This is the developer pack. Ten Claude Code subagents shaped by real solo-dev work on a TypeScript / SST v3 / DynamoDB / React stack, plus two skills that extend what Claude Code can do in any project.
The agents are opinionated. They push back, they assume a specific stack, and they're focused on one job each. That focus is the point. A generic "code review" agent produces generic feedback. An agent that knows your stack produces feedback worth reading.
This is Claude Code, not claude.ai. You run these in the terminal via the Claude Code CLI. If that sentence makes no sense, you probably want the Business Agents pack instead.
Code and architecture
architectArchitecture and design reviews. Abstractions, data modeling, service boundaries, dependency direction, build-vs-buy. The "should we even build it this way?" agent.
challengeCode correctness review. Bugs, edge cases, race conditions, type safety holes, error handling gaps. Not architecture, not security. Just correctness.
securityAdversarial security review. Auth bypasses, injection vectors, exposed secrets, overly broad IAM, client-side trust issues. Thinks like an attacker.
dynamoDynamoDB single-table design. Access pattern analysis, GSI planning, query optimization, idempotency ledgers. Invoke when designing new tables or adding access patterns.
lambdaLambda performance. Cold starts, bundle size, memory and timeout tuning, SDK initialization, execution patterns. Run when functions feel slow.
Frontend and SEO
uiReact component reviews. Component API design, Tailwind consistency, shadcn/ui integration, React 19 patterns. Invoke when building or reviewing UI components.
seo-contentLocal-search content review. Heading hierarchy, keyword placement, NAP consistency, service-area coverage, search intent match. Run when writing or reviewing client copy.
seo-technicalTechnical SEO audit for 11ty/Nunjucks sites. Meta tags, LocalBusiness schema, canonical URLs, Open Graph, sitemap/robots, internal linking. Run before deploying.
Writing
human-voiceCopy editor that strips AI writing patterns from a draft. Targets significance inflation, negative parallelisms, contaminated vocabulary, uniform rhythm, and em dash overuse; doesn't change meaning.
Each agent ships with worked examples in agents/<name>/examples/: real reviews from real projects, not synthetic demos.
Skills
Two slash commands that work in any Claude Code session:
/debateSpawns 3 parallel agents to argue a decision from opposing angles, then synthesizes into a concrete recommendation. Use when you're unsure about an architecture or design call.
/human-voiceRewrites any text to strip AI writing patterns and read as unmistakably human. Applies automatically to all writing tasks in the session.
Installation
One-liner to copy all agents into ~/.claude/agents/:
mkdir -p ~/.claude/agents && \
for a in architect challenge security cost dynamo lambda ui seo-content seo-technical human-voice; do \
cp "agents/$a/agent.md" "$HOME/.claude/agents/$a.md"; \
done
Then invoke any agent from Claude Code with /agent:architect, /agent:challenge, etc. See INSTALL.md in the pack for the full setup, including skill installation and optional team configuration.
License and use
Proprietary. Licensed for the named purchaser's own use: your own products, your internal tooling, your paid client work. Not for resale, redistribution, or sharing with co-workers or clients.
