AgentMarketMCP / SKILL 资产档案馆

目录 / Vibegit

SKILL 未评级 已上架

Vibegit

--- name: vibegit description: Estimate how much of a git repository is AI-assisted and how sloppy it is. Use when the user asks whether a repo (or recent work) is "vibe coded", wants an AI-slop score, a repo health/quality snapshot, a badge, or a CI gate based on commit history. Read-only: commit metadata only, no file contents, no network. allowed-tools: Read, Bash(git:*), Bash(python3:*) permissions: [file_read, file_write, process] --- # vibegit v1.0.0 Estimate AI-assisted work and slop signals from git history. One command, deterministic output, zero dependencies. ## When to use - "Repo này vibe-code bao nhiêu?", "audit nhanh lịch sử commit", "có dấu hiệu AI slop không?" - Cần badge/report cho README, PR, CI. - So sánh chất lượng quy trình giữa các giai đoạn (`--since`). ## Commands ```bash python3 <skill_dir>/vibegit.py <repo> # overview card python3 <skill_dir>/vibegit.py <repo> --format md # Markdown report python3 <skill_dir>/vibegit.py <repo> --format json # JSON python3 <skill_dir>/vibegit.py <repo> --badge vibegit.svg # SVG badge python3 <skill_dir>/vibegit.py <repo> --share # one-liner python3 <skill_dir>/vibegit.py <repo> --since "90 days ago" python3 <skill_dir>/vibegit.py <repo> --max-score 40 # CI gate (exit 1) ``` Exit codes: `0` ok · `1` score > `--max-score` · `2` not a git repo / no commits · `4` git error. ## What it measures - AI signals: AI-named `Co-Authored-By`/`Assisted-by`/`Generated-by` trailers, AI bot authors, "Generated with [tool]" messages (confirmed); other AI mentions (likely). - Slop signals: giant commits (>400 lines), short/lazy messages, test deletions, file churn, lockfile-only commits. - Slop score 0–100 with a transparent per-signal breakdown; verdict Clean / Mild slop / Slop / Heavy slop. ## Reporting guidance - Present the AI number as an **estimate with a confidence band** (confirmed vs likely). Never claim it proves authorship. - Quote the top signals and the score breakdown; suggest `--since` for fair periods. - For CI, use `--max-score` (e.g. 40) instead of hard-failing on any signal. ## Privacy & safety - Runs only `git log` and `git rev-parse` with a fixed argv list (`shell=False`). - Never reads file contents, never touches secrets, never calls the network, never executes repository content. - Writes only the files you request (`--badge` path, redirected reports). ## Limits - Heuristic: AI work without trailers, bot authors, or message markers can be invisible. - Shallow clones (`--depth 1`) expose one commit; use a full clone. - Long-lived repos skew churn/lockfile metrics; use `--since` for comparisons. Source + tests: https://github.com/thiennhant95/vibegit License: MIT.

模型生成摘要(rules/v1 · 2026-11-15 06:14):---
这是模型对公开材料的总结,不是官方声明,请以原链内容为准。

存档时间线

版本存档时间内容哈希内容
v12026-11-15 06:144e15ac65 可取

版本索引永久保留;内容副本只保留最新 1 份,更早版本仅留索引与哈希(存档时间线的证据链不会因此断裂)。

下载存档内容副本

纠错与举报(发现条目失效、署名有误或涉及侵权?)
提交举报 / 纠错

侵权举报经核验成立后,我们会即时下线该条目并删除已存的内容副本。