enable-1m-context

name	enable-1m-context
description	Enable or repair a 1M token context window for Codex Desktop/CLI by safely patching the local Codex model catalog and config. Use when a user asks to raise Codex context length, change 258k/272k context to 1M/1000K, fix the status bar still showing 258k, package or apply the local 1M context workaround, or restore the backed-up Codex context configuration.

Enable 1M Context

Core Behavior

Treat this as a local Codex configuration workflow, not as a guaranteed model entitlement change. