implement-codex

Implement approved multi-phase software plans by running Codex in a writable sandbox with Claude verification.

Updated Aug 27, 2026
One-click install
npx skills add https://github.com/KMakayee/playbook --skill implement-codex
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: implement-codex
Source: https://github.com/KMakayee/playbook/tree/main/.claude/skills/implement-codex
Command: npx skills add https://github.com/KMakayee/playbook --skill implement-codex

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It helps you implement an already-approved plan by having Codex generate code changes phase-by-phase in a sandbox while Claude verifies that each phase stays aligned with the plan and produces coherent, expected outputs.

Core Features & Use Cases

  • Phase-by-phase implementation: Executes each unchecked phase in tasks/plan.md with a controlled workflow that captures baselines, runs Codex, and evaluates results.
  • Tight plan adherence via contracts: Uses explicit mismatch and blocked contracts to prevent Codex from editing when the plan premise is wrong or when the phase needs capabilities like network access.
  • Robust verification gates: Enforces output schema checks, parses Codex JSON logs, detects verification-command violations, cross-checks diff scope, and updates phase checkmarks and metrics.
  • Recovery and escalation: Supports mismatch re-research, blocked-state Claude takeover, and crash diagnosis with either manual escalation or a limited retry path.

Quick Start

Run implement-codex after you have a finalized tasks/plan.md and the required tasks/research-codebase.md in place, so each phase can be implemented with Codex and verified against the plan.

Frequently Asked Questions about implement-codex

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I implement a multi-phase coding plan with Codex and verify the diffs?

To implement a multi-phase coding plan with Codex, execute unchecked phases in a writable sandbox while a separate verifier checks JSON logs, validates output schemas, and cross-checks diff scope to ensure strict surgical edits.

What is needed to start automated code implementation using a sandboxed agent workflow?

Automated code implementation requires a finalized tasks/plan.md defining phase structure, plus tasks/research-codebase.md and tasks/design-decision.md citing relevant files and outlining success criteria before execution begins.

How do agent workflows handle plan adherence when Codex makes unexpected code changes?

Agent workflows enforce plan adherence by applying mismatch and blocked contracts that prevent Codex from editing when the plan premise is wrong, triggering a re-research phase or a manual escalation path instead.

Can I use network access or custom test commands during sandboxed code generation?

Sandboxed code generation applies strict network and test-command constraints, blocking execution if a phase requires capabilities like network access and detecting verification-command violations to maintain diff scope integrity.

What happens when Codex crashes or produces invalid JSON output during phase implementation?

When Codex crashes or outputs invalid JSON, the workflow supports crash diagnosis with either manual escalation or a limited retry path, while a blocked-state allows a verifier takeover to recover the phase.

Does implement-codex work for updating metrics and checkmarks after each coding phase?

Yes, implement-codex updates phase-level metrics and checkmarks in tasks/plan.md after verifying output schemas and cross-checking real-diff scope against the approved plan's success criteria.