codex-issue-plan-execute_skill

Overview

This skill provides an autonomous issue planning and execution workflow for Codex, combining planning, queuing, and execution in a single stateful tool. It supports batch issue processing and maintains persistent planning and execution agents to reduce overhead. The skill outputs unified planning and execution result files and a final report for auditability.

How this skill works

The workflow spawns two persistent agents: a Planning Agent that converts issues into validated solution JSON, and an Execution Agent that implements and tests those solutions. Issues are processed sequentially: planning results are appended to planning-results.json, then successful solutions are queued for execution and recorded in execution-results.json. State is tracked in a JSON schema to allow resuming, reporting, and minimal agent recreation through send_input() routing.

When to use it

• Batch-processing multiple issues that need structured plans and automated implementation.
• When you need repeatable, auditable records of planning and execution (planning-results.json / execution-results.json).
• Integrating CLI-driven AI orchestration into developer workflows (Gemini/Qwen/Codex).
• Resuming interrupted pipelines or running long-lived planning/execution agents across many tasks.
• Automating enforcement of solution schema and quality standards across a backlog.

Best practices

• Read the issue handling and solution schema specs before running to ensure valid inputs.
• Feed issues in small batches initially to validate agent prompts and quality thresholds.
• Keep the persistent agents alive for the full run to avoid context loss and reduce latency.
• Monitor state schema and result files to detect and recover from failed steps quickly.
• Use snapshot/resume features for long-running pipelines or when iterating on prompts.

Example use cases

• Convert a list of tracked issues into concrete implementation tasks and corresponding commits.
• Automate triage and solution generation for security fixes, then execute patches automatically.
• Run a nightly batch that plans and implements low-risk refactors across multiple repositories.
• Resume a previously interrupted pipeline from a saved snapshot to complete remaining issues.
• Enforce a standardized solution JSON format across teams and collect unified execution metrics.