dzhng/jevgrep

dzhng 资料

最近记录的来源活动: 2026年10月15日 23:42

SkillsMP 收录数据更新: 2026年10月17日 00:09

已收集 skills: 29

GitHub 星标: 2,724
GitHub Forks: 203

仓库 README

仓库 README 来源：README.md

查看 README 源文件

展开完整 README

![jevgrep — Same intelligence. About 30% lower cost.](assets/cover.png) # jevgrep [![npm](https://img.shields.io/npm/v/@dzhng/jevgrep?style=flat-square&color=ef5638)](https://www.npmjs.com/package/@dzhng/jevgrep) [![MIT license](https://img.shields.io/badge/license-MIT-blue?style=flat-square)](LICENSE) [![Node.js 22+](https://img.shields.io/badge/Node.js-22%2B-339933?style=flat-square)](apps/cli/README.md) [![Release](https://img.shields.io/github/actions/workflow/status/dzhng/jevgrep/publish.yml?style=flat-square&label=release)](https://github.com/dzhng/jevgrep/actions/workflows/publish.yml) **Same intelligence. ~30% lower cost.** Find code by asking what it does. In our ten-task SWE-bench comparison, Jevgrep successfully completed the same 8 of 10 tasks as the baseline, at lower cost. Coding agents spend part of every unfamiliar task finding the right files. Jevgrep gives them a place to start: ask a repository question, and `jg` returns relevant files, reading leads, and verbatim source excerpts in one stdout response. It uses [Jev](https://vercel.com/ai-gateway/models/jev) to judge relevance across folders, files, and declarations. Your coding agent then implements and tests the change. ```sh npm install -g @dzhng/jevgrep jg auth jg skill jg "How are telemetry events recorded and sent?" ./my-project ``` Requires **Node.js 22+**, **macOS, Linux, or Windows**, and a key for **Vercel AI Gateway, TypeSafe, OpenRouter, OpenCode Zen, or a custom TypeSafe-compatible endpoint**. No separate Python, Bun, or ripgrep installation is required to use `jg`. ## Install the agent skill — required for agent setup Installing the CLI alone does not teach your coding agent to use it. **Install the skill as well**, from the project where your agent works: ```sh jg skill ``` The installer detects your coding agents (Claude Code, Codex, OpenCode and others) and asks where to install. Add `--global` for a user-wide install, or `--yes` for unattended installation. The [skill](skills/jevgrep/SKILL.md) explains installation, invocation and the meaning of returned context. It leaves research and implementation decisions to the calling agent. The current repository skill checks for `jg` and installs the CLI if it is missing; authentication still needs your selected provider’s key. The skill installer itself does not configure credentials. `jg skill` installs the `jevgrep` skill from the repository's `main` branch using a pinned [skills CLI](https://github.com/vercel-labs/skills), with npm lifecycle scripts disabled. Skill installation needs Node 22.20+ and npm/npx plus network access for the installer. Rerun `jg skill` to get the current skill without waiting for a new npm release. ### Upgrade There is currently no `jg upgrade` command. Upgrade the CLI with npm: ```sh npm install -g @dzhng/jevgrep@latest jg --version ``` After updating the CLI, refresh the installed skill by rerunning `jg skill`. Updating the npm package does not overwrite skill files in your projects. The installer records the GitHub source and skill path in `skills-lock.json`; rerun `jg skill` on another machine or after changing your Node installation. See the [package guide](apps/cli/README.md) for authentication details. ## Start with a question, leave with source Use `jg` when you know the behavior you need to understand but not where it lives: ```sh jg "Where is authentication checked before a request reaches a handler?" . jg "How are database connections created, pooled, and closed?" ./src jg "Which tests cover retry behavior when a request times out?" . ``` Jevgrep explores the repository hierarchy and follows qualifying branches. It selects files using content previews, then identifies useful source units and surrounding context. It keeps qualifying file locations even when it cannot confidently return an excerpt; it does not force every search into a fixed top-two list. The summary and compact file list come first, followed by selected source with line references, then detailed declaration and call locations. Python, TypeScript/JavaScript, Go and Rust support declaration parsing, with extensionless Python and Node scripts recognised by their shebang; other text uses a fallback. The output is evidence for the agent to use, not a generated answer or a guarantee that every relevant file was found. [See a recorded output example](specs/done/jevgrep/assets/stdout-example.txt). When you already know an exact symbol or path, a direct read or `rg` search may be all you need. Jevgrep is most useful for questions that span unfamiliar files. ## What we measured **Same intelligence, ~30% lower coding-agent cost.