Skip to content

CHANGELOG.md

Latest commit

 

History

History
86 lines (74 loc) · 11.2 KB

CHANGELOG.md

File metadata and controls

86 lines (74 loc) · 11.2 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Added

  • Added herdr (terminal multiplexer for coding agents, https://herdr.dev) to sf-toolbox: GitHub-releases binary install on Arch (no AUR), Homebrew on Debian/Ubuntu, with automatic version management like rtk
  • sf-toolbox now installs and refreshes global OpenSpec skills and prompt commands (~/.claude/skills/openspec-*, ~/.claude/commands/opsx/*) for Claude during provisioning by running ajust sf-openspec-install-global claude, mirroring the macOS sparkdock provisioning; idempotent and skipped when the OpenSpec CLI is absent (sparkfabrik/sparkdock#521)
  • Enable the Claude Code gh skill gate automatically in the sf-toolbox role by running the sparkdock-installed claude-gh-gate.py enable (blocks raw gh commands until the gh skill is loaded), with a fallback warning to run ajust claude-gh-gate-enable manually if the sparkdock script is missing. Mirrors the existing caveman/rtk setup pattern and the macOS sparkdock provisioning.
  • Install spark-http-proxy on Linux via a git clone + symlink (under ~/.local/spark/http-proxy/src) instead of a standalone downloaded script, mirroring the macOS sparkdock mechanism. This makes spark-http-proxy self-update work, keeps the compose file in sync with the CLI, and deterministically updates to the latest main on every provision. Migrates existing installs by replacing the standalone CLI file with a symlink and removing the now-stale standalone compose.yml that would otherwise shadow the clone
  • Added http-proxy-install-update ajust recipe (group http-proxy) for parity with macOS sjust; updates spark-http-proxy via the CLI's own self-update (git-pulls the clone it was installed from), so it needs no provisioner checkout or config
  • Fixed the provision-tags ajust recipe default provisioner_path ($HOME/provisioner → /opt/archlinux-provisioner, where bootstrap.sh/install.linux install the provisioner) so provisioner-tag recipes find the checkout without manually setting AJUST_PROVISIONER_PATH
  • Configured *.loc local DNS resolution on all OSes (including Debian/Ubuntu) by invoking spark-http-proxy configure-dns during sf-toolbox provisioning
  • Added ripgrep (rg) to sf-toolbox packages for both Arch Linux (pacman) and Debian/Ubuntu (Homebrew)
  • Added language server binaries for Claude Code's official code-intelligence plugins to the sf-toolbox role so org-level enabledPlugins can wire them in without per-machine setup: intelephense, typescript, typescript-language-server, pyright via npm (covers php-lsp, typescript-lsp, pyright-lsp plugins) and gopls via pacman on Arch / Homebrew on Debian/Ubuntu (covers gopls-lsp plugin). LSP processes only spawn when matching file extensions are present in the workspace, so devs not working in a given language pay no runtime cost
  • Switched Claude Code from claude-code (stable) to claude-code@latest cask on Debian/Ubuntu, which tracks latest releases instead of pinned stable versions; includes migration task to auto-uninstall old cask
  • Added /usr/local/bin/sparkfabrik-claude-code-otel-headers — Claude Code OTLP otelHeadersHelper script, sourced from sparkdock (sparkfabrik/sparkdock#483).
  • Added automatic caveman configuration via sparkdock setup script (sjust/scripts/caveman/setup.sh) in sf-toolbox role, with fallback warning to run ajust sf-caveman-install manually
  • Added automatic rtk configuration via sparkdock setup script after installation, with fallback warning to run ajust sf-rtk-setup manually
  • Added rtk-ai (Rust Token Killer) installation to sf-toolbox role with automatic version management from GitHub releases
  • Added GitHub Copilot CLI as a managed sf-toolbox coding agent (npm on Arch, brew cask on Debian/Ubuntu)
  • Added Claude Code as a managed sf-toolbox coding agent (curl installer on Arch, brew cask on Debian/Ubuntu)
  • Added Homebrew cask support to sf-toolbox packages task for Debian/Ubuntu
  • Added --help / -h flag to sf-toolbox with colored output and OS detection display
  • Added playbooks/roles/sf-toolbox/ role with per-tool task files (packages, gcloud, ai, glab, ajust, http-proxy)
  • Added detect section in toolbox package definitions for binary detection in the installer
  • Added clean arch/debian separation in toolbox package definitions with dotted-path support in the parser

