The Quality Pipeline
Every change an agent makes passes through 6 quality gates before it reaches your codebase. Nothing ships without passing all of them.
Gate 1: Specification
Before any code is written, Dispatch generates a functional specification from your goal. The spec defines what will be built, the acceptance criteria, and the constraints. You review and approve the spec before implementation begins.
This ensures the agent's understanding matches your intent. No surprises, no wasted implementation cycles. If the spec doesn't look right, request changes before a single line of code is written.
Gate 2: Tests Before Code
After the spec is approved, Dispatch generates test requirements that the implementation must satisfy. Tests are defined before implementation starts, not after. This test-first approach means agents write code to pass concrete criteria, not vague descriptions.
When the implementation completes, the test suite runs automatically. If tests fail, the agent iterates until they pass or escalates for your review.
Gate 3: Security Scanning
Every change is scanned for security issues before it can proceed. Dispatch checks for common vulnerability patterns, credential exposure, injection risks, and unsafe operations. Security scanning runs automatically on every agent run with no configuration required.
When a security issue is detected, the task is blocked and the findings are surfaced for your review. You cannot approve a task with unresolved security findings.
Gate 4: Clean Code Enforcement
Dispatch enforces code quality standards on every change. The clean code gate checks for structural issues, complexity problems, and style violations. It catches the kinds of issues that pass tests but degrade the codebase over time.
Clean code rules are fully customizable. Bring your own rules, disable defaults, or let Dispatch learn your preferences from your review feedback over time.
Gate 5: Performance Checks
Before changes reach review, Dispatch checks for common performance issues. Database access patterns, resource usage, blocking operations, and algorithmic concerns are flagged automatically.
Performance findings are surfaced as warnings, not hard blocks, so you can make informed decisions about tradeoffs.
Gate 6: Human Review
The final gate is you. Every change that passes the automated gates is presented in a review workspace where you can inspect the diff, check the spec compliance, and approve or request changes.
Low-risk changes (lint fixes, documentation) can be auto-approved. High-risk changes (cross-boundary, irreversible, security-adjacent) require explicit approval with a spec checklist. You control the risk thresholds.
Self-Improving
The pipeline learns from your feedback. When you request changes or reject a task, Dispatch records the correction and applies it to future runs. Over time, agents make fewer mistakes on the patterns you care about.
You don't need to write rules manually. The pipeline observes your preferences and adapts. If you do want explicit rules, you can add them in Settings.
Fully Customizable
Every gate is configurable. Disable gates you don't need. Add your own rules to any gate. Adjust risk thresholds for auto-approval. Configure which agents handle verification. The pipeline works out of the box, but it's yours to shape.
Bring your own linting rules, security policies, or style guides. Dispatch integrates them into the pipeline alongside its built-in checks.