Skip to content
BranchesTags

Latest commit

Β 

History

290 Commits

Folders and files

NameName
Last commit message
Last commit date
.claude
.claude
Β 
Β 
docs
docs
Β 
Β 
editors
editors
Β 
Β 
fonts
fonts
Β 
Β 
git
git
Β 
Β 
herdr
herdr
Β 
Β 
mac
mac
Β 
Β 
shell
shell
Β 
Β 
.editorconfig
.editorconfig
Β 
Β 
.gitignore
.gitignore
Β 
Β 
README.md
README.md
Β 
Β 

Repository files navigation

bertini36/dotfiles ο£Ώ

My personal Mac setup and configurations

πŸš€ Setup

  • Download code:

    git clone https://github.com/bertini36/dotfiles.git ~/.dotfiles/

  • Brew packages installation:

    brew bundle --file=mac/Brewfile
    Package Description
    bat cat with syntax highlighting
    eza Modern ls replacement
    fzf Fuzzy finder for the terminal
    gh GitHub CLI
    pre-commit Git hook manager
    graphviz Graph visualization tools
    jq JSON processor
    libmagic File type detection library
    gotop Terminal system monitor
    copilot-cli GitHub Copilot CLI (cask)
    mole macOS disk space cleaner and system optimizer
    postgresql@18 PostgreSQL database
    pyenv Python version manager
    uv Fast Python package manager
    [email protected] Python interpreter
    tldr Simplified man pages with practical examples
    karabiner-elements Keyboard remapper (cask)
    fd Fast find replacement
    ripgrep Fast grep replacement
    semgrep Static analysis (SAST) scanner
    gitleaks Secret detection in git commits
    nvm Node version manager
    pnpm Fast Node package manager
    claude Anthropic Claude desktop app (cask)
    claude-code Anthropic Claude CLI (cask)
    granola AI meeting notepad that captures and summarizes meetings (cask)
    rtk CLI proxy that reduces LLM token consumption by 60-90%
    herdr Terminal multiplexer that keeps coding agents running in panes
    handy Speech-to-text utility

  • Extra configuration (not available through Brew):

    bash mac/config_extras.sh
    Config Description
    gitleaks hook Global git pre-commit hook for secret detection

  • Add fonts (fonts/) to Font Book

  • Configure Karabiner

    • Change Caps Lock to CMD + CTL + Option + Shift
    • Map F4 to CMD + Space (Raycast)

  • Install Oh My ZSH

    • Source shell/.zshrc from ~/.zshrc so installer appends and machine-specific aliases stay out of the repo:
    echo 'source ~/.dotfiles/shell/.zshrc' > ~/.zshrc

    • Install plugins

      git clone https://github.com/zsh-users/zsh-syntax-highlighting.git ${ZSH_CUSTOM:-~/.oh-my-zsh/custom}/plugins/zsh-syntax-highlighting
git clone https://github.com/zsh-users/zsh-autosuggestions ${ZSH_CUSTOM:-~/.oh-my-zsh/custom}/plugins/zsh-autosuggestions
git clone https://github.com/agkozak/zsh-z ${ZSH_CUSTOM:-~/.oh-my-zsh/custom}/plugins/zsh-z

  • Install Chrome

  • Install Youtube Music (as browser pwa)

  • Install WhatsApp

  • Install Telegram

  • Install Slack

  • Install Claude

  • Install Notion

  • Install Notion Calendar

  • Install Obsidian

  • Install Jetbrains Toolbox and Pycharm

  • Install Visual Studio Code

  • Install Iterm2

  • Install Docker

  • Install Raycast

    • Disable Spotlight shortcut to enable Raycast one (System Preferences -> Keyboard -> Shortcuts -> Spotlight -> Uncheck Show Spotlight search)
    • Configure shortcuts following keymap.md

  • Install Amphetamine and set it to keep the computer awake indefinitely

  • Grant Screen Recording permission to your terminal app (System Settings -> Privacy & Security -> Screen Recording): screencapture needs it, and Claude Code uses screencapture to screenshot visual changes for a PR (see the create-pull-request skill)

  • Enable auto-focus: defaults write com.apple.Terminal FocusFollowsMouse -bool true

  • Install reviewr, the herdr plugin that reviews an agent's diff in a pane beside it and sends the line comments back

    herdr plugin install persiyanov/herdr-reviewr

  • Link the rest of configuration files (install Claude Code first so ~/.claude/ exists)

    ln -s ~/.dotfiles/git/.gitignore_global ~/.gitignore_global
git config --global core.excludesfile ~/.gitignore_global

ln -s ~/.dotfiles/editors/vim/.vimrc ~/.vimrc

mkdir -p ~/.config/herdr
ln -s ~/.dotfiles/herdr/config.toml ~/.config/herdr/config.toml
herdr config check   # must print 'config: ok'

mkdir -p ~/.config/herdr/plugins/config/persiyanov.reviewr
ln -s ~/.dotfiles/herdr/reviewr.toml ~/.config/herdr/plugins/config/persiyanov.reviewr/config.toml

ln -s ~/.dotfiles/.claude/settings.json ~/.claude/settings.json
ln -s ~/.dotfiles/.claude/statusline-command.sh ~/.claude/statusline-command.sh
ln -s ~/.dotfiles/.claude/CLAUDE.md ~/.claude/CLAUDE.md
ln -s ~/.dotfiles/.claude/RTK.md ~/.claude/RTK.md
ln -s ~/.dotfiles/.claude/skills ~/.claude/skills
ln -s ~/.dotfiles/.claude/rules ~/.claude/rules
ln -s ~/.dotfiles/.claude/agents ~/.claude/agents
ln -s ~/.dotfiles/.claude/hooks ~/.claude/hooks

🧠 Claude Configuration

All Claude Code configuration lives under .claude/ and is symlinked into ~/.claude/.

Note

Remote control is enabled via remoteControlAtStartup in the global settings. Each session auto-starts the bridge, so it can be driven from claude.ai/code or the Claude mobile app.

Workflow

Spec-driven development, opt-in. /start-feature "<task>" runs the full pipeline: a spec, a grilled plan, and a verified implementation, worth it for work that changes architecture, persistent data, or a public contract. Everything else stays a direct conversation, no ceremony paid unless it is asked for explicitly. The superpowers plugin supplies the stages; this repo supplies grilling and fix-until-green around them. Full walkthrough in .claude/skills/start-feature/SKILL.md.

/start-feature "<task>" walks the pipeline, stopping only where it needs you, and always ends on the same Verify β†’ Review β†’ PR β†’ Address feedback β†’ Finish tail.

πŸ™‹ orange waits for you Β· πŸ€– gray runs alone Β· ❓ blue runs alone but can interrupt Β· πŸ“„ green are the documents it writes.

