Claude Code Dotfiles — Ethical Capital

Configuration, skills, and plugin stack for Claude Code as used at Ethical Capital for investment research, compliance, and software development.

Published for transparency. Clients and prospective clients can see exactly how we instruct our AI tools — what standards we hold them to, what we check for, and how we think about quality.

Plugin Stack

We run 16 plugins. The principle: LSPs for code navigation, workflow plugins for discipline, domain plugins only when they earn their token budget.

Always On

Plugin	Source	What It Does
superpowers	claude-plugins-official	14 workflow skills (brainstorming, TDD, code review, debugging, etc.)
episodic-memory	superpowers-marketplace	Cross-session memory via semantic search
elements-of-style	superpowers-marketplace	Strunk's rules applied to all prose
brand-voice	knowledge-work-plugins	Brand guideline enforcement + generation
productivity	knowledge-work-plugins	Task management, memory management
frontend-design	claude-plugins-official	Anti-generic design direction
claude-md-management	claude-plugins-official	CLAUDE.md auditing and improvement
dead-code	jko-claude-plugins	Dead code detection and removal
rust	jko-claude-plugins	Rust critique, hardening, type strengthening

Language Servers (LSP)

Plugin	Language	Notes
typescript-lsp	TypeScript	Symbol nav, type checking
pyright-lsp	Python	Type checking, symbol nav
rust-analyzer-lsp	Rust	Full Rust IDE support

Domain-Specific

Plugin	Source	What It Does
database-design	claude-code-workflows	PostgreSQL schema design best practices
database-migrations	claude-code-workflows	Zero-downtime migration strategies
python-development	claude-code-workflows	Async patterns, testing, packaging, uv
debugging-toolkit	claude-code-workflows	Structured debugging methodology

Disabled (Available When Needed)

Plugin	Why Disabled
marketing-skills	~25 skills, rarely needed — re-enable for campaigns
posthog	~20 skills — re-enable for analytics work
serena	Replaced by per-language LSPs (TS/Python/Rust)
postgres-best-practices	Duplicate of project-level Supabase MCP
data-engineering	Not currently needed
code-review	Superpowers handles this
textual-tui	Not currently needed

MCP Servers

Global (~/.claude/.mcp.json)

Server	What
research-mcp	Railway-hosted web research (fetch, crawl, PDF, screenshot, search)
devin	DeepWiki documentation + Devin session management

Project-Level (.mcp.json)

Server	What
supabase	Database management, migrations, SQL execution
linear-server	Issue tracking, project management

Cloud Integrations (Account-Level)

Gmail, Magic Patterns, Mesh (contacts), Chrome automation.

What We Removed (and Why)

Server	Replaced By
filesystem	Claude Code built-in Read/Write/Edit/Glob/Grep
cloudflare-observability	workers-observability skill + wrangler CLI
cloudflare-bindings	workers-observability skill + wrangler CLI
serena (global MCP)	serena plugin (then replaced by per-language LSPs)
vectorbtpro	Configured but never enabled — removed dead config

Skills

Investment Research & Compliance

How we screen companies, write exclusion narratives, and check our own work.

Skill	Purpose
writing-exclusions	Voice and structure for exclusion narratives (public-facing on ethicic.com)
exclusions-manager	Full exclusion lifecycle — sources, approval workflow, taxonomy
evidence-crawl	Multi-source evidence gathering for exclusion research
research-reviewer	Research note quality review, staleness detection, MD-DB consistency
research-notes	Research note management for ~7,600 investment research notes
pipeline-runner	Orchestrates the overnight research pipeline
compliance-analyst	RIA marketing compliance — SEC Marketing Rule, Utah R164, CAN-SPAM
investment-analysis	Structured logical analysis — thesis validation, bias detection, decision audit trails

Writing & Argumentation

How we hold our own prose to account. These two are complementary: buzz-saw audits logic and evidence, elements-of-style (plugin) refines prose clarity.

Skill	Purpose
buzz-saw-editor	Ruthless argument editor — cuts filler, audits logic chains, flags unsupported claims
fact-checker	Multi-source fact-checking with structured verdicts and interactive edit approval
anti-racist-review	Checks writing about marginalized communities for encoded assumptions
veil-of-ignorance	Forces analysis through Rawlsian lens — who bears consequences the analyst doesn't?

Brand & Content

Skill	Purpose
ecic-brand	Brand voice, visual identity, and tone enforcement
serper-seo	SERP research, keyword analysis, content framing
signup-flow-cro	Signup and registration flow conversion optimization

Infrastructure & Observability

Skill	Purpose
deploy-worker	Cloudflare Worker deployment with pre-deploy checks and rollback
workers-observability	Worker health, binding inspection, metrics, debugging (replaces CF MCPs)
workers-best-practices	CF Workers production patterns — streaming, error handling, secrets

Code Quality & Review

Skill	Purpose
cynical-schema-review	Ruthless database schema review — type mismatches, missing constraints, edge cases
data-app-ui-review	UI/UX review for data apps, dashboards, analytics tools
accessibility-review	WCAG 2.2 AA + PDF/UA across all output types
auditing-claude-installation	Audit Claude Code setup for conflicts, duplicates, security risks
typeset-pdf	Branded PDF generation from markdown/Typst

Thinking Tools

Skill	Purpose
thought-partner	Structured reasoning for tricky, uncertain, or strategic problems

Configuration Principles

Token Budget Management

Every skill description, MCP tool name, and plugin instruction block consumes tokens in every prompt. We audit aggressively:

• Skills: ~95 total (was 130). Removed duplicates from plugins re-exporting the same skills.
• MCP tools: ~105 deferred tools (was 152). Removed redundant servers, fixed Serena context flag.
• Plugin MCP servers: 18 (was 27). Disabled servers that duplicate project-level MCPs.
• LSP servers: 3 (TypeScript, Python, Rust). Replaced Serena's single-language MCP with per-language LSPs.

Credential Management

All secrets flow through Doppler (project: rawls, env: dev). MCP configs never contain hardcoded credentials — they wrap commands in doppler run --.

Permission Philosophy

Pattern-based allow rules (Bash(git:*), Bash(doppler run:*)) rather than per-command approvals. Session-specific one-off permissions are cleaned periodically to prevent settings.local.json bloat.

Installation

Copy any skill directory into your .claude/skills/ folder:

# Single skill
cp -r buzz-saw-editor/ ~/.claude/skills/buzz-saw-editor/

# All skills
for d in */; do [ -f "$d/SKILL.md" ] && cp -r "$d" ~/.claude/skills/; done

Skills are loaded automatically by Claude Code when their trigger description matches your task.

Philosophy

We believe in radical transparency. These skills encode our actual working standards — not aspirational documentation, but the instructions we give our tools every day. If a skill says "never skip verification," that is a rule we enforce on ourselves.

The investment research and compliance skills are especially relevant for due diligence. They show how we screen companies, write exclusion narratives, check facts, and review marketing materials for regulatory compliance.

License

CC BY-NC 4.0 — use freely, but not for resale.