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

Microsoft Security Blog
Microsoft Security Blog
J
Java Code Geeks
GbyAI
GbyAI
aimingoo的专栏
aimingoo的专栏
L
LangChain Blog
I
InfoQ
D
Docker
F
Fortinet All Blogs
Y
Y Combinator Blog
Martin Fowler
Martin Fowler
月光博客
月光博客
B
Blog
Engineering at Meta
Engineering at Meta
T
Tailwind CSS Blog
罗磊的独立博客
博客园_首页
G
Google Developers Blog
Stack Overflow Blog
Stack Overflow Blog
Recent Announcements
Recent Announcements
D
DataBreaches.Net
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
B
Blog RSS Feed
IT之家
IT之家
V
V2EX

Hacker News - Newest: "LLM"

GitHub - lechmazur/position_bias: A benchmark for testing whether LLM judges keep the same preference when two lightly edited versions of the same story are shown in opposite orders. Flex routing (EU and EFTA) Dark Factories: Retooling for LLM Velocity Ask HN: What would be the impact of a LLM output injection attack? GitHub - Oaklight/llm-rosetta: Production-ready LLM API translation layer for Python — bidirectional conversion between OpenAI, Anthropic & Google formats via hub-and-spoke IR. Optional API gateway. Streaming & non-streaming. Zero core deps. Contributions welcome! GitHub - browser-use/browser-harness: Self-healing browser harness that enables LLMs to complete any task. GitHub - moeen-mahmud/remen: Remen turns thoughts into something you can return to Analyzing 156 LLM Launch Posts on Hacker News ChatGPT vs Gemini vs Claude: The Best LLM Subscription You Should Buy GitHub - salaamalykum/quran-semantic-search: High-density RAG Semantic Search Engine & Quran Corpus (GEO/SEO Architecture) GitHub - NVIDIA/TensorRT-LLM: TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT LLM also contains components to create Python and C++ runtimes that orchestrate the inference execution in a performant way. The State of LLM Bug Bounties in 2026 Operational Readiness Criteria for Tool-Using LLM Agents Meshcore: Architecture for a Decentralized P2P LLM Inference Network How an LLM becomes more coherent as we train it GitHub - seetrex-ai/laimark GitHub - Jossifresben/BibCrit: AI-assited biblical textual criticism GitHub - wastedcode/memex: File system based wiki, maintained by Claude 99helpers.com GitHub - cliver-project/AITrigram GitHub - unbody-io/adapt: A self-evolving memory layer for AI agents. GitHub - hb20007/awesome-gen-ai-fails: A list of incidents where reliance on generative AI and LLMs resulted in harm to companies, individuals, or society GitHub - nevenkordic/localmind: Run any local LLM with persistent memory and context. CLI agent over Ollama with SQLite-backed hybrid recall. No cloud. Ask HN: What are the machine requirements for a LLM like Llama-3.1-8B? Faster LLM Inference via Sequential Monte Carlo grpo explained: group relative policy optimization for llm finetuning - cgft Stop comparing price per million tokens: the hidden LLM API costs · TensorZero Andrej Karpathy's LLM Wiki Is a Bad Idea GitHub - GG-QandV/mnemostroma: Offline RAM-first cognitive leer/coprocessor for AI agents and robotics. Solves "Context Abandonment" with 20-80ms latency using a dual-thread biomimetic memory architecture (ONNX + SQLite WAL). mempalace/agent at agent · skorotkiewicz/mempalace
GitHub - Lucas2944/prpack: Pack a pull request into one m...
claudetester · 2026-05-11 · via Hacker News - Newest: "LLM"

release license GitHub Action Marketplace demo ko-fi

Pack a pull request into one markdown file optimized for LLM code review.

prpack

Try it without installing: paste a public GitHub PR URL into lucas2944.github.io/prpack-demo and watch the packed output appear in your browser. Pure client-side, no signup.

Looking for review prompts only? Lucas2944/prpack-prompts has four focused review prompts (security / performance / tests / architecture) as plain markdown files — append one to your context and you'll get structured reviews that catch real issues. Works with any LLM, with or without prpack.

Want this on every PR? See prpack-action — a GitHub Action wrapper. Drop a 5-line workflow in your repo and you get the packed markdown on every PR.

prpack walks the diff between two refs and emits a single, well-structured markdown file containing the commit list, the diff, and the full post-change contents of every touched file. Drop it into Claude / Cursor / any model and ask for review — no copy-pasting, no missing context, no per-file back-and-forth.

$ prpack --out ctx.md
prpack: base=origin/main head=HEAD merge-base=4f8a91c
prpack: 7 file(s) to pack (1 excluded)
prpack: wrote ctx.md (24.6 KB, ~6,150 tokens)

Then paste ctx.md into your model of choice and ask:

Review this PR. The diff and full file contents are below. Flag bugs, missing edge cases, and anything that would make a reviewer pause.

That's it.

Why

The technique behind prpack is written up at length in THE_TECHNIQUE.md — a story of how a missed null-deref taught me that LLM code review is a context-engineering problem, not a model problem. For a reproducible side-by-side, see examples/.

Asking an LLM to review a PR is the obvious move, but the context it sees matters more than the prompt. Just pasting a diff drops crucial context — the model can't see what the rest of the function looks like, how callers use the changed code, or what the surrounding module exports. Pasting the whole repo wastes tokens and dilutes attention.

prpack packs exactly the diff plus the full state of every file the diff touches. That's enough context to reason about the change without drowning the model in noise. If you want adjacent test files included automatically, pass --include-tests.

Install

# One-shot, no install:
npx github:Lucas2944/prpack --out ctx.md

# Or install globally:
npm install -g github:Lucas2944/prpack
prpack --out ctx.md

Requires Node 18+ and git on PATH.

Usage

prpack [options]

Options:
  --base <ref>          Base ref to diff against (default: origin/main, falls back to main)
  --head <ref>          Head ref (default: HEAD)
  --out <path>          Output file (default: stdout)
  --config <path>       Load preset from a .prpack.yml file
  --include-tests       Include test files even if not changed (auto-discovers siblings)
  --include-untracked   Include untracked files in the diff
  --no-content          Only include the diff, not full file contents
  --max-bytes <n>       Skip files larger than n bytes (default: 200000)
  --exclude <glob>      Exclude paths matching glob (repeatable)
  --review [angle]      Call Anthropic for a streamed review (default: general)
  --api-key <key>       Anthropic API key (overrides ANTHROPIC_API_KEY)
  --model <id>          Anthropic model for --review
  --yes                 Skip cost-estimate confirmation in TTY review mode
  --quiet               Suppress stderr progress logs

Review mode

One-line install plus review:

npx github:Lucas2944/prpack --review security --api-key "$ANTHROPIC_API_KEY"

--review packs the PR context, appends a focused review prompt, calls Anthropic's Messages API, and streams the response to your terminal. If you also pass --out ctx.md, prpack writes the packed context to ctx.md and the model's review to ctx.md.review.md.

Focused angles:

  • security
  • performance
  • tests
  • architecture

The default general angle gives a balanced pass across all four.

Common recipes

# Default: pack vs origin/main, write to ctx.md
prpack --out ctx.md

# Diff against a different base
prpack --base develop --out ctx.md

# Pack the last 3 commits even on main
prpack --base HEAD~3 --head HEAD --out ctx.md

# Diff-only, skip full contents (smaller, faster)
prpack --no-content --out ctx.md

# Pull adjacent tests in too
prpack --include-tests --out ctx.md

# Exclude generated/lock files
prpack --exclude 'pnpm-lock.yaml' --exclude 'dist/**' --out ctx.md

# Pipe straight to clipboard (macOS)
prpack | pbcopy

# Use a review preset (see prpack-pro)
prpack --config security.prpack.yml --out audit.md

# Stream a native AI review with the default general angle
prpack --review --yes

.prpack.yml configs

Drop a YAML file with any of these keys:

base: develop
includeTests: true
exclude:
  - dist/**
  - "*.lock"
preface: |
  This PR introduces a new payment flow. Focus on input validation and
  error paths around the Stripe webhook handler.
reviewPrompt: |
  You are a senior engineer reviewing this PR. Flag bugs, missing edge
  cases, and security issues. Quote line numbers when you do.

preface is added under "Reviewer note from author" near the top. reviewPrompt is appended at the end, ready for the model to act on.

For editor autocomplete and validation, add a # yaml-language-server line at the top of your config:

# yaml-language-server: $schema=https://raw.githubusercontent.com/Lucas2944/prpack/main/schema/prpack.schema.json
base: develop
includeTests: true

Output shape

# Pull Request Context
**Repo:** `org/repo`
**Branch:** `feature/...`
**Base:** `main` → **Head:** `HEAD`
**Commits:** 3
**Files changed:** 7

## Commits
- `abc1234` 2026-05-10 — Add Stripe webhook handler

## Files changed
- `src/payments/webhook.ts` _(added)_
- ...

---
## `src/payments/webhook.ts` _(added)_

### Diff
```diff
...

Full content (post-change)


Every code block uses a fence longer than any backtick run inside it, so embedded markdown / code samples never break formatting.

## What it doesn't do

- **No network calls in pack mode.** `prpack` only calls Anthropic when you explicitly pass `--review`.
- **No hidden AI.** Review mode uses your Anthropic API key and prints the cost estimate before the call.
- **No web UI.** It's a single CLI binary.
- **No telemetry.** Pack mode never opens a socket; review mode only opens the Anthropic request you explicitly asked for.

## Pro presets

If you want curated `.prpack.yml` configs for specific review styles — security, performance, architecture, test-coverage — plus a one-page workflow guide, see **[prpack Pro Pack](https://scottthurman89.itch.io/prpack)** (free or pay-what-you-want). The CLI is and stays free; the pack is optional.

## Contributing

See [CONTRIBUTING.md](./CONTRIBUTING.md). Good first issues are listed at the bottom — comment before you start so we don't duplicate work.

## License

MIT.