AI Coding Agent SOP

# ai-coding-agent-sop

AI Coding Agent SOP

## Overview
AI Coding Agent SOP is the operating system for a one-person company that ships product changes with AI assistance but still protects reliability, security, and velocity. Most solo founders do not fail because AI is weak. They fail because there is no repeatable method between prompt and production. This skill gives you a concrete system to turn AI coding output into production-safe releases without creating long-term code debt.

## When to Use This Skill
Use this skill when you are shipping code weekly and one or more of these patterns appears:
- AI-generated changes are fast but regressions keep showing up.
- Features ship, but post-release bug cleanup consumes the next sprint.
- You cannot explain why a change is safe beyond "the code looked fine."
- Pull requests are too large to review confidently as a solo operator.

Use it before incidents happen, not after. The highest leverage moment is when your release pace starts to increase.

## What This Skill Does
This skill defines a full release loop with explicit quality gates:
1. Scope gate: every task has a business outcome and done condition.
2. Implementation gate: tasks are sliced into small, auditable changes.
3. Verification gate: tests and validation run before merge.
4. Release gate: deployment is staged with rollback instructions.
5. Learning gate: recurring failures are converted into updated SOP rules.

The result is simple: higher shipping confidence per week, lower incident cost, and faster compounding product improvement.

## How to Use
Step 1: Write an executable task brief.
Define one user-visible outcome, one measurable success signal, and one explicit non-goal. Add file constraints, dependency limits, and risk notes before asking any AI agent to edit code.

Step 2: Enforce task slicing.
Break implementation into isolated units such as schema, API, UI, tests, and docs. One unit per PR whenever possible. Small PRs reduce review fatigue and rollback blast radius.

Step 3: Require an acceptance checklist.
Each task needs concrete checks:
- Behavior check: exact expected user outcome.
- Test check: command list and pass criteria.
- Safety check: auth, billing, data integrity, and migration impact.
- Observability check: log or metric you will verify post-release.

Step 4: Run pre-merge quality gates.
Do not merge on prompt confidence. Merge on evidence:
- Tests pass locally or in CI.
- Diff is readable and scoped.
- Edge case identified for failure path.
- Rollback command documented and tested for plausibility.

Step 5: Ship in controlled stages.
Release to low-risk traffic first when possible. Monitor errors, latency, and one business KPI (signup completion, checkout success, task completion) for 30-60 minutes before full rollout.

Step 6: Run a weekly reliability retro.
Review the past seven days:
- Which changes caused avoidable rework?
- Which prompts produced ambiguous output?
- Which review checklist item would have caught the issue earlier?
Then update the SOP document. The skill compounds only when updated from evidence.

## Output
A strong output from this skill includes:
- A reusable AI coding task brief template.
- A five-gate release checklist.
- A lightweight PR rubric for solo review.
- A rollback playbook with commands and owner actions.
- A weekly retro template that turns incidents into permanent process improvements.

## Common Mistakes
- Treating AI speed as a replacement for engineering discipline.
- Merging large multi-layer diffs because "it seems to work."
- Skipping explicit rollback steps in deployment notes.
- Measuring productivity only by features shipped, not change failure rate.
- Switching tools weekly before stabilizing one workflow.

## What Good Looks Like
Good looks like consistent weekly shipping with fewer surprise regressions, shorter incident windows, and clearer confidence in each release decision. You should be able to answer:
- Why is this change safe?
- How do we know it works?
- What is the rollback path?
- What did we learn and update in the SOP?

If you can answer those four questions every release week, AI coding has moved from random acceleration to a durable business asset.