Skip to content

CLAUDE.md

Latest commit

 

History

History
70 lines (53 loc) · 3.1 KB

CLAUDE.md

File metadata and controls

70 lines (53 loc) · 3.1 KB
Raw

CLAUDE.md — Automatic Context for Claude Code

This file is read automatically at the start of every Claude Code session. It exists because conversation history does NOT persist between sessions.

What This Repo Is

t3code-mobile is an unofficial Android and iPhone companion app + optional HTTPS/PWA proxy for T3 Code. Three deliverables: native Android WebView wrapper (apk/), iPhone PWA via proxy (IPHONE-GUIDE.md), and HTTPS proxy (server.js). Built by ChatGPT 5.4, reviewed by Claude Opus.

Current State

  • Branch: main
  • Version: 1.1.0 (in package.json)
  • GitHub: https://github.com/JSvandijk/t3code-mobile
  • Owner: Jesse van Dijk (JSvandijk)

Key Files (read these first in a new session)

  1. README.md — public-facing overview
  2. docs/CHAT-CONTEXT.md — operational handoff with rules and failure modes
  3. package.json — scripts, deps, version
  4. server.js — HTTPS/HTTP proxy (the main Node.js code, supports --http flag)
  5. IPHONE-GUIDE.md — iPhone PWA setup guide (from scratch)
  6. apk/app/src/main/java/com/t3code/app/MainActivity.java — Android app (single activity)
  7. sw.js — service worker
  8. webview-harness.js — local test harness
  9. smoke-proxy.js — proxy smoke test
  10. check-release.js — release validation

Useful Commands

npm test                          # full test suite (syntax + lint + manifest + repo checks + unit tests + proxy smoke)
npm run test:units                # unit tests only (lib/html.js)
npm run harness:http              # local WebView test server (HTTP)
npm run harness:https-bad-cert    # local WebView test server (bad cert)
npm run harness:redirect          # local WebView test server (redirect)
npm run start:iphone              # PWA proxy in HTTP mode for iPhone

Windows APK build: build-apk.bat iPhone PWA: npm run start:iphone + tailscale serve https / http://localhost:3780

Working Rules

  • GitHub is the real end state. Work is not done until committed and pushed.
  • Do not treat documentation as proof. Prefer executable checks and runtime evidence.
  • Always do an adversarial pass after implementation (info disclosure, escaping, overbroad config, deprecated APIs).
  • Read code, docs, and audits against each other. Mismatches are bugs.
  • Remove dead code. Do not let old and new implementations coexist without explanation.
  • Evidence must match the exact commit being represented.
  • A self-audit can miss issues introduced by the same authoring pass. Be explicit about this limitation.

Known Failure Modes (from prior sessions)

  • Same-pass contradiction: code, docs, and self-audit all wrong together
  • Documentation-as-proof fallacy: strong README != strong tests
  • Add-more-than-delete bias: dead files remain across iterations
  • Evidence drift: evidence from one SHA misrepresented for a later SHA
  • Security theater: document exists but code doesn't implement it

Context About This Project

This project was built primarily by ChatGPT 5.4 and reviewed by Claude Opus. The owner (Jesse) uses this as a comparative evaluation of AI code quality. Be honest, strict, and accurate in all assessments.

@AGENTS.md