VibeWise for T3 Code and Codex

Learn to design software while Codex implements the scope you agree on. This independent adaptation of VibeWise by Noah Kim (@nykooi1) targets Codex and T3 Code.

Experimental alpha — 0.1.0-alpha.1. Local helpers and native skill/hook discovery are tested. Full T3 conversations, educational quality, and restoration after an actual resume or compaction still need end-to-end validation. Try a disposable project first. This is not an official OpenAI or T3 product.

Features

$vibe-wise-learn: onboarding, learner-led design, explanations, and implementation checkpoints.
$vibe-wise-reset: preview and confirmed reset of learning notes, with a backup.
A read-only SessionStart hook for active projects on startup, resume, clear, and compact.
T3 Code Plan/Default instructions that respect the active mode and available tools.
A preview-first installer that preserves existing hooks and refuses to overwrite different skills.

You can ask for hints, skip a decision, pause, or explicitly request direct implementation. Existing authorization is respected without redundant confirmations. These are model instructions, not a security boundary or a guarantee of checkpoint compliance.

Requirements

Python 3.8+, available as python3; helpers use only the standard library.
Codex with native skills/hooks. Discovery checked with 0.159.3 during development and 0.162.0 for this alpha.
T3 is optional: integration source inspected at v0.0.44. Use the same Codex profile as the CLI. Full conversational testing remains pending.
Linux/macOS-style shell environment. Linux is tested; macOS, Windows, and WSL are unverified.

These are verified baselines, not the oldest compatible versions. See compatibility and limitations.

Install

Review the code before running it:

git clone https://github.com/AndrewwAM/vibe-wise-t3-codex.git
cd vibe-wise-t3-codex
git checkout v0.1.0-alpha.1
python3 -B scripts/install.py
python3 -B scripts/install.py --apply

The first command is read-only. --apply installs both skills under ~/.codex/skills or the existing CODEX_HOME, and merges the hook into hooks.json. Existing hook configuration is backed up before modification. Repeating the same installation adds no duplicates. Select another profile with --codex-dir /absolute/path/to/codex-profile.

Review the hook: open codex, run /hooks, inspect “Restoring VibeWise learning context”, and trust it if you accept its behavior. Enabled but untrusted hooks are skipped. Installation does not bypass review or change authentication, sandbox, model, or other settings. See official hook documentation.

Skills are discovered on a subsequent turn. If T3's selector is stale, start a new conversation or refresh its provider. Confirm that T3 uses the same Codex profile.

Use

Inside a project:

$vibe-wise-learn I want to build a command-line tool for organizing my notes.

Codex asks for your approach, explains unfamiliar concepts, discusses tradeoffs, and implements the scope you authorize. Say “pause learning mode” to pause. Invoking Learn again resumes existing notes without resetting them.

State lives in .vibe-wise/profile.md, progress.md, and project-map.md. Legacy .sensible-vibes/ is supported in place. The hook does not activate new projects, write notes, or copy notes into its output; it points Codex to the files. Lookup stops at Git/worktree boundaries and rejects linked state paths.

$vibe-wise-reset previews the selected state and files. After confirmation of that snapshot, it backs up notes and restarts onboarding. Application code and Git history are preserved. Backups remain in the local state directory.

In T3 Plan, inspect and discuss without writing code or notes. Switch to Default to persist state or implement the agreed step. The skill does not change T3's mode for you.

Privacy

The Python helpers make no network requests and collect no telemetry. Codex reads notes as model context, so notes and paths can reach your configured model provider. Local notes do not imply offline inference. Keep credentials, private URLs, customer data, and conversation transcripts out of them.

Ignore .vibe-wise/ and .sensible-vibes/ in every learning project's .gitignore before publishing it. Installation does not edit other projects' ignore rules. See security and data handling.

Validation

python3 -B -m unittest discover -s tests -v
python3 -B scripts/check_release.py
python3 -B scripts/smoke_app_server.py

The release check inspects Git-tracked files; stage intended files first. The native smoke requires codex and queries discovery in an isolated temporary project. It starts no model turn, persists no trust, and does not exercise a real conversational lifecycle. Unit tests execute the hook directly with synthetic events.

CI runs tests, package/privacy checks, Gitleaks, and native Codex discovery. See test results and the manual checklist and CONTRIBUTING.md.

Updates, removal, and packaging

This alpha has no automatic upgrade/uninstall command. Different installed skills cause the installer to stop. Compare and back up existing skill folders before replacing them; do not delete a different adaptation.

To remove this adaptation, inspect and remove only its SessionStart registration from your Codex profile's hooks.json, then remove its two skill folders after backing up edits. Preserve unrelated hooks, configuration, and project notes. Use /hooks to disable restoration without uninstalling.

Root plugin.json and hooks/hooks.json also support portable local packaging. Marketplace installation is not validated; the standalone installer is the tested alpha path. Use one method to avoid duplicate hooks. This GitHub release is not a submission to OpenAI's public plugin directory.

Credits and license

The learning workflow, checkpoints, and Markdown state originate in VibeWise by Noah Kim. Base: upstream 0.1.43, commit c5fc8d813ce6789ebaf631ff4dd2a69658da3f31. The original MIT license and copyright notice are preserved in the root and each installed skill. See LICENSE and ATTRIBUTION.md.

Other public adaptations include djolex999/vibe-wise-codex and yava-code/vibe-wise-mcp. This project derives from the original, not those ports, and does not claim to be the first Codex port or endorsed by their authors.