** Both Jevgrep and the no-Jev baseline solved **8/10 tasks**. Full Sol cost fell from **$7.62 to $5.44**—a measured **28.6% reduction**, rounded to ~30%—including failed attempts and excluding Jev cost. ![Jevgrep workflow: about 30% lower coding-agent cost, with 8 of 10 tasks solved both with and without Jevgrep.](assets/how-it-works.png) This comparison uses ten tuned Python SWE-bench tasks, one frozen installed package and the exact public skill in this repository. It measures task success and cost, not a speed improvement or guaranteed savings on every repository. See the [results and methodology](evals/results/relevance-threshold-2026-09-27.md) for per-task costs, artifact identities and limitations. A separate [speed study](evals/results/speed-2026-09-28.md) measures the follow-up local optimizations with Jev’s native TypeSafe endpoint. The [0.4.3 total-cost rerun](evals/results/total-cost-2026-09-28.md), including Jev, measured **25.8% lower total cost with the same 8/10 tasks solved**. The older ~30% graphic above reports Sol-only cost. Future benchmark totals include Jev. The [0.5.0 evaluation](evals/results/combined-cost-research-2026-09-28.md) retained 8/10 solves while reducing native Jev cost by about 59% versus that 0.4.3 run. Combined Sol-plus-Jev cost was 2–3% higher, accepted as a small tradeoff for this release. These single-run observations do not establish statistical equivalence or a speed improvement. Current protocols and subsequent experiments are in the [evaluation records](evals/README.md). ## Source, credentials, and local state Searches send eligible source content to Jev through the provider selected during auth. Default filesystem filtering respects ignore files and excludes hidden, dependency/build, binary, and obvious credential files, including `.env` variants and any path component whose name ends in `.env`. It also excludes a whole file as `sensitive_content` when its text matches a conservative credential heuristic: a known provider token prefix (such as `npm_` or `glpat-`), a Slack webhook URL, or a credential-shaped assignment with a long, high-entropy value. UUID-shaped identifiers are excluded from that entropy signal. This is best-effort detection, not redaction or a guarantee that all sensitive information has been removed; choose a search root you intend to send. Use `--include-sensitive` when a search must include sensitive names, private-key markers, and content-level matches. `jg files [root]` counts the files a search under that root may read, grouped by top-level directory, with no provider key or network request. It takes the same filtering flags as search. To skip paths inside that root for one search, pass `--exclude` with a gitignore pattern relative to the root, for example `--exclude '**/*.test.ts' --exclude 'src/generated/'`. `jg auth` asks for your provider, then saves its key in an owner-only config file. Re-running auth replaces that setup; searches always use the saved provider. `jg doctor` checks it with synthetic input. Existing saved keys without a provider remain Vercel keys. Environment-based credentials and endpoint overrides are not used; run `jg auth` if you previously relied on them. Evaluation answers are cached locally by default. The CLI writes its output to stdout and does not create report files. Use `jg --help` for cache controls, search overrides, and incomplete-result behavior. ## Development The repository uses TypeScript, Bun workspaces, and Turborepo. From a checkout: ```sh bun install --frozen-lockfile bun run dev --help bun run verify ``` Verification includes Docker tests of the installed Node-only package. For the reasoning behind retrieval, parsing, caching, and failure handling, start with the [architecture](docs/architecture.md) and [implementation record](specs/done/jevgrep/README.md). [Release guidance](scripts/RELEASING.md) covers tag-triggered npm publication and verification of the exact public package. See [Contributing](CONTRIBUTING.md) for the pull-request review and maintenance policy. [MIT](LICENSE). [Artwork and generation prompts](assets/README.md). ## Ratchet Files Source files are limited to 8,000 estimated tokens (bytes divided by four). The pre-commit hook runs `scripts/check-file-size.mjs` through lint-staged, so a clone must install the repository dependencies before commits. Existing oversized authored files are frozen at their current size; generated and vendored paths are excluded through `.gitattributes`. A file may use `@ratchet-files-ignore` in its first 10 lines only when splitting it would cost more than it saves, with the reason beside the pragma.

正在加载仓库 README…

这个仓库中的 skills

职业分类待补全

用 AI 助手安装

复制这段 Prompt，发给你正在使用的 AI 助手。

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/dzhng/jevgrep.

dzhng/jevgrep

仓库 README

相关分享

这个仓库中的 skills

用 AI 助手安装

仓库 README