CLITrigger
CLAUDE.md14 starsChanged 7 months agoMITView on GitHub
CLITrigger is a full-stack app that automates AI-powered task execution. Users write TODO items in a web UI, and the system spawns isolated git worktrees for each task, running Claude/Antigravity/Codex CLI tools in parallel. Built with Express + React + SQLite + WebSocket. These encode constraints that aren't obvious from the code. Apply when touching related areas. 시각 디자인 결정은 .claude/docs/design.md를 기준으로 참조할 것. Tooltips, dropdowns, popovers, context menus, and "more" menus MUST use createPortal(..., document.body) with position: fixed +…
- Reads credentials
In this repository8 other files it gives its agents.
- lessonSkill
- manage-skillsSkill
- merge-worktreeSkill
- releaseSkill
- remove-worktreeSkill
- update-docsSkill
- update-wikiSkill
- verify-implementationSkill
The file
We split this CLAUDE.md into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 80
Use grep/glob to find relevant files FIRST.
- Distinctive: line 81
Only read files you intend to modify or that are directly needed.
- Distinctive: line 82
Do NOT launch Agent/Explore subagents for simple tasks.
- Distinctive: line 88
Do not perform additional refactoring, optimization, or testing beyond what was requested.
- Distinctive: line 89
Do not add comments, docstrings, or type annotations to unchanged code.
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Project Overview
CLITrigger is a full-stack app that automates AI-powered task execution. Users write TODO items in a web UI, and the system spawns isolated git worktrees for each task, running Claude/Antigravity/Codex CLI tools in parallel. Built with Express + React + SQLite + WebSocket.
Commands
# Development (runs server + client concurrently)
npm run dev
# Build
npm run build # both client and server
npm run build:server # server only (→ dist/server/)
npm run build:client # client only (→ src/client/dist/)
# Tests
npm run test # all tests (server + client)
npm run test:server # server tests only (vitest, node env)
npm run test:client # client tests only (vitest, jsdom env)
npx vitest run src/server/path/to/file.test.ts # single server test
cd src/client && npx vitest run src/path/to/file.test.tsx # single client test
# Type checking
npm run typecheck # server + client
# Desktop (Windows)
scripts/build-win.bat # --skip-install/-s skips npm ci; --msix for appx targetArchitecture
Monorepo Layout
src/server/— Express backend (TypeScript, ESM). Entry:src/server/index.ts. DB: SQLite (better-sqlite3, WAL) insrc/server/db/. Services insrc/server/services/, routes insrc/server/routes/, plugins insrc/server/plugins/.src/client/— React frontend (Vite + TailwindCSS). Separatepackage.json. Entry:src/client/src/main.tsx→App.tsx.bin/clitrigger.js— npm CLI entry.plugin/— Hecaton TUI plugin (CommonJS, Deno-compatible).
Key Patterns & Gotchas
These encode constraints that aren't obvious from the code. Apply when touching related areas.
- Sandbox Mode — Claude absolute path gotcha:
.claude/settings.jsonpatterns MUST use absolute paths (${workDir}/**) — Claude resolves relative paths internally and the match fails. Also normalizeworkDirto forward slashes (workDir.replace(/\\/g, '/')) on Windows — mixed separators silently reject every Edit/Write match. - Wiki naming convention: UI was renamed Memory→Wiki but DB tables, routes (
/api/projects/:id/memory/*), files (MemoryList.tsx,memory-injector.ts), and the<long_term_memory>prompt tag intentionally retain "memory" — don't rename them. The XML tag is a semantic hint to LLMs. - Floating Window state gating:
SessionWindowsHostpersistsOpenGroup[]tosessionGroups:{projectId}localStorage. Persist must be gated onsessionshaving loaded so empty-during-load doesn't nuke restored state. Hydrate inuseStateinitializer (synchronous), not a post-mount effect — a persist effect can race and write[]first. - Raw Binary Streaming: DB
session_raw_chunksare the single source of truth for replay —session:subscribeflushes the in-flight buffer then sends only DB chunks. Never replay from both ring buffer + DB (causes duplicate output). - Native Focus Handoff (Electron): React
element.focus()doesn't reclaim native HWND focus from xterm.js. Use the IPC bridge:electronAPI.imeReset()→mainWindow.webContents.focus()in main process, then RAF twice before focusing the target.
UI Guidelines
시각 디자인 결정은
.claude/docs/design.md를 기준으로 참조할 것.
Floating elements must render via portal
Tooltips, dropdowns, popovers, context menus, and "more" menus MUST use createPortal(..., document.body) with position: fixed + viewport clamping. Do NOT use position: absolute — parent containers have overflow/transform that clip.
Checklist:
createPortalintodocument.bodyposition: fixed, top/left from anchor'sgetBoundingClientRect()- Clamp within viewport (flip/shift), ≥8px from edges
- Recompute on
scroll(capture) andresize z-tooltipz-index class
Environment
Config via .env (see .env.example), ~/.clitrigger/config.json, or Electron userData/config.json.
Key vars: PORT (default 3000), DB_PATH, TUNNEL_ENABLED, HEADLESS, DISABLE_AUTH.
Language
UI and documentation are primarily in Korean. Commit messages: Korean or English. Codebase (identifiers, comments): English.
Task Execution Guidelines
Efficiency
- Use grep/glob to find relevant files FIRST. Do NOT read files one by one to explore.
- Only read files you intend to modify or that are directly needed.
- Do NOT launch Agent/Explore subagents for simple tasks. Use direct grep → read → edit.
- Make all related edits in a single pass. Prefer
replace_all: truefor repetitive changes. - Aim for under 15 tool calls for simple tasks, under 30 for complex ones.
Completion
- Once done, commit and stop immediately.
- Do not perform additional refactoring, optimization, or testing beyond what was requested.
- Do not add comments, docstrings, or type annotations to unchanged code.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

