Skillget

8v

mcp
Guvenlik Denetimi
Gecti
Health Gecti
  • License — License: NOASSERTION
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 21 GitHub stars
Code Gecti
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested
Purpose
This tool is a unified, all-in-one binary designed for AI coding agents like Claude Code. It replaces the agent's native file manipulation and execution tools with optimized commands for reading, writing, searching, building, and testing codebases to save processing tokens.

Security Assessment
Overall Risk: Medium. The core functionality of this tool inherently requires broad system access. It is explicitly designed to read, write, and modify local files, execute builds, run tests, and spawn subprocesses. Because it intercepts and replaces the agent's standard tools, it sits in a highly privileged position.

While the automated code scan found no hardcoded secrets or dangerous patterns in the scanned files, there is a significant installation risk: the primary setup method uses an unsigned remote script (`curl | sh`). If you do not trust the publisher, this is a potential attack vector. To mitigate this, you can safely review the underlying Rust source code and compile the binary yourself.

Quality Assessment
The project demonstrates high engineering quality and active maintenance. It is written in Rust, which provides strong memory safety guarantees, and the codebase is organized into a clean workspace of modular, purpose-built crates. It was updated very recently, indicating active development. However, community trust is currently limited given its small footprint of 21 GitHub stars. The licensing structure requires attention: while some core libraries are MIT, the main binary uses BSL-1.1 (converting to Apache 2.0 in 2030), meaning it is currently source-available rather than strictly open-source.

Verdict
Use with caution: while the code is clean and well-structured, the tool inherently requires extensive file system and execution permissions, and you must fully trust the developer to allow it to act as a proxy for your AI agent.
SUMMARY

8v is one binary that reads, writes, searches, checks, builds, tests, and formats code, using fewer input and output tokens than the agent's native tools.

README.md

8v

You and your AI coding agent — same tool, same rules.

8v is one binary that reads, writes, searches, checks, builds, tests, and
formats code, using fewer input and output tokens than the agent's native tools.

Today, 8v supports Claude Code. More agents next.

Install

curl -fsSL https://install.8vast.io | sh

Or build from source:

cargo build -p o8v --release
# binary at target/release/8v

Setup

In any project:

8v init

8v init replaces Claude Code's native Read, Edit, Write, Grep, and Glob with 8v. An MCP server is also available: 8v mcp. Bash stays available for git, processes, and environment.

Commands

8v ls                       # list files and projects in the tree
8v read src/main.rs         # symbol map (functions, types) — fast
8v read src/main.rs:40-80   # exact line range
8v read src/main.rs --full  # whole file
8v search "fn parse"        # regex across the codebase
8v write path:12 "..."      # replace a line
8v write path --insert      # insert, --delete, --append
8v write path --find a --replace b
8v check .                  # lint + type-check + format-check, every stack
8v fmt .                    # auto-format
8v build .                  # invoke the project's build tool
8v test .                   # invoke the project's test runner
8v log                      # what ran in this session
8v stats                    # what failed most often
8v upgrade                  # update 8v itself

Every command accepts --json for structured output.

Stacks

Rust, TypeScript, JavaScript, Python, Go, Deno, .NET, Ruby, Java, Kotlin,
Swift, Terraform, Dockerfile, Helm, Kustomize, Erlang. Shell files always
go through shellcheck and shfmt.

Benchmark

Each scenario gives Claude a broken codebase and asks it to fix it.
Native tools vs. 8v, one variable at a time.

Scenario Input tokens Output tokens
fix-failing-test (Rust) −14% −42%
fix-go −21% −52%
fix-python −66% −66%
fix-typescript −12% −39%

N=6 per condition. Tests pass 6/6 in every scenario, both arms.

cargo test -p o8v --test demo_agent_benchmark -- --ignored --nocapture --test-threads=1

Workspace

cargo build -p o8v --release   # binary
cargo test --workspace         # full suite
8v check .                     # 8v on itself

Crates: o8v (binary), o8v-core (commands and rendering), o8v-fs
(safe filesystem), o8v-process (safe subprocess), o8v-stacks (stack
detection and dispatch), o8v-check (lint/type/format orchestration),
o8v-testkit (test infrastructure).

License

o8v-fs and o8v-process are MIT.
All other crates are BSL-1.1, converting to Apache 2.0 on April 5, 2030.

Copyright (c) 2026 Soheil Alizadeh / 8Network. All rights reserved.

Yorumlar (0)

Sonuc bulunamadi