365 Five-Step Dev (Codex Edition): A Light Governance Layer for Non-Coding Users
Doesn't duplicate Codex's native dev workflow; adds a thin business-governance layer at the key decision points: goal capsule, A/B/C risk grading, five-step execution, evidence-based acceptance — so business users who don't read code can still safely drive AI development.
The problem: I want to use Codex for development, but I don't read code
Many business owners face a real dilemma: they know what they want and can judge whether the result is right, but the middle part — how Codex writes, tests, and commits — is a black box. Either they let go completely, or they ask at every step. Neither extreme works.
The 365 Five-Step Dev (Codex edition) has a clear positioning: a thin business-governance layer around Codex, not a second dev workflow. Codex's native planning, coding, debugging, review, testing, and Git delivery stay with Codex; this skill only governs what business decisions need approval, what needs verification, and when to stop.
Goal capsule: align once, valid throughout
Derive or confirm five fields from the request:
- Business outcome
- Requested delivery level (local / commit / push / PR / merge / deploy)
- Exact protected target and effect
- Acceptance evidence
- Stopping condition
When the user clearly names the action, target, and business outcome, that's goal-scoped authorization. Preflight, technical fixes, safe rollback, propagation monitoring, and serial continuation toward the same outcome don't re-ask. Reconfirm only when the target, scope, material effect, or permission boundary changes.
A/B/C risk grading: by actual impact, not domain words
| Grade | Actual boundary | Depth |
|---|---|---|
| A — Small | Safe, contained, reversible; no live effect | Inspect → change → focused verify |
| B — Standard | A bounded product or business behavior with no unresolved protected effect | Run all five steps, add no ceremony |
| C — Protected | Actual production/infrastructure effect, bulk or destructive live-data action, unresolved permission/compliance decision | Scale only to the unresolved risk |
Key principle: database code, migration files, permission implementation, payment-flow code, or a commit by itself is not C-grade. What matters is whether it actually touches a live effect or an unresolved business risk.
Five steps: Research → Plan → Implement → Review → Verify
- Research — Read the nearest AGENTS.md, relevant source, tests, and the smallest necessary runbook; establish current behavior, affected people/systems, smallest coherent change, known facts, assumptions, and unresolved business decisions.
- Plan — Use the goal capsule as the plan when sufficient; read a requirement brief only when outcome, scope, acceptance, rollback, or multi-milestone ownership remains materially unresolved.
- Implement — Let native Codex and the most specific project skill handle the smallest complete change; reuse existing contracts, verify incrementally, avoid unrelated cleanup, keep changes isolated and attributable.
- Review — Review the actual diff against the business outcome: requirement coverage, unintended behavior, scope expansion, data/permission/privacy/compatibility risk, failure paths, rollback, evidence gaps, delivery completeness.
- Verify — Map each important business outcome to observed evidence; distinguish passed, failed, and not verified. Prefer the real affected flow over implementation claims.
Ready protected execution
When the goal capsule names the protected action, exact target, and outcome, and the project provides exact reviewed source, guarded operator, rollback, and readback:
- Run one complete preflight
- Consume the active goal authorization without a second approval
- Execute one serial guarded attempt
- Read back the target and close success, or run the fixed safe rollback and report the terminal result
Never blind-retry or execute protected mutations concurrently.
Install
bash git clone https://github.com/laojin1900/365Skill.git ./install.sh 365-five-step-dev-codex
Repo: github.com/laojin1900/365Skill/tree/main/skills/365-five-step-dev-codex
Made by Laojin · AI that ships
365SkillAn agent-skills lab: 13 in-house skills
365Skill is our public lab for agent skills: a standard SKILL.md format, a deny-by-default publish policy, and an evals harness. It holds 13 original 365 skills — 11 public and 2 internal. Apache-2.0 — star it, install it, file issues.
More from Laojin: Sellenca · 365AIOrg · AllModelsAPI · 365Loopa · 365 Ops