off-the-wall — Claude Code hook: feed your past corrections back to Codex
The off-the-wall hook quotes your earlier fixes into the prompt before every Codex answer. The model sees what you corrected before and stops repeating the same mistakes.
MIT ★ 1,596 updated 2026-10-29
.claude/settings.jsonWhat off-the-wall does
The plugin keeps a local record of your corrections and feeds them back word for word. Codex gets the context of what went wrong last time before it starts writing code.
It fires on four events: PreToolUse, UserPromptSubmit, SessionStart, and Stop. The PreToolUse hook has a matcher for spawn_agent|collaborationspawn_agent, so it hooks into subagent launches.
The otw report command shows which corrections you make most often. A routing rule asks the orchestrator to name a model for each subagent. No API key required, everything runs on local data. The plugin is written in Node.js, so Node must be on the system.
What it helps with
- Repeating the same fixes after every Codex run
- Remembering what exactly you corrected last time
- Finding which model mistakes you fix most
- Assigning a model per subagent through the orchestrator
Example prompts for the agent
- Run Codex and check it respects my previous corrections from the local history
- Show the
otw report— what do I fix most often? - Set up the routing rule so the orchestrator names a model for each subagent
How to install off-the-wall
Manually: put the files in place
.claude/settings.jsonThe hooks block: event (PreToolUse, PostToolUse, Stop…) → command. Personal ones go to ~/.claude/settings.json
Step by step, with commands for each platform: Claude Code hooks: what they are, where to configure them, how to block a risky command
Questions about off-the-wall
How do I install off-the-wall in Claude Code?
The hook goes into .claude/settings.json. Add the hooks section from the plugin README: each event (SessionStart, UserPromptSubmit, Stop, PostCompact, PreToolUse) runs node "$PLUGIN_ROOT/src/hook.mjs" codex.
What does off-the-wall do?
It quotes your past corrections back to Codex before every answer. It also provides the otw report on your most frequent fixes and a routing rule for picking subagent models.
Do I need an API key?
No. The plugin works only with your local correction history. You do need Node.js to run the hooks.
Source text
codex.json · MIT
Hooks bundle shipped with the off-the-wall plugin for Claude Code.
Quotes your own past corrections back to Codex, word for word, before each answer; otw report shows what you correct most from your local history, and a routing rule asks the orchestrator to name a model for each subagent. No API key.
Triggers on: PreToolUse, UserPromptSubmit, SessionStart, Stop.
```json
{
"hooks": {
"SessionStart": [
{
"hooks": [
{
"type": "command",
"command": "node \"$PLUGIN_ROOT/src/hook.mjs\" codex",
"timeout": 30,
"commandWindows": "node \"%PLUGIN_ROOT%\\\\src\\\\hook.mjs\" codex"
}
]
}
],
"UserPromptSubmit": [
{
"hooks": [
{
"type": "command",
"command": "node \"$PLUGIN_ROOT/src/hook.mjs\" codex",
"timeout": 30,
"commandWindows": "node \"%PLUGIN_ROOT%\\\\src\\\\hook.mjs\" codex"
}
]
}
],
"Stop": [
{
"hooks": [
{
"type": "command",
"command": "node \"$PLUGIN_ROOT/src/hook.mjs\" codex",
"timeout": 30,
"commandWindows": "node \"%PLUGIN_ROOT%\\\\src\\\\hook.mjs\" codex"
}
]
}
],
"PostCompact": [
{
"hooks": [
{
"type": "command",
"command": "node \"$PLUGIN_ROOT/src/hook.mjs\" codex",
"timeout": 30,
"commandWindows": "node \"%PLUGIN_ROOT%\\\\src\\\\hook.mjs\" codex"
}
]
}
],
"PreToolUse": [
{
"matcher": "spawn_agent|collaborationspawn_agent",
"hooks": [
{
"type": "command",
"command": "node \"$PLUGIN_ROOT/src/hook.mjs\" codex",
"timeout": 30,
"commandWindows": "node \"%PLUGIN_ROOT%\\\\src\\\\hook.mjs\" codex"
}
]
}
]
}
}
```
© Junebeom Seo · MIT · hashgraph-online/awesome-codex-plugins