Tired of Cursor cutting off context, missing your files and folders, and spitting out empty responses?
Save your context with Shotgun! → Prepare a truly GIGANTIC prompt → Paste it into Google AI Studio and receive a massive patch for your code. 25 free queries per day! → Drop that patch into Cursor or Windsurf and apply the entire diff in a single request.
That means you get 25 huge, fully coherent patches per day for your codebase—absolutely free, thanks to complete context transfer.
Perfect for dynamically-typed languages:
Python JavaScript
One‑click codebase "blast" for Large‑Language‑Model workflows.
Shotgun is a tiny desktop tool that explodes an entire project into a single, well‑structured text payload designed for AI assistants. Think of it as a rapid‑fire alternative to copy‑pasting dozens of files by hand:
- Select a folder → get an instant tree + file dump
in a predictable delimiter format (
*#*#*...*#*#*begin … *#*#*end*#*#*). - Tick check‑boxes to exclude noise (logs, build artifacts,
node_modules, …). - Paste the result into ChatGPT, Gemini 2.5, Cursor, etc. to ask for multi‑file edits, refactors, bug fixes, reviews, or documentation.
- Receive a diff‑style reply and apply changes with your favourite patch tool.
Shotgun trades surgical, single‑file prompts for a "whole‑repository blast" – hence the name.
| Scenario | Pain Point | Shotgun Benefit |
|---|---|---|
| Bulk bug fixing | "Please fix X across 12 files." | Generates a complete snapshot so the LLM sees all usages. |
| Large‑scale refactor | IDE refactors miss edge cases. | LLM gets full context and returns a patch set. |
| On‑boarding review | New joiner must understand legacy code. | Produce a single, searchable text file to discuss in chat. |
| Doc generation | Want docs/tests for every exported symbol. | LLM can iterate over full source without extra API calls. |
| Cursor / CodePilot prompts | Tools accept pasted context but no filesystem. | Shotgun bridges the gap. |
- ⚡ Fast tree scan (Go + Wails backend) – thousands of files in milliseconds.
- ✅ Interactive exclude list – skip folders, temporary files, or secrets.
- 📝 Deterministic delimiters – easy for LLMs to parse and for you to split.
- 🔄 Re‑generate anytime – tweak the excludes and hit Shotgun again.
- 🪶 Lightweight – no DB, no cloud; a single native executable plus a Vue UI.
- 🖥️ Cross‑platform – Windows, macOS, Linux.
(This describes the UI flow. The core GenerateShotgunOutput Go function remains the primary backend logic for creating the text payload based on the selected root and exclusions.)
- Step 1: Prepare Context
- User selects a project folder.
- The file tree is displayed in the
LeftSidebar. - User can mark files/folders for exclusion.
- The application automatically (or via a button) triggers context generation in Go (
GenerateShotgunOutput). - The resulting context (tree + file contents) is stored in
shotgunPromptContextand passed toCentralPanel.vue, which in turn makes it available toStep2GenerateDiff.vue.
- Step 2: Compose Prompt
Step2GenerateDiff.vueis shown.- It displays the
shotgunPromptContext(likely in a read-only textarea). - User types their instructions for the LLM into another textarea (the prompt).
- User clicks "Compose Prompt" (was "Generate Diff").
MainLayout.vue(simulates) sending theshotgunPromptContextand the user's prompt to an LLM.- (Simulated) LLM returns a diff, which is then displayed in the "Diff Viewer" section of
Step2GenerateDiff.vue.
- Step 3: Execute Prompt
Step3ExecuteDiff.vueis shown.- User clicks "Execute Prompt" (was "Execute Diff").
MainLayout.vue(simulates) the "execution" of this prompt/diff. This step is more conceptual in the current stubbed implementation but would represent running or applying the changes indicated by the LLM.- Logs appear in the step-specific console within
Step3ExecuteDiff.vueand/or theBottomConsole.vue.
