product: scan, review, patch, re-scan, then deploy with confidence.