flowchart TD
    K["πŸ™‹ 0 Β· Kick off<br><b>/start-feature</b><br><i>you describe the task, plus the Jira ticket</i>"]
    W["πŸ™‹ 1 Β· Branch or worktree<br><b>superpowers:using-git-worktrees</b><br><i>the hook provisions the virtualenv<br>in Python worktrees</i>"]
    B["πŸ™‹ 2 Β· Brainstorm<br><b>superpowers:brainstorming</b><br><i>you answer questions one at a time,<br>then approve the design section by section</i>"]
    SPEC[/"πŸ“„ SPEC Β· what and why<br>docs/superpowers/specs/&lt;date&gt;-&lt;topic&gt;-design.md<br><i>you review the file</i>"/]
    P["πŸ™‹ 3 Β· Plan<br><b>superpowers:writing-plans</b>"]
    PLAN[/"πŸ“„ PLAN Β· how, task by task<br>docs/superpowers/plans/&lt;date&gt;-&lt;feature&gt;.md<br><i>you read it</i>"/]
    G["πŸ™‹ 3 Β· Grill<br><b>grilling</b><br><i>you answer a round at a time until<br>no decision in the plan is fuzzy,<br>then you confirm</i>"]
    I["❓ 4 Β· Implement<br><b>superpowers:executing-plans</b> Β· in session<br><b>superpowers:test-driven-development</b> Β· every task<br><b>superpowers:dispatching-parallel-agents</b> Β· fan-out only<br><i>runs task to task without checking in</i>"]
    Y["πŸ€– 5 Β· Verify<br><b>superpowers:verification-before-completion</b><br>fix-until-green Β· on failing checks<br>superpowers:systematic-debugging Β· on surprises<br><i>you get the evidence: tests + pre-commit output</i>"]
    R["πŸ™‹ 6 Β· Review<br><b>code-reviewer</b> agent on the diff<br><i>plus /security-review when the change<br>touches auth, secrets, or user input</i>"]
    PR["πŸ™‹ 7 Β· PR<br><b>create-pull-request</b> skill<br>writing-clearly Β· superpowers:finishing-a-development-branch<br><i>you read the title and body before they go out</i>"]
    F["πŸ™‹ 8 Β· Feedback<br><b>pr-reviewer</b> agent<br><i>you paste the PR link; it closes threads,<br>and you answer human reviewers yourself</i>"]
    Z["πŸ™‹ 9 Β· Finish<br><i>you merge the PR, then switch to main, pull,<br>and remove the feature branch (and worktree)</i>"]

    K --> W --> B --> SPEC --> P
    P --> PLAN --> G
    G --> I
    I -- "stops only on a plan conflict,<br>an implementer question, or BLOCKED" --> Y
    Y --> R --> PR --> F --> Z

    classDef you fill:#FFE3C2,stroke:#C2410C,color:#111827
    classDef auto fill:#E5E7EB,stroke:#6B7280,color:#111827
    classDef ask fill:#DBEAFE,stroke:#1D4ED8,color:#111827
    classDef doc fill:#DCFCE7,stroke:#15803D,color:#111827
    class K,W,B,P,G,R,PR,F,Z you
    class Y auto
    class I ask
    class SPEC,PLAN doc
Loading

This pipeline only runs behind /start-feature. Nothing routes a task into it automatically; anything not started that way is a direct conversation, no spec, plan, or grill imposed.

Five rules never bend:

  • No fuzzy decisions. grilling interviews you in rounds until every decision is settled. It ends when you confirm the understanding is shared, not when it runs out of questions.
  • A plan names its harness. Every behavior the plan promises says which test fails when it breaks. A spec is only as solid as the thing that checks it.
  • Subagents fan out, they do not relay. A dependent chain stays in one session. Subagents take work that shares no state and no ordering: sweeping for references, auditing an area, getting oriented.
  • The diff is reviewed once. code-reviewer audits it at stage 6. pr-reviewer then owns the open threads, not a second pass over the same code.
  • Humans answer humans. pr-reviewer closes your threads and bot threads. Another person's thread stays yours, even when you asked for the fix.

Skills

Reusable AI agent skills that Claude invokes autonomously when a task matches their description. Any skill can also be invoked explicitly as a slash command (/skill-name).

The MCPs column lists the MCP servers a skill needs. Required servers must be connected before the skill runs, otherwise it stops. Optional servers only add enrichment, or cover one input type among several, and the skill degrades gracefully without them. See Per-project plugins and Installing and updating for how to connect each one.