- Step 4: Apply Patch
Step4ApplyPatch.vueis shown.- User interacts with a (currently stubbed) patch editor.
- User clicks "Apply Selected" or "Apply All & Finish".
MainLayout.vue(simulates) applying these patches.
- Go ≥ 1.20
go version - Node.js LTS
node -v - Wails CLI
go install github.com/wailsapp/wails/v2/cmd/wails@latest
git clone https://github.com/glebkudr/shotgun_code
cd shotgun_code
go mod tidy # backend deps
cd frontend
npm install # Vue deps
cd ..wails devHot‑reloads Vue; restart the command for Go code changes.
wails build # binaries land in build/bin/- Run
wails dev. The app window will open. - Step 1: Prepare Context
- Click "Select Project Folder" and choose your repository root.
- In the left pane (
LeftSidebar), expand folders and un-tick any items you wish to exclude from the context. - Click the "Prepare Project Context & Proceed" button (typically in
Step1CopyStructure.vueor a similar area for Step 1). - The generated context (project tree and file contents) will be prepared internally.
- Step 2: Compose Prompt
- The view will switch to Step 2 (
Step2GenerateDiff.vue). - The generated project context from Step 1 will be displayed (usually read-only).
- Enter your instructions for the LLM in the "Prompt Editor" textarea.
- Click "Compose Prompt".
- A (mock) diff will be generated and shown in the "Diff View 8000 er".
- The view will switch to Step 2 (
- Step 3: Execute Prompt
- The view will switch to Step 3 (
Step3ExecuteDiff.vue). - Click "Execute Prompt".
- (Mock) execution logs will appear in the console areas.
- The view will switch to Step 3 (
- Step 4: Apply Patch
- The view will switch to Step 4 (
Step4ApplyPatch.vue). - Interact with the (stubbed) patch editor.
- Click "Apply Selected" or "Apply All & Finish" to (simulate) completing the process.
- The view will switch to Step 4 (
- You can navigate between completed steps using the top
HorizontalStepperor theLeftSidebarstep list.
app/
├── main.go
├── app.go
└── frontend/
├── App.vue
└── components/
└── FileTree.vue (example)
<file path="main.go">
package main
...
</file>
<file path="frontend/components/FileTree.vue">
<template>
...
</template>
</file>
- Tree View – quick visual map for you & the LLM.
- XML-like File Blocks – ... for easy parsing by models.
- Trim the noise – exclude lock files, vendored libs, generated assets. Less tokens → cheaper & more accurate completions.
- Ask for diffs, not whole files – keeps responses concise.
- Iterate – generate → ask → patch → re‑generate if needed.
- Watch token limits – even million‑token models have practical caps. Use Shotgun scopes (root folder vs subfolder) to stay under budget.
| Symptom | Fix |
|---|---|
wails: command not found |
Ensure $GOROOT/bin or $HOME/go/bin is on PATH. |
Blank window on wails dev |
Check Node version & reinstall frontend deps. |
| Output too large | Split Shotgun runs by subdirectory; or exclude binaries/tests. |
-
✅ Step 1: Prepare Context
Basic ability to select a project, exclude items, and generate a structured text context. -
✅ Step 2: Compose Prompt
- ✅ Watchman to hot-reload TreeView
- ✅ Custom rules
-
☐ Step 3: Execute Prompt
"Executing" the prompt and showing logs. -
☐ Step 4: Apply Patch
Enable applying patches inside Shotgun.- ☐ Direct API bridge to send output to OpenAI / Gemini without copy-paste
- ☐ CLI version for headless pipelines
- Watch token limits – even million-token models have practical caps. Use Shotgun scopes (root folder vs subfolder) to stay under budget.
PRs and issues are welcome!
Please format Go code with go fmt and follow Vue 3 style guidelines.
Custom MIT-like – see LICENSE.md file.
Shotgun – load, aim, blast your code straight into the mind of an LLM. Iterate faster. Ship better.