Step 01
Customize the 'Suggest Skills' Automation
In Codex, select the 'From recent PRs and reviews suggest next skills to deepen' automation template. Configure the automation to run on a recurring schedule, such as weekly on Fridays at 10:00 a.m.
AdvancedEngineering
Build a Self-Improving AI to Generate Agent Skills in Codex
Use a Codex Automation to create a 'meta-agent' that analyzes your codebase, identifies opportunities for new reusable agent skills, and then spawns sub-agents to automatically test and validate those new skills.
Try this with your AI assistant
Download the skill for this workflow. Version 1.0.1.
From 19:00 to 25:27, Claire builds a weekly Codex loop that inspects recent work, identifies reusable skills, and uses subagents to produce them. Clip range: 19:00 to 25:27.
Before you start
What you need
- Codex account with automation templates enabled
- GitHub repository with recent pull request history
- Weekly automation schedule
- Permission to create agent threads and update the skill library
- Repository main branch for validation testing
What you’ll make
A validated set of reusable agent skills with test results recorded against the repository base branch.
Tools used
- CodexVisitCodex
OpenAI's cloud-based AI software engineering agent that can execute code, run tests, and handle complex multi-file tasks autonomously.
- GitHub CopilotVisitGitHub Copilot
AI pair programmer
The workflow
Follow the sequence once, then adapt the prompts, checks, and handoffs to your own setup.
4 steps
Step 02
Enhance the Prompt with a Validation Loop
Modify the template's prompt to instruct the agent to identify potential new skills, such as those from a command line tool, and then validate them. The prompt should require the main agent to spawn a sub agent for each potential skill, giving it a specific goal to test the skill against the repository's main branch.
Example prompt
If you identify a skill, spin up its own thread and use that skill... validated against the base branch of the repo. We want to confirm that the skill actually works and outputs high quality... You should use a goal when validating the skill.
Step 03
Run the Automation and Observe Sub-Agents
Run the automation. The main agent will scan your GitHub history and start creating sub agents for each potential skill it identifies. Monitor the chat list in Codex to observe these new sub agents appearing, each assigned a specific goal to validate one skill.
Step 04
Review and Implement Validated Skills
After a sub agent completes its goal, it will report whether the new skill is functional. Review the results from each sub agent. For each validated skill, add it to your agent's reusable skill library.
What good looks like
- Automation launches on the configured weekly schedule
- Sub-agents are created for each proposed skill
- Each skill is tested against the repository main branch with a defined goal
- Validated skills are added to the reusable skill library
Build your next product with ChatPRD
Turn an idea into a PRD, user stories, and a plan.
Runbook notes
How to recover when the loop fails and where human judgment helps.
If it goes sideways
- Suggested skills fail on the repository base branch
- Record the failing commands and restrict the skill to supported environments
- Sub-agent validation lacks measurable success checks
- Define explicit repository tasks such as passing tests or successful builds
- Automation analyzes stale pull request data
- Limit scans to merged PRs from a recent date range
Related workflows
intermediateProductCodexClaude
Build a Personal PM Operating System with a Shared Knowledge Repo
Create a repository-based system to provide an AI assistant with durable context about your team, projects, and communication channels. This enables it to perform routine tasks like summarizing daily updates and identifying stakeholders, acting as a personalized chief of staff.
Open
intermediateProductMeta Muse
Create a Personalized Family Newspaper with Meta's Muse
Use Meta's Muse to generate a printable, multi-page family newspaper that includes a weekly schedule, weather forecasts, and activity ideas. This workflow leverages Muse's strength in creating visually appealing, family-oriented artifacts.
Open
advancedProductCursorSlack
Build an Automated Growth Experiment Loop with Multiple AI Agents
Coordinate agents in Slack and Cursor to propose growth experiments for human review. Ezra’s Auto XP system was still being developed at the time of the episode.
Open
Start shipping better products.
Join 250,000+ product managers who use ChatPRD to write better docs, align teams faster, and build products users love.
Free to start · 14-day Pro trial · SOC 2 Type II
