Skip to content
Back to search

nexpath

SkillWorks with Claude CodeGitHub

Local-first AI coding workflow for vibe coders, indie hackers, technical founders and product managers — catch missing tests and safety checks across Claude Code, Cursor, Windsurf, Replit, Lovable and Bolt.new.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/hi0001234d/nexpath with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/hi0001234d/nexpath'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/hi0001234d-nexpath/badge.svg)](/listings/hi0001234d-nexpath)

HTML

<a href="/listings/hi0001234d-nexpath"><img src="/api/listings/hi0001234d-nexpath/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 245 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # Nexpath CLI — Build Fast. Ship with Confidence.
  2. [![License: Apache 2.0](https://img.shields.io/badge/license-Apache%202.0-blue.svg)](LICENSE) [![Built at MLH AI Hackfest 2026](https://img.shields.io/badge/MLH-AI%20Hackfest%202026-orange.svg)](https://devpost.com/software/nexpath) [![Follow on X](https://img.shields.io/badge/X-@Nex__path-black?style=flat&logo=x&logoColor=white)](https://x.com/Nex_path) [![Discord](https://img.shields.io/badge/Discord-Join-7289DA?style=flat&logo=discord&logoColor=white)](https://discord.gg/za6JPBHrkV) [![Extension](https://img.shields.io/badge/VS%20Code-Extension-007ACC?style=flat&logo=visual-studio-code&logoColor=white)](https://marketplace.visualstudio.com/items?itemName=nexpath.nexpath-vscode) [![npm version](https://img.shields.io/badge/npm-v0.1.58-cb3837.svg?style=flat)](https://www.npmjs.com/package/nexpath-cli) [![Chrome Web Store](https://img.shields.io/badge/Chrome-Web%20Store-4285F4?style=flat&logo=googlechrome&logoColor=white)](https://chromewebstore.google.com/detail/nexpath/gdkknhjgflkkeajbhalkaakohpoflolc) [![Firefox Add-ons](https://img.shields.io/badge/Firefox-Add--ons-FF7139?style=flat&logo=firefoxbrowser&logoColor=white)](https://addons.mozilla.org/en-US/firefox/addon/nexpath/) [![Demo](https://img.shields.io/badge/YouTube-Demo-red?style=flat&logo=youtube&logoColor=white)](https://youtu.be/pNejtPA5DPU?si=4vhclFLQxAGSRjin)
  3. > **A local-first behaviour guidance layer that helps builders catch missing checks before AI-generated work becomes shipped risk.**
  4. AI coding agents make it easy to move fast. Nexpath helps reduce the chance that speed quietly becomes an unverified bug, an unsafe change, an unmaintainable codebase, or hours of manual review later.
  5. ## Prompt Enhancement — A Practical Twist
  6. Within Nexpath's broader behaviour-guidance vision, Prompt Enhancement is the first feature we're introducing as a practical twist: it saves vibe coders from writing every prompt detail by hand while preserving their original intent. When a task needs more rigour, it can suggest missing development practices, verification, or confirmation steps — encouraging users to give a mature development flow the time it needs to finish.
  7. **Demo:** [Prompt Enhancement in action, on YouTube](https://youtu.be/pNejtPA5DPU)
  8. ---
  9. ## Why Nexpath Exists
  10. - AI can introduce a bug without making the risk obvious.
  11. - Fast generation can leave a codebase harder to maintain after every change.
  12. - Security, rollback, regression, and acceptance checks are easy to miss while momentum is high.
  13. - Reviewing every AI-generated decision manually can cost more time than the generation saved.
  14. Nexpath reduces these risks by noticing what the current workflow is missing and bringing the relevant check into the work at the right moment. It does not replace developer judgement or promise bug-free code; it helps make that judgement faster, more structured, and harder to skip.
  15. ---
  16. ## A Workflow, Not Another Tool
  17. 1. **Build with your AI coding agent** — Work in the tools and flow you already use.
  18. 2. **Review with Nexpath** — Let the Prompt Quality Layer surface missing verification, safety, scope, or maintenance steps while the context is still fresh.
  19. 3. **Ship confidently** — Move forward after the important checks are visible, reviewed, and included in the task.
  20. Nexpath is one focused step inside a better shipping workflow — not a new place where development has to happen.
  21. Built during AI Hackfest 2026 by MLH.
  22. ---
  23. ## The Prompt Quality Layer — How It Works
  24. 1. **Understand** — Nexpath reviews the prompt, the current development stage, and relevant workflow signals while preserving the complete original request.
  25. 2. **Structure** — It builds one editable prompt with the sections the task needs, such as scope, constraints, acceptance expectations, verification, or missing-practice guidance.
