Retulo 0.5.0rc1 — one-command install and agent setup

Retulo 0.5.0rc1

Install Retulo without keeping a source checkout or manually creating a virtual environment:

curl -fsSL https://thestack-ai.github.io/retulo/install.sh | sh
retulo setup codex --project /absolute/path/to/project --apply
# Or: retulo setup claude --project /absolute/path/to/project --apply

If the installer reports that the executable directory is missing from PATH, use its printed export command first. It does not edit shell profiles.

• A checksum-verified release wheel runs in a persistent uv tool environment. Missing uv is installed privately; uv can obtain Python automatically. This release is distributed on GitHub, not npm or PyPI.
• The wheel includes Codex and Claude Code integration templates, guarded setup, diagnosis and disconnection. Agent settings use the installed runtime, independent of a source directory.
• Setup and disconnect default to previews; --apply writes changes. Existing unrelated settings and core workspaces are preserved. Claude setup manages project .mcp.json and a marked CLAUDE.md block; Codex setup manages the user-level MCP connection, skill, guidance and context hooks.
• retulo doctor codex|claude --project /absolute/path/to/project --connect checks the configured local MCP server without an AI model call.
• Isolated Python execution prevents project files and PYTHONPATH from shadowing Retulo. Unowned entries, edited components and incomplete installation records fail safely.
• English, Simplified Chinese and Korean READMEs and the documentation site describe the same packaged workflow. See the installation guide for custom paths and development installation.

Qualification includes regression checks, source/distribution inspection, an independent installer review, fresh installed-wheel task/version/restore checks, both SDK MCP connections, persistent bootstrap/reinstall, Codex CLI configuration parsing and Claude Code CLI health checking. Client trust and a complete agent conversation adopting the workflow remain separate checks. No AI model calls were made for these integration probes. Windows and network filesystems are not qualified by this release.

Assets include the universal Python wheel, source distribution, install.sh and SHA256SUMS. For downloaded assets, verify with shasum -a 256 -c SHA256SUMS on macOS or sha256sum -c SHA256SUMS on Linux.