Agentic CodingGuide

Agentic Coding Guide

Tighten the feedback loop — principles, workflows, and guardrails for engineering with coding agents.

LLMs are tool calls in a loop. What makes them effective is feedback. Signals at every layer that help the agent evaluate its own work and course-correct. Linters, types, tests, hooks, agent browsers, sub-agents: each one is a feedback mechanism that tightens the loop. This guide is about setting those up.

Feedback signals

Each iteration
  1. On edit
    Formatting, type hints, lint autofix on every tool call
  2. Pre-commit
    Lint errors, secret detection, complexity limits
  3. Tests & build
    Unit tests, integration tests, build, security scan
  4. Agentic review
    Architecture violations, pattern drift, judgement calls
  5. Agentic QA
    UI flows, visual regressions, network behaviour

Before the Loop

Infrastructure and questioning discipline — the prerequisites for a productive feedback loop.

Code Quality

Deterministic feedback signals: linters, types, complexity limits, and custom lint rules.

Agent Workflows

The plan→execute→review→test loop, skills, and hooks.

Agent Teams

Multiple feedback perspectives in parallel — UX, testing, security sub-agents.

CI/CD & Automation

Feedback at scale: layered CI, event-driven agents, and security scanning.

Testing & QA

Verification feedback: unit, E2E, and agentic QA.

Troubleshooting

When the feedback loop breaks — common failure modes and what to do.

Before the Loop

Infrastructure and questioning discipline — the prerequisites for a productive feedback loop.