Skill Description MCPs
create-pull-request Create a GitHub PR following project conventions using gh CLI, with a screenshot attached when the change has a visual surface None
django-patterns Django architecture, REST APIs with Pydantic, ORM best practices, caching, and signals None
domain-service-layer Where business logic lives in a Django app: domain services own the rules, views do HTTP, factories wire the dependencies None
explain Turn a link into a local HTML page that explains it visually, with diagrams built from pure CSS and inline SVG, then open it in Chrome via /explain Optional: atlassian for Jira and Confluence links, notion for Notion links. Other link types use WebFetch and gh
fix-until-green Loop project checks and pre-commit until green or 5 iterations, fixing each failure against its full output None
grilling Grill the user relentlessly about a plan, decision, or idea, interviewing in rounds until every decision is settled (source) None
herdr Drive herdr through its CLI: inspect panes, tabs and workspaces, split layout, start sibling agents and read their output (source) None
investigate-sentry Investigate a Sentry exception down to root cause and propose a fix Required: sentry. Optional: datadog-mcp to correlate the request behind the exception
memento Morning briefing from the previous working day's Granola meetings and Slack conversations: up to 5 importance-sorted points, action-flagged, with the review window resolved against Google Calendar Required: granola, slack, google-calendar (all three checked in a preflight gate)
python-code-style Python type safety, generics, protocols, and advanced type annotations None
save-session Save a high-density summary of the current session to .claude_sessions.md None
start-feature Start the feature development pipeline, via /start-feature only None
tldr Quick bullet summary of a URL (article, blog post, video) via /tldr None
writing-clearly Clear prose for docs, commits, error messages, and UI text None

Evals

Some skills carry an evals/evals.json file that defines test cases to measure skill effectiveness: create-pull-request, django-patterns, python-code-style, and writing-clearly. To run the evals, paste the following steps into your AI agent prompt.

  1. Read the eval definitions in .claude/skills/<skill>/evals/evals.json
  2. Generate outputs - run each eval prompt twice per skill (once with the skill loaded, once without) and save the results to .claude/skills-workspace/iteration-1/<eval-id>/with_skill/outputs/ and without_skill/outputs/
  3. Create eval_metadata.json - record the assertions from each eval's expectations array alongside references to the output files
  4. Compare outputs in with_skill/outputs/ vs without_skill/outputs/
  5. Verify each assertion from eval_metadata.json against the corresponding output

Agents

Specialized subagents that run in isolated context windows with restricted tools.

Agent Description
code-reviewer Read-only diff audit with an A-F graded report (architecture, security, performance, quality, testing) and explicit agent-authored red flags
pr-reviewer Closes out a PR's open review comments: fetches, triages, fixes, commits, pushes, replies, resolves threads, and verifies CI

Rules

Path-scoped rules that load automatically only when working on matching files.

Rule Scope
python **/*.py - Python 3.13+ conventions, ruff, uv, naming, imports
django Django files (views, models, urls, admin, etc.) - architecture plus N+1 prevention, batch writes and query-count guards
tests Test files - no comments, self-explanatory naming

Hooks

Shell scripts in .claude/hooks/ that the harness runs around tool calls and session events. Registered in the hooks block of .claude/settings.json.

Hook Event Description
python-worktree-venv.sh SessionStart Provisions the virtualenv when a session starts inside a Python git worktree
git-protected-branches.sh PreToolUse (Bash) Denies any push to main or master and any rewrite of their history (amend, rebase, moving reset, forced branch moves, filter-repo); stays silent for everything else

rtk hook claude also runs on every Bash call to rewrite commands through the rtk proxy.

πŸ”Œ Claude Plugins

Plugins are split into two tiers to keep session context lean: a small global set enabled for every session, and domain-specific plugins enabled only in the projects that need them. All marketplaces are registered globally in extraKnownMarketplaces of .claude/settings.json.

Global plugins

Enabled in enabledPlugins of the global settings:

Plugin Description
superpowers Spec driven development (SDD) based on brainstorming, planning, subagent-driven execution, TDD, and code review skills
skill-creator Create, modify, and benchmark custom skills, including eval runs and description optimization
context7 Up-to-date documentation and code examples for any library
caveman Caveman-speak mode that cuts ~75% of output tokens while keeping technical accuracy
last30days Research any topic across Reddit, X, YouTube, HN, Polymarket, and the web, scored by upvotes, likes, and real money

Per-project plugins

Enable these in the project's .claude/settings.json:

Plugin Description
sentry-skills Sentry engineering skills: PR writing, code review, Django patterns, security review, and more
notion Read and manage Notion pages and databases
figma Read Figma designs and generate code from them
atlassian Jira and Confluence: issues, backlogs, status reports, and knowledge search
datadog-mcp Datadog observability: logs, metrics, traces, incidents, monitors, and dashboards 
{
  "enabledPlugins": {
    "figma@claude-plugins-official": true,
    "sentry-skills@sentry-skills": true
  }
}

Installing and updating

On a fresh machine, install the third-party plugins (the marketplaces are already registered through the tracked settings):

claude plugin marketplace add JuliusBrussee/caveman
claude plugin install caveman@caveman

claude plugin marketplace add mvanhorn/last30days-skill
claude plugin install last30days@last30days-skill

datadog-mcp is an MCP server rather than a plugin; install it with:

claude mcp add --transport http datadog-mcp https://mcp.datadoghq.eu/api/unstable/mcp-server/mcp

Granola (installed via the Brewfile cask) also ships an official MCP server that exposes meeting notes and transcripts. Register it globally; the first connection opens an OAuth flow in the browser, no API key needed. Requires a signed-in Granola app (macOS 12+, Google or Microsoft account, microphone permission):

claude mcp add --transport http --scope user granola https://mcp.granola.ai/mcp

sentry, slack, and google-calendar are hosted connectors rather than local plugins. Enable them from the connectors directory at claude.ai and authenticate once; they then appear as MCP tools in Claude Code sessions.

To update everything, ask Claude in a session: Update installed plugins.

🧰 Companion Tools

CLI tools that complement Claude Code. They are installed outside the plugin system but configured from this repository.

rtk

rtk proxies common dev commands and strips their output down to what the model needs (60-90% token savings). A PreToolUse hook in the global settings rewrites Bash commands through it transparently. Activate with:

rtk init -g

πŸ–₯️ Claude Desktop

Only the mcpServers block is tracked. The rest of claude_desktop_config.json (account UUIDs, window state, onboarding flags) is local runtime state that shouldn't be versioned.

  • Real config: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Tracked copy: mac/claude/mcp_servers.json

Setup on a fresh machine:

mkdir -p ~/Library/Application\ Support/Claude

# File doesn't exist yet:
cp ~/.dotfiles/mac/claude/mcp_servers.json \
  ~/Library/Application\ Support/Claude/claude_desktop_config.json

# File already exists: merge mcpServers into it instead of overwriting
# (shallow merge, so a tracked server entry fully replaces any stale match)
jq -s '.[1] + {mcpServers: ((.[1].mcpServers // {}) + .[0].mcpServers)}' \
  ~/.dotfiles/mac/claude/mcp_servers.json \
  ~/Library/Application\ Support/Claude/claude_desktop_config.json \
  > /tmp/claude_desktop_config.json \
  && mv /tmp/claude_desktop_config.json \
  ~/Library/Application\ Support/Claude/claude_desktop_config.json

Restart Claude Desktop for the change to take effect. After adding a new MCP server by hand, copy the updated mcpServers block back into mac/claude/mcp_servers.json to keep it tracked.

google_health

google-health-mcp-unofficial reads user-authorized Google Health API v4 data (Fitbit, Pixel Watch) locally over OAuth; tokens never leave the machine and live at ~/.google-health-mcp/tokens.json.

Setup on a fresh machine, after the MCP entry above is in place:

  1. Create a Google Cloud OAuth client, enable the Google Health API, and add the redirect http://127.0.0.1:3000/callback.

  2. Run:

    npx -y google-health-mcp-unofficial setup --scope-preset full
npx -y google-health-mcp-unofficial auth
npx -y google-health-mcp-unofficial doctor

    On a headless host (SSH, no browser) use auth --manual instead; see docs/oauth.md.

  3. doctor --live calls safe Google Health identity/profile endpoints to confirm the API is reachable end to end.


Built with ❀️ from Mallorca

About

My personal Mac setup and configurations

Resources

Readme
Activity

Stars

2 stars

Watchers

1 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages