Skip to content

CLAUDE.md

Latest commit

 

History

History
172 lines (128 loc) · 5.43 KB

CLAUDE.md

File metadata and controls

172 lines (128 loc) · 5.43 KB

TizenClaw — Claude Code Project Rules

This file is the authoritative guide for Claude Code operating in this repository. All rules here are binding.

Language Rule: Always respond in the same language as the user's input. Korean input → Korean response. English input → English response.

Project Overview

TizenClaw is a Rust-based Autonomous AI Agent daemon for Tizen OS (embedded Linux) and Ubuntu/WSL host development. The default workflow uses ./deploy_host.sh; the Tizen GBS workflow uses ./deploy.sh when explicitly requested. The project is a Cargo workspace with 9 crates under src/.

Active branch: develRust
Target device: emulator-26101 (x86_64) — auto-detected via sdb

Absolute Rules (Never Violate)

Build & Test

  • NEVER run cargo build, cargo check, cargo test, or cargo clippy directly. Default development builds/tests must go through ./deploy_host.sh.
  • NEVER run cmake . or any local CMake build.
  • Use ./deploy.sh only when the user explicitly asks for Tizen, emulator, or device validation.
  • Architecture focus: x86_64 only.

Commits

  • NEVER use git commit -m "...". Write the message to .tmp/commit_msg.txt first, then run git commit -F .tmp/commit_msg.txt.
  • Commit messages must be in English.
  • Title: ≤ 50 characters, imperative sentence, capitalized.
  • Body: each line ≤ 80 characters. No feat:, fix: prefixes. No explicit Why: / What: headers.
  • Push target: git push origin develRust.

Temporary Files

  • Use .tmp/ (project root) for all temporary files — not /tmp/.
  • Delete temp files when no longer needed.
  • .tmp/ is in .gitignore and must never be committed.

Local Build Cleanup

If a local build is accidentally triggered, immediately clean up:

rm -rf target/
rm -f CMakeCache.txt Makefile cmake_install.cmake
rm -rf CMakeFiles/ build_local/
find ./src -name 'CMakeFiles' -type d -exec rm -rf {} + 2>/dev/null
find ./src \( -name '*.o' -o -name '*.d' \) -delete 2>/dev/null

Development Cycle (6 Stages)

All tasks must follow these stages sequentially. Skipping is forbidden.

1. Planning → 2. Design → 3. Development →
4. Build/Deploy → 5. Test/Review → 6. Commit

Each stage has a corresponding skill in .agent/skills/. After each stage, update .dev_note/DASHBOARD.md with the stage status.

Stage Skill Key Output
1. Planning .agent/skills/planning-project/SKILL.md Module objectives, execution mode classification
2. Design .agent/skills/designing-architecture/SKILL.md FFI boundaries, async topology docs
3. Development .agent/skills/developing-code/SKILL.md TDD Red→Green→Refactor cycle
4. Build/Deploy .agent/skills/building-deploying/SKILL.md ./deploy_host.sh succeeded, or explicit ./deploy.sh -a x86_64
5. Test/Review .agent/skills/reviewing-code/SKILL.md Host or device logs as evidence
6. Commit .agent/skills/managing-versions/SKILL.md Clean commit via .tmp/commit_msg.txt

Deploy Commands

Tizen (GBS build → emulator/device)

# Full build + deploy (auto-detect arch)
./deploy.sh -d emulator-26101

# Fast rebuild (skip GBS init)
./deploy.sh -d emulator-26101 -n

# Fastest incremental rebuild
./deploy.sh -d emulator-26101 -n -i

# Skip build, deploy existing RPM only
./deploy.sh -d emulator-26101 -s

# Build only, no deploy
./deploy.sh -d emulator-26101 -S

Host Linux (default Ubuntu/WSL workflow)

# Release build + install + run (Generic Linux mode)
./deploy_host.sh

# Debug build
./deploy_host.sh -d

# Build only (no install/run)
./deploy_host.sh -b

# Run tests (offline, vendored)
./deploy_host.sh --test

# Daemon management
./deploy_host.sh --status   # check status
./deploy_host.sh --log      # follow logs
./deploy_host.sh --stop     # stop daemon

Code Quality Rules

  • Zero build warnings: All compiler warnings must be resolved at the code level. Do not suppress with #![allow(...)] except for C bindgen FFI.
  • No .unwrap() in production paths: Use proper error propagation.
  • Minimal FFI: Core AGI logic must be pure Rust. FFI only where Tizen-specific hardware/API is unavoidable.
  • Dynamic loading: Tizen .so symbols must be loaded via libloading. The daemon must never panic if a native library is absent — always fall back gracefully.
  • Send + Sync on all async types: Explicitly declare ownership bounds.

Workspace Crates

Crate Role
tizenclaw Main daemon — AgentCore, PromptBuilder, IPC
tizenclaw-cli User-facing CLI (stdio → IPC)
tizenclaw-tool-executor Secure native tool runner daemon
libtizenclaw C-ABI bridge for legacy C/C++ callers
libtizenclaw-core Core shared library
tizenclaw-metadata-* Plugin metadata crates

Operation Log

All stage progress and Supervisor audit records are tracked in: .dev_note/DASHBOARD.md

Supervisor Gate

After each stage, the Supervisor validates compliance before authorizing the next stage. See .agent/skills/supervising-workflow/SKILL.md.

Rollback is triggered if:

  • Local cargo commands were used
  • The wrong script path was used for the cycle
  • Commit used inline -m flag
  • No host/device logs provided in Test/Review stage
  • Extraneous build artifacts are staged for commit

Maximum 3 retries per stage gate before escalating to the user.