Changed

  • sf-toolbox now runs gcloud components update on every provisioning pass, so an existing Google Cloud SDK in /opt/google-cloud-sdk is kept current instead of staying pinned at the version installed on first provision
  • Delegated *.loc systemd-resolved configuration to the spark-http-proxy CLI and removed the bespoke docker role drop-in (docker-dev-dns.conf, 172.17.0.1:19322); the CLI now writes http-proxy.conf (127.0.0.1:19322). Legacy ~docker/dnsdock routing is no longer configured
  • Replaced the obsolete Python yq package (pacman) with go-yq, the mikefarah Go yq v4 (extra repo); the old yq is now removed first since the two packages conflict
  • Homebrew formulae, casks, and rtk now install with state: latest so packages upgrade on every run
  • Split rtk installation by OS: GitHub releases on Archlinux, Homebrew on Debian/Ubuntu
  • Moved github-cli (gh) installation from packages role to sf-toolbox role with Debian/Ubuntu support via Homebrew
  • Moved GitHub Copilot CLI installation from packages/tasks/development.yml to sf-toolbox role
  • Disabled Google Cloud SDK usage reporting both at install time (--usage-reporting false) and persistently via gcloud config set core/disable_usage_reporting true to avoid sending telemetry to Google
  • Moved opencode base configuration from ~/.config/opencode/opencode.json to /etc/opencode/opencode.json (user-owned, in a root:root directory) to support user-local overrides via ~/.config/opencode/opencode.json
  • Added automatic cleanup of duplicate ~/.config/opencode/opencode.json when identical to the shipped source, with a warning when the file contains non-custom content
  • Standardized all shell variable references in bin/install.linux to use curly braces syntax (${VAR})
  • Moved config/toolbox-packages.yml into playbooks/roles/sf-toolbox/vars/main.yml (auto-loaded by Ansible, no more include_vars)
  • Renamed playbooks/toolbox.yml to playbooks/sf-toolbox.yml for consistent naming with the sf-toolbox role and installer
  • Restructured toolbox package config from flat keys to nested arch/debian/common sections
  • Moved src/scripts/parse-packages.py to bin/common/parse-toolbox-packages.py with support for dotted key paths
  • Moved just/gum package installation from sparkdock role to sf-toolbox role
  • Moved ajust setup (wrapper, justfile, completion, recipes) from sparkdock role to sf-toolbox role
  • Stripped sparkdock role down to git clone + agent resources sync only
  • Simplified bin/install.linux — removed conflict detection/removal, removed plan presentation, added detect/inform status display
  • Removed invasive apt package and PPA source removal on Debian (Homebrew PATH precedence handles conflicts)
  • Updated system.yml to import sf-toolbox role (alongside existing roles)

Fixed

  • Fixed sf-toolbox Claude Code never upgrading on Arch Linux. The claude --version detection ran without ~/.local/bin on PATH, so it failed even when Claude Code was installed there, causing the official installer to re-run on every provision while a real upgrade never happened. The detection now includes ~/.local/bin on PATH, and when Claude Code is already present the role runs claude update, matching the state: latest upgrade behavior of the Debian/Ubuntu Homebrew cask path
  • Fixed sf-toolbox Homebrew package installs failing on Debian/Ubuntu with "Bubblewrap is installed but cannot create a rootless sandbox" when building source-only formulae (e.g. opencode). Homebrew 6.0 enables a Bubblewrap build sandbox on Linux by default, which fails on distros that restrict unprivileged user namespaces (Ubuntu 24.04+ AppArmor). The brew install tasks now set HOMEBREW_NO_SANDBOX_LINUX=1, and the managed Homebrew shellenv (/etc/profile.d/homebrew.sh, ~/.zshrc, ~/.bashrc) exports it so manual brew commands work too, without relaxing kernel or AppArmor hardening
  • Fixed sf-toolbox failing with "PyYAML is not installed" when a Homebrew python3 shadows the system interpreter on PATH; the installer now selects a python3 that has PyYAML (preferring /usr/bin/python3) for the toolbox package parser
  • Fixed sf-toolbox npm packages (e.g. @fission-ai/openspec) never upgrading after first install by switching the Arch and Debian/Ubuntu npm install tasks from state: present to state: latest, matching the existing Homebrew behavior
  • Fixed rtk not detected by sf-toolbox conflict system; added rtk to toolbox.detect list and is_sf_managed(), removed redundant per-role which -a detection that didn't feed into the conflict resolver
  • Fixed resolve_repo_dir in bin/install.linux always preferring /opt/archlinux-provisioner over local checkout, preventing local development testing
  • Fixed sf-toolbox OS detection to treat CachyOS as Archlinux
  • Fixed sf-toolbox symlink breaking SCRIPT_DIR resolution by resolving symlinks with readlink -f before computing the directory

Removed

  • Removed playbooks/roles/packages/tasks/ai.yml, glab.yml, gcloud.yml, homebrew.yml (merged into sf-toolbox)

  • Removed playbooks/roles/docker/tasks/sparkfabrik-http-proxy.yml (merged into sf-toolbox)

  • Removed playbooks/roles/sparkdock/tasks/packages-arch.yml and packages-debian.yml (merged into sf-toolbox)

  • Removed sparkdock zshrc sourcing (replaced by ajust shell integration)

  • Added GitHub Actions CI workflow testing the toolbox playbook on Ubuntu 24.04, Ubuntu 26.04, and Arch Linux

  • Added bin/install.linux single-command installer/updater (sf-toolbox) with detect→plan→confirm→execute flow, gum integration, and conflict removal on Debian

  • Added playbooks/toolbox.yml lightweight playbook for company tooling only (sparkdock, AI, glab, gcloud, http-proxy) without full system provisioning

  • Added config/toolbox-packages.yml as single source of truth for package definitions consumed by both shell and Ansible

  • Added src/scripts/parse-packages.py stdlib-only YAML parser for shell consumption

  • Added bin/common/logging.sh shared logging helpers (sourced from sparkdock) with gum integration and ANSI fallback

  • Added Debian/Ubuntu compatibility for sparkdock, AI tooling, and shell setup using Homebrew as the package manager for dev tools on Debian systems

  • Added Homebrew bootstrap task that auto-installs linuxbrew on Debian/Ubuntu systems

  • Added OS-specific task file split pattern (e.g., packages-arch.yml / packages-debian.yml) to keep platform logic separated

  • Added gum spin wrapper for headless sparkdock-agents-sync execution in non-interactive Ansible context

  • Added CHANGELOG.md following Keep a Changelog conventions