How I Use Codex

I do not use Codex as one infinitely flexible assistant. I give it named modes with specific jobs, outputs, and stop conditions.

This is not my whole skill library. These are the six files I return to when I want to take an idea from a vague seed to something I would actually build. Each one interrupts a different failure mode: empty ideation, shallow agreement, generic taste, polite criticism, endless planning, or generated-looking residue.

---
name: hatch
description: Find strong ideas for what to build next. Use when the user asks for a new idea, asks what they should make or work on next, or wants untapped applications of current AI capabilities. Default to 1 to 3 tight pitches. Treat pitching, saving, specifying, building, and publishing as separate authority gates.
---

# Hatch

Find ideas that feel inevitable in hindsight: specific applications of current technology that should exist, but mostly do not yet.

## Workflow

1. Understand the user's abilities, taste, current work, constraints, and available tools.
2. Look for a specific observable behavior, frustration, or capability gap. Do not begin from a trend category.
3. When a technical capability is load-bearing, verify that it exists now.
4. Generate 1 to 3 ideas. For each, give:
   - a memorable working name;
   - the core insight;
   - why it is possible now;
   - what a focused weekend version looks like.
5. Compare against the user's existing work and obvious crowded territory. Remove recycled ideas and thin wrappers.
6. Commit to the strongest direction and explain why it clears the bar.

## The Bar

- The enabling technology is the mechanic, not a decorative feature.
- The idea comes from a concrete gap, not a vague market trend.
- One capable person can build a meaningful first version.
- The first version creates evidence instead of simulating a full company.
- The pitch is restrained, specific, and free of press-release language.

## Authority Gates

A request for ideas authorizes a pitch, not file creation or implementation. Save, specify, build, commit, deploy, pay for services, or publish only when the user asks for that stage.

## Guardrails

Do not produce a giant list. Do not pitch generic verticals with AI attached. Do not mistake novelty for usefulness. If no idea clears the bar, say what is missing instead of padding the answer.

The Loop

The sequence is simple, but I do not run every skill every time.

1. Hatch finds a small number of ideas with a real gap, a reason they are possible now, and a weekend-sized first version.
2. Think Deeper challenges the frame and exposes the assumption carrying the idea.
3. Make It Weirder pushes the promising version away from the generic middle without destroying its usefulness.
4. Grill Me gives it an honest verdict and finds the weaknesses that could actually kill it.
5. Make It Real converts the surviving direction into a spec, prototype, test, draft, or other usable artifact.
6. Remove AI Slop strips out the choices that feel generated by default and makes the result belong to this particular project.

A real session might start with $hatch to find a product idea, use $think-deeper on the strongest pitch, call $make-it-weirder when the concept still feels too familiar, and then ask $grill-me whether it is worth continuing. If it survives, $make-it-real produces the first buildable artifact. $remove-ai-slop is the finishing pass, not a substitute for having a strong idea.

Why Skill Files

A skill is just a Markdown file that gives the agent a reusable behavioral contract. There is no magic phrase here and no claim that a skill makes the model more intelligent. The value is that the job becomes explicit.

Instead of hoping the model understands what I mean by "be more creative" or "give me honest feedback," the file defines what to inspect, what to preserve, what the answer should contain, and what bad behavior to avoid. The difference is consistency. I can invoke the same mode in a new task without rebuilding the working relationship from scratch.

Think Deeper has its own case study with a before-and-after example of how one of these files changes a product idea.

Install

For Codex, put a downloaded file at .agents/skills/<skill-name>/SKILL.md in your project, then mention the skill by name in a prompt. Codex can also select a skill when the request matches its frontmatter description.

For Claude, use the same SKILL.md in your skill setup or project context and ask Claude to apply it. These public versions are intentionally portable: one file each, no scripts, no private paths, and no missing reference files.

Start with one. The useful part is not collecting skill files. It is noticing a repeated failure mode in your own work and giving the agent a better job for that moment.