  26. 3. **Safeguard** — Higher-risk work can receive confirmation, safety, rollback, or evidence requirements. Complex work can also receive a sequence-aware breakdown when multiple prompts would be more effective.
  27. 4. **Review** — You inspect and edit the result before sending it, or return to the original prompt. Nexpath provides the quality layer; you keep the final decision.
  28. ---
  29. ## Benchmarked on SWE-bench Verified
  30. Prompt quality is easy to demonstrate and hard to prove. To check whether Nexpath actually changes
  31. what a coding agent produces, we ran it against a public benchmark where the answer is graded by
  32. each project's own test suite — not by a person, and not by another model.
  33. **Setup:** Claude Code (`sonnet`) attempted 40 tasks from SWE-bench Verified twice — once on its
  34. own, once with the Nexpath-reviewed prompt. The task, starting commit, container, model, time
  35. limit, and tools were identical in both runs. The prompt is the only thing that changed. Grading
  36. used the official SWE-bench harness.
  37. | | Claude Code | Claude Code + Nexpath |
  38. |---|---|---|
  39. | Tasks solved | 27/40 (67.5%) | **29/40 (72.5%)** |
  40. | Issue's own tests pass | 27/40 (67.5%) | 29/40 (72.5%) |
  41. | Ran the project's tests before submitting | 38/40 (95.0%) | 39/40 (97.5%) |
  42. **Paired outcomes:** solved by both — 27 · solved only with Nexpath — **2** · solved only without
  43. Nexpath — **0** · solved by neither — 11
  44. ### Full results
  45. Every number above traces to a file below. The full benchmark report (Nexpath_benchmark/README.md) has the method, the two tasks Nexpath won, and the eleven neither run could solve. These are the raw files behind it — all 40 tasks, both runs, and the list we locked before the first run.
  46. | | |
  47. |---|---|
  48. | Complete report | [Benchmark report](Nexpath_benchmark/README.md) |
  49. | Every task, both runs | [`final_per_task.csv`](Nexpath_benchmark/final_per_task.csv) |
  50. | All measurements | [`final_summary.json`](Nexpath_benchmark/final_summary.json) |
  51. | Task list and settings, locked before the run | [`strat40-manifest.json`](Nexpath_benchmark/strat40-manifest.json) |
  52. | Prompt checks run before starting | [`preflight-strat40.json`](Nexpath_benchmark/preflight-strat40.json) |
  53. | Raw data for the confirmation-line test | [before](Nexpath_benchmark/gate-experiment-BEFORE-fix.jsonl) · [after](Nexpath_benchmark/gate-experiment-AFTER-fix.jsonl) |
  54. ---
  55. ## How Nexpath Reduces Risk
  56. ### Fewer Silent Bugs and Missed Checks
  57. - Verification and test expectations can be added for debugging, maintenance, planning, review, and other tasks that need proof of completion.
  58. - Absence signals can surface missing regression checks, acceptance criteria, reproduction evidence, or project grounding.
  59. - Weak or unrelated signals are not used as filler; guidance must match the current task.
  60. The core interaction keeps your request and the added workflow guidance together:
  61. ### Safer High-Risk Changes
  62. - Sensitive actions can receive explicit confirmation, rollback, backup, or safety requirements before the reviewed prompt is used.
  63. - Double confirmation, cross-confirmation, and final confirmation are used only when the task's complexity or risk justifies them.
  64. - The quality layer does not silently expand authority, auto-send work, or treat an agent response as proof of completion.
  65. ### A More Maintainable Development Flow
  66. - Scope, constraints, acceptance expectations, affected surfaces, and behaviour-preservation needs stay visible instead of being lost in a fast request.
  67. - Complex work can be decomposed into a clear current task plus a compact, ordered sequence plan.
  68. ### Less Review Overhead, Without Losing Control
  69. - Your complete original request stays visible inside one editable, quality-reviewed prompt.
  70. - You can edit the result, use it, or return to the original request before the reviewed version is sent.
  71. ---
  72. ### Supported AI Coding Agents & Developer Tools
  73. Nexpath CLI is built for prompt capture across AI coding agents.
  74. | Agent | Status in v0.1.58 |
  75. |-------|-----------------|
  76. | [**Claude Code**](#add-nexpath-to-your-development-workflow--installation) | Fully supported — end-to-end tested |
  77. | [**Cursor**](https://marketplace.visualstudio.com/items?itemName=nexpath.nexpath-vscode) | Fully supported — end-to-end tested |
  78. | [**Windsurf**](https://marketplace.visualstudio.com/items?itemName=nexpath.nexpath-vscode) | Fully supported — end-to-end tested |
  79. | [**Replit**](https://chromewebstore.google.com/search/nexpath) | Fully supported — end-to-end tested |
  80. | [**Lovable**](https://chromewebstore.google.com/search/nexpath) | Fully supported — end-to-end tested |
  81. | [**Bolt.new**](https://chromewebstore.google.com/search/nexpath) | Fully supported — end-to-end tested |
  82. ---
  83. ## Add Nexpath to Your Development Workflow — Installation
  84. ```bash
  85. # Install
  86. npm install -g nexpath-cli
  87. # Register with your coding agent
  88. nexpath install
  89. # Verify
  90. nexpath --version
  91. ```
  92. Setup notes:
  93. - The npm package is **`nexpath-cli`**; the commands it installs are **`nexpath`** and
  94. `nexpath-serve`. Requires Node 18 or later.
  95. - Building from source instead? See the [contribution guide](CONTRIBUTING.md).
  96. - **Install asks how Nexpath should run — one credential, two ways to give it:** your own
  97. **OpenAI API key** ([get one](https://platform.openai.com/api-keys)), or a **Nexpath token**
  98. from a free account at [parseos.tech/nexpath](https://parseos.tech/nexpath/) if you would
  99. rather not create an OpenAI account. If both are configured, your OpenAI key is the one used.
  100. - During install you pick your project role (what kind of work you do) so Nexpath tailors its guidance to how you build.
  101. - Nexpath's core is Prompt Enhancement (PE), with Multi-Prompt Sequence (MPS) and Prompt Enhancement Feedback (PEF) — these work automatically as you code.
  102. ### Uninstalling
  103. ```bash
  104. # Remove the Nexpath CLI (package name: nexpath-cli, binary: nexpath)
  105. nexpath uninstall
  106. npm uninstall -g nexpath-cli
  107. npm unlink -g nexpath-cli # only if you installed via npm link
  108. # Verify it's gone
  109. npm list -g nexpath-cli
  110. which nexpath
  111. # Clear local data and caches
  112. rm -rf ~/.nexpath
  113. rm -rf ~/.config/nexpath
  114. rm -rf ~/.local/share/nexpath
  115. rm -rf ~/.cache/nexpath
  116. # Clear the npm cache
  117. npm cache clean --force
  118. ```
  119. `nexpath uninstall` disconnects Nexpath from all detected agents and offers to clear the stored
  120. API key. The remaining steps remove the global package, any leftover binary, and all local
  121. data and caches.
  122. ---
  123. ## Configuration and Privacy
  124. ### Privacy Controls
  125. All data is stored **locally only** at `~/.nexpath/`. Only targeted LLM calls used to classify or
  126. prepare relevant guidance leave your machine.
  127. - **Automatic secret redaction** — API keys (`sk-*`, `ghp_*`, `ghu_*`), bearer tokens, and
  128. PEM blocks are automatically stripped from prompts before storage.
  129. - **Telemetry is off unless you turn it on** — install does not ask and does not enable it;
  130. it starts off and stays off until you run `nexpath config set telemetry.enabled true`.
  131. Local prompt capture and remote telemetry are independent — disable capture anytime via
  132. `nexpath store disable` (if you do this, nothing will work).
  133. ### Deleting Stored Prompts
  134. ```bash
  135. # Delete prompts for a specific project you no longer want to keep
  136. nexpath store delete --project <path>
  137. # Delete all stored prompts permanently
  138. nexpath store delete -y
  139. ```
  140. ---
  141. ## Troubleshooting
  142. ### No Popup, Or A Suggestion That Looks Wrong?
  143. Run `nexpath status`. It reports which credential is in effect and the last LLM call that
  144. failed — for example a key that has run out of credit, been revoked, or been replaced. When a
  145. call fails Nexpath falls back to local detection and goes quiet, which looks like nothing
  146. happening rather than like an error.
  147. ### Where Is My API Key Stored?
  148. | Platform | Default location | Inspect with |
  149. |---|---|---|
  150. | macOS | Keychain | Keychain Access.app → search "nexpath" |
  151. | Linux | Secret Service (libsecret) | `secret-tool lookup service nexpath account openai_api_key` |
  152. | Windows | Credential Manager | Control Panel → Credential Manager → Web Credentials |
  153. | Fallback (any OS) | `~/.nexpath/config.json` (mode 0600) | `cat ~/.nexpath/config.json` |
  154. Use `nexpath config show-key-source` to confirm which layer is currently active.
  155. ---
  156. ## Contributing
  157. [Contribution guide](CONTRIBUTING.md)
  158. ---
  159. ## License
  160. [Apache License 2.0](LICENSE)
  161. ---
  162. ## Acknowledgements
  163. - **Major League Hacking (MLH)** — For organizing AI Hackfest 2026
  164. - **Anthropic** — For Claude Code, our primary development environment
  165. - **OpenAI** — For models used in targeted classification and prompt-quality tasks
  166. - **Google** — For Gemini AI, planned as an alternative LLM provider alongside OpenAI
  167. Built with insights from the vibe coding community and developers building real projects with AI coding agents, coding AI tools, and AI developer tools.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.