Philosophy

The VisiTrans Claude Toolkit is built on one core idea: compound engineering.

Compound engineering

Each unit of engineering work should make subsequent units of work easier — not harder.

Most development workflows are linear: you plan, build, ship, and move on. Knowledge stays in people's heads or gets lost in Slack threads. The next project starts from scratch.

Compound engineering changes this. Every project generates artifacts — decisions, patterns, solutions, runbooks — that accumulate and compound. The tenth project is dramatically faster than the first because the toolkit has learned from the previous nine.

graph TD
    A[Project 1] -->|journal + consolidate| B[Knowledge Base]
    C[Project 2] -->|journal + consolidate| B
    B -->|compound-docs| D[Institutional Memory]
    D -->|Step 0: Load Knowledge| E[Project 3]
    E -->|faster, fewer mistakes| F[Better Outcomes]

Design principles

1. Workflows, not tools

The toolkit doesn't give you a bag of disconnected tools. It gives you three end-to-end workflows that guide you from start to finish:

• Development (/0-start through /6-operate) — for building software
• Product Design (/pd-0-start through /pd-6-handoff) — for designing products
• Knowledge Work (/kw-0-start through /kw-4-publish) — for non-code deliverables

Each workflow is a sequence of phases. Each phase has a command, a set of agents, and relevant skills.

2. Orchestrators over individual agents

Complex tasks need coordination. Instead of invoking agents one by one, you invoke an orchestrator that dispatches the right sub-agents in parallel and aggregates their results.

Example: /review invokes the implementation orchestrator, which dispatches 6 reviewers simultaneously (security, TypeScript quality, race conditions, complexity, performance, patterns) and produces a single quality gate report.

3. Skills activate automatically

Skills are context-sensitive knowledge modules. You don't invoke them directly — they activate when the work you're doing matches their domain.

Writing an API endpoint? The defense-in-depth skill activates for input validation. Running tests? The test-driven-development skill guides your approach.

4. Security is non-negotiable

Two hook scripts run on every file write:

• Secret scanner — blocks hardcoded credentials, API keys, tokens
• Security lint — warns about eval(), innerHTML, raw SQL

These aren't optional. They run automatically, every time.

5. Branch-based git workflow

The toolkit enforces branch-based development:

• Never commit to main or master directly
• Always create a feature branch first
• Commit messages follow conventions
• Hooks verify compliance

6. Plugins for specialization

The core toolkit provides workflows, agents, and skills that every team needs. Department-specific concerns (financial reporting, support workflows, marketing content) live in separate plugins that can be installed independently.

This keeps the core lean while allowing unlimited specialization.

The learning loop

Every session contributes to institutional memory:

1. Journal (/journal) — capture decisions and learnings during work
2. Consolidate (/consolidate) — turn journal entries into specs, PRDs, and ADRs
3. Compound (/compound) — build up the solutions library
4. Load (Step 0 of every orchestrator) — consult past knowledge before starting

Over time, the toolkit accumulates solutions to problems your team has already solved. New team members benefit from this knowledge immediately.