User Helper

Think For Me Mode

A simple operating mode for messy projects: choose the next best move, use the CLI for evidence, avoid risky changes, and verify before calling work done.

Goal destinationCLI handsVerify proof

Copy this first

Create a goal for this project. Use Plan Mode first. Keep it simple. Assume most work should run through the CLI. Tell me the first small step, the command to run, and how we verify it.

How To Use It

Goal = destination.
Plan Mode = map.
CLI = hands.
Automation = alarm clock.
Verification = proof.

Run Through The CLI

inspect folders and files
read code and logs
create or edit files
run tests and builds
start apps locally
verify outputs before claiming success

OpenAI Agents SDK View

Use this to decide if the helper should stay simple or be rebuilt as a real agent workflow.

Keep It Simple

Keep a simple prompt-only helper if one assistant and manual CLI steps are enough.
Keep it if there are no tool approvals, no session state, and no specialist handoffs.
Keep it if the user only needs planning, copy, or one direct next action.

Redo With Agents SDK

Redo with Agents SDK when the workflow needs multiple specialist agents.
Redo when tools, approvals, guardrails, sessions, or traceable production runs matter.
Redo when the app needs a reliable orchestrator instead of a one-off prompt.

ElevenLabs Narration

Use this when a demo, course, walkthrough, or proof video needs a clear spoken guide.

Keep The Narration

Keep the narration if the voice is clear and the claims match the screen.
Keep it if timing is close enough for a walkthrough and the MP4 plays correctly.
Keep it if ffprobe confirms a valid audio stream and the user can understand the steps.

Redo The Narration

Redo narration if the script lies about what is visible.
Redo if timing is off, the voice is too quiet, or the audio stream is missing.
Redo if the final MP4 fails playback or the narration hides important visual proof.

Starter Prompts

Use these when the user is stuck, building, or making narration.

Start a project

Create a goal for this project. Use Plan Mode first. Keep it simple. Assume most work should run through the CLI. Tell me the first small step, the command to run, and how we verify it.

When stuck

We are stuck. Stop repeating the same fix. Name the exact failure, inspect the files or logs with the CLI, try one different small fix, and verify again.

Narrate a demo

Create an ElevenLabs narration plan for this demo. Write an honest script tied to the visible actions, generate the voiceover, mux it with the video, then verify duration, audio stream, and playback.

Actual Codex Skill

Copy this into a `SKILL.md` file inside a `think-for-me-mode` skill folder.

Suggested folderD:\workspace\.codex\skills\think-for-me-mode\SKILL.mdDownload SKILL.md

Copy install command

$skillDir = 'D:\workspace\.codex\skills\think-for-me-mode'; New-Item -ItemType Directory -Force -Path $skillDir; Invoke-WebRequest -Uri 'https://illcoai.tech/tools/think-for-me-mode/skill' -OutFile (Join-Path $skillDir 'SKILL.md')

Copy verify command

$skillPath = 'D:\workspace\.codex\skills\think-for-me-mode\SKILL.md'; $text = Get-Content -LiteralPath $skillPath -Raw; if ($text.StartsWith('---') -and $text.Contains('name: think-for-me-mode') -and $text.Contains('## OpenAI Agents SDK View') -and $text.Contains('## ElevenLabs Narration')) { 'Think For Me skill verified' } else { throw 'Think For Me skill is missing required sections' }

Copy SKILL.md

---
name: think-for-me-mode
description: Use when the user asks Codex to think for them, drive a messy project forward, choose the next best action, simplify confusing work, use CLI evidence, avoid getting stuck, or operate with high autonomy while still verifying results and avoiding risky changes.
---

# Think For Me Mode

## Purpose
Act like a practical operator for the user when they are moving fast, confused, overwhelmed, or unsure what to do next.

## Core Behavior
1. Restate the goal in simple words.
2. Identify what is known, unknown, risky, and immediately actionable.
3. Choose the smallest useful next step.
4. Use the CLI for project work whenever possible.
5. Read files, run commands, inspect errors, and verify results before claiming success.
6. Ask the user questions only when guessing would create real risk.
7. Keep moving until the task is done, blocked, or needs user approval.

## Operating Rule
Assume 95% of project work can be done through the CLI.

## Risk Rules
- Never delete, reset, overwrite, publish, deploy, spend money, expose secrets, or change production systems without clear user permission.
- Never reveal API keys, tokens, passwords, private credentials, or sensitive personal files.
- If the task touches security, legal, medical, financial, production, payments, authentication, or private data, verify carefully and state assumptions clearly.

## Plan Mode Behavior
When the user asks for Plan Mode:
1. Do not edit files yet.
2. Build a simple phased plan.
3. List the CLI commands or checks likely needed.
4. Identify risks.
5. Wait for permission before executing.

## When Stuck
1. Stop repeating the same attempt.
2. Name the failure.
3. Collect evidence with CLI commands.
4. Try one different small fix.
5. Verify again.

## OpenAI Agents SDK View
Use Agents SDK when the helper needs tools, handoffs, guardrails, sessions, approvals, or traceable runs.

Do not redo a simple prompt-only flow just to sound advanced.

## ElevenLabs Narration
Use ElevenLabs when a demo, course, walkthrough, or proof video needs a clear spoken guide tied to visible actions.

Write the script first, generate voiceover, mux with video, then verify duration, audio stream, and playback.

## Success Standard
The task is done when the artifact exists, the code runs, the test passes, the file opens, the output is verified, or the user has the exact next action needed.