My personal development operating system — a curated repository of skills, sub-agents, workflows, output styles, hooks, and plugins that power my daily software development workflow with Claude Code.
This is where I document and evolve my Vibe Coding process.
Skills live in skills/, grouped by activity. The categories exist only in this repo: import-skills.sh shows them as headers in its picker and installs every skill flat into .claude/skills/<name>/, because Claude Code only discovers skills one level deep.
Legend: /name = invoked only as a slash command (disable-model-invocation: true, the former commands/); global = always installed into ~/.claude/skills/.
| Skill | Description |
|---|---|
/end-session global |
Generate a session log with summary, changes, and next steps |
/handover |
Save a structured context dump to continue in a fresh session |
/logbook global |
Update the daily PT-BR work logbook at .claude/logbook.md |
napkin-runbook global |
Read and curate the repo's runbook of reusable rules (.claude/napkin.md) |
/skill-audit |
Audit a skill to test, benchmark, optimize, or remove it |
/skill-optimize-description |
Optimize a skill's description to improve triggering accuracy |
skill-reviewer |
Review and improve skills against official best practices |
| Skill | Description |
|---|---|
/ai-agent-security-analysis |
Audit an AI agent's security posture, MCP governance, and guardrails |
/prompt-optimizer |
Rightsize prompts, CLAUDE.md, SKILL.md, and tool descriptions |
| Skill | Description |
|---|---|
coupling-analysis |
Analyze module coupling by strength, distance, and volatility |
solid-checker |
Analyze and fix SOLID principle violations |
| Skill | Description |
|---|---|
/create-update-makefile |
Create or update a self-documenting Makefile |
frontend-design |
Create distinctive, production-grade frontend interfaces |
frontend-testing |
Generate Vitest + React Testing Library tests |
playwright-expert |
Write and debug Playwright E2E tests |
python-async-postgres |
Set up async PostgreSQL with SQLAlchemy and test fixtures |
python-cashews-cache |
Add caching with cashews and a diskcache backend |
python-fastapi-api-key-auth |
Add X-API-Key header authentication to FastAPI apps |
python-fastapi-projects |
Create production-ready FastAPI projects with async SQLAlchemy |
python-testing |
Generate pytest suites with fixtures, parametrization, and mocking |
| Skill | Description |
|---|---|
ce-code-review global |
Structured review for bugs, regressions, tests, and standards |
frontend-code-review |
Review .tsx/.ts/.js files for quality, performance, and business logic |
/pr-review-to-subtasks global |
Create work-item subtasks from a review file |
/pr-review-workflow |
Review an Azure DevOps PR and create subtasks from the review |
pr-security-review global |
Security review of an explicit base...head revision range |
pr-description global |
Write a PR title and description |
| Skill | Description |
|---|---|
/blitzy-create-codebase-docs |
Generate a codebase-ingestion doc at docs/02-codebase.md |
/blitzy-create-comprehensive-documentation |
Add module- and code-level documentation across the project |
/blitzy-create-product-description |
Create a structured product description |
/create-simple-feature-prd |
Create a single-feature PRD through clarifying questions |
/create-simple-feature-tasks |
Generate a phased task list from requirements |
/create-update-changelog |
Create or update CHANGELOG.md and tag the release |
create-update-readme |
Create or update the project README.md (PT-BR) |
docs-generator global |
Generate or update a standardized docs/ folder |
user-story-spec-writer |
Turn vague user stories into specs ready for AI coding agents |
| Skill | Description |
|---|---|
/create-commit-message |
Create a commit message from the current diff |
| Skill | Description |
|---|---|
azure-devops-cli global |
Manage Azure DevOps resources via the az devops CLI |
planecli global |
Manage Plane.so work items, cycles, and modules via planecli |
| Skill | Description |
|---|---|
/explainer-graphic |
Create visual infographics using real-world analogies |
knowledge-distiller |
Distill transcripts, articles, or notes into a PT-BR knowledge document |
| Skill | Description |
|---|---|
cognicode-sow |
Generate a Cognicode project scope document (SOW) |
competitors-analysis |
Evidence-based competitor analysis from actual cloned code |
copywriting |
Write and improve marketing copy for web pages |
/html-it |
Produce HTML output instead of markdown |
improve-tech-writing |
Apply clear technical writing patterns in PT-BR |
macos-cleaner |
Analyze and reclaim macOS disk space |
/mermaid-diagram-generator |
Generate interactive HTML diagrams with Mermaid.js |
sop-creator |
Create runbooks, playbooks, and technical documentation |
unslop global |
Cut AI tells from writing and restore human voice |
workflow-visualizer |
Map a system or workflow as an interactive HTML diagram |
Third-party assets that are installed, not copied, are declared in two YAML catalogs at the repo root. To add one, add an entry to the YAML; the scripts build the install and update commands from its fields.
Plugins — external-plugins.yaml, installed with scripts/install-plugins.sh.
| Plugin | Marketplace | Scope | Description |
|---|---|---|---|
github |
official | user | GitHub plugin |
claude-md-management |
official | user | CLAUDE.md management |
skill-creator |
official | user | Skill scaffolding helper |
claude-code-setup |
official | user | Claude Code setup helper |
thermos |
dev-os (this repo, plugins/) |
user | Thermo-nuclear branch review |
compound-engineering |
EveryInc/compound-engineering-plugin | user | Compound Engineering pipeline |
andrej-karpathy-skills |
multica-ai/andrej-karpathy-skills | user | Andrej Karpathy guideline skills |
mattpocock-skills |
official | user | Matt Pocock skills |
pyright-lsp |
official | project | Python type checking |
typescript-lsp |
official | project | TypeScript type checking |
frontend-slides |
zarazhangrui/frontend-slides | project | Frontend slides |
Mods — external-mods.yaml, installed and uninstalled with scripts/manage-mods.sh install|uninstall. A mod is a plugin built from function hooks that changes the Claude Code UI (panes, bands, status lines). Uninstall leaves the marketplace registered. A mod with marketplace: builtin ships with Claude Code and is enabled/disabled instead.
| Mod | Marketplace | Scope | Description |
|---|---|---|---|
savvy-progress |
JohnnyVizz/claude-kit | user | Progress bar above the prompt and subagents panel (/agents-info) |
Skill packages — external-skills.yaml, installed with scripts/install-external-skills.sh through npx skills add.
| Package | Source | Scope | Description |
|---|---|---|---|
langchain-skills |
langchain-ai/langchain-skills | project | LangChain, LangGraph and Deep Agents skills |
shadcn-ui |
shadcn/ui | project | shadcn/ui skills |
Third-party skills copied into skills/ (such as unslop) record their upstream in frontmatter: metadata.source (a GitHub tree/<ref>/<dir> URL of the skill directory, or a gist URL with a #file-<name> anchor), metadata.last_synced (the date of the last copy from upstream, which is what marks a skill as third-party) and metadata.local_changes (what was adapted locally, or "None."). Own skills that are only inspired by external material cite it in metadata.reference instead.
scripts/sync-third-party-skills.sh lists them with the upstream changes since last_synced, and shows the upstream diff for the skills you pick. With --apply it merges those changes into the local copy (a three-way merge per file, so local adaptations survive; conflicts are left as markers) and bumps last_synced. It needs an authenticated gh and never commits.
Sub-agents in agents/, imported per project with scripts/import-agents.sh.
| Agent | Description |
|---|---|
ce-code-simplicity-reviewer |
Final review pass for YAGNI violations and simplification opportunities |
framework-docs-researcher |
Gather official docs, version constraints, and best practices for a dependency |
frontend-code-reviewer |
Review recently changed frontend code (components, styles, hooks) |
Multi-agent workflow scripts in workflows/, installed globally into ~/.claude/workflows/ with scripts/import-workflows.sh.
| Workflow | Description |
|---|---|
dev-flow |
Implement a markdown plan, open a PR, review it with 5 reviewers, apply the fixes, and audit the docs |
code-review-flow |
Review an existing PR with 5 reviewers and produce a verified final report |
Both rely on the thermos and mattpocock-skills plugins and on the ce-code-review and pr-security-review skills.
Output styles in output-styles/, installed globally with scripts/import-output-styles.sh.
| Style | Description |
|---|---|
ELI5 |
Keep explanations simple |
Escrita Técnica Clara |
Direct PT-BR technical writing, without anglicisms or filler |
hooks/log-skill.sh logs every Skill invocation (timestamp, skill, args) to the project's .claude/skill-usage.log. Run scripts/setup-skill-hook.sh once per machine to copy it into ~/.claude/hooks/ and register it as a PreToolUse hook. Requires jq.
dev-os/
├── skills/ # Reusable skills, grouped by category
│ ├── claude-code/ # Session and skill maintenance
│ ├── engineering/ # build, code-review, documentation, git, ...
│ ├── learning/ # Knowledge distillation and explainers
│ └── productivity/ # Writing, diagrams, ops, misc
├── agents/ # Sub-agent definitions
├── hooks/ # Hook scripts (single source of truth)
├── workflows/ # Multi-agent workflow scripts
├── output-styles/ # Output styles
├── plugins/ # Dev-OS's own plugin marketplace (thermos)
├── scripts/ # Import/install helper scripts
├── docs/ # Reference notes (LangGraph practices)
├── external-plugins.yaml # External plugins to install
├── external-mods.yaml # Claude Code mods to install or uninstall
├── external-skills.yaml # External skill packages to install
├── CONTEXT.md # Domain glossary
└── .claude/ # Local session state + agents symlink
This repo is my single source of truth for development configurations. I clone it and import the skills, agents, and other assets into each project with the scripts in scripts/. When I learn a new pattern or refine a workflow, I update it here so every future session benefits.
The skills cover the full stack I work with daily — from FastAPI backends and React frontends to project management, documentation, and code quality. The slash-only skills automate repetitive tasks like writing PRDs, changelogs, and PR summaries.
- Clone the repo into
~/dev-os - From any project (including Dev-OS itself), run the import scripts to install assets:
~/dev-os/scripts/import-skills.sh,import-agents.sh(per-project),import-workflows.sh,import-output-styles.sh(global, into~/.claude/).import-skills.shlists the skills grouped by category in an interactive picker (--allskips it) and flattens the categorizedskills/tree into.claude/skills/<name>/, since Claude Code only discovers skills one level deep. The skills listed inGLOBAL_SKILLS(e.g./end-session) always go to~/.claude/skills/so every project sees them. - Install external plugins and skill packages with
~/dev-os/scripts/install-plugins.shandinstall-external-skills.sh(picker;--allskips it). Both update entries already installed, so rerun them to refresh. Mods go through~/dev-os/scripts/manage-mods.sh install(oruninstall). - Once per machine, run
~/dev-os/scripts/setup-skill-hook.shto log skill usage - Open the project with Claude Code
- Use
/create-simple-feature-tasksto break down a feature,/pr-descriptionto write a PR description, or any other skill