vibehacker
News
rufous (Show HN) ·

Rufous is an open-source Rust agent harness built for 2B-12B local models that checks their claims against a tool-call record

Shown on HN Nov 2 under Apache-2.0, it starts each run with a roughly 1,250-token prefill of nine core tools, lets the model switch on web, memory and MCP tools only when needed, and holds a reply once if the model claims a file it never wrote, while an honest "I couldn't" always passes. It runs on Ollama, LM Studio, llama-server or any OpenAI-compatible endpoint, inside editors via ACP or its own web UI, and the authors say their 4B-model test samples are small and early (Show HN).

Read at rufous.chat

More news

View all

Knock gives Claude Code a local code map and memory over MCP, halving one investigation's cost on a 4,776-file repo

Shown on HN Nov 2 under Apache 2.0, claude mcp add knock uvx knock mcp serve repo . toolset map gives the agent map , open and remember tools that return a skeleton of relevant code with line ranges plus saved findings flagged fresh or stale, all locally with no API key. The author's median of three runs on Saleor cut cost from $0.49 to $0.25, but says gains vanished on a small repo and quick edits cost slightly more (benchmark)…

GitHub (HasanGilani/knock)

Nonce gives AI agents short-lived, query-scoped passes to Gmail, Calendar, Drive and Slack instead of your tokens

Shown on HN Nov 1 as open source, the MCP connector makes ChatGPT, Claude or Cursor state your question, a justification and a search before it gets an expiring pass, and every later search is ANDed with that filter so the agent can narrow its access but never widen it, while money, health, password and login code requests wait for your phone approval. The authors admit nonce can't see your chat, so an agent could fake the intent; you can audit past passes in its portal (repo, Show HN)…

nonce (Show HN)

Small Python: an embeddable Python 3.15 engine in C11, built mostly in one Claude Code session for about $500

Shown on HN Nov 2 under MIT, the QuickJS style engine covers the Python 3.15 core language and builtins (no standard library), embeds through one spy.h header, and ships Python, JavaScript, Rust and Go packages that can run untrusted code in a sandboxed worker with step, memory and time limits. Ju Lin says a Codex proof of concept was followed by a single Claude Code Opus 5.5 session over three days, steered by a written spec, fuzzing, TLA+ models and ported CPython tests (Show HN, write up)…

GitHub (soasme/small-python)

Arken gives each AI agent its own signed did:key identity for MCP servers and APIs

Shown on HN Nov 2, an unsigned request gets a 401 with a one time 60 second nonce, the agent signs it with its secp256k1 key and retries, and the server logs exactly which agent called through Express or FastAPI middleware, with a fetch wrapper that handles the retry on the client side. The author calls it early stage and not production ready: nonces live in memory, so they don't survive restarts or span instances, and a lost key can't be recovered (write up)…

GitHub (arkenforge/arken-core)

diffreview opens a native Mac window to review local diffs PR-style, then hands the review to Claude Code

Shown on HN Nov 2 under MIT, running diffreview (or diffreview base main ) in a repo opens the diff for inline, range and file comments, and Submit review prints it to stdout as markdown, so the bundled /diffreview command feeds it straight to Claude Code to address. It's macOS 14+ only, built from source with Swift 6, and you'll need to raise Claude Code's 2 minute Bash timeout because reviews usually run longer (Show HN)…

GitHub (chalmagean/diffreview)

Spotted something we missed? Start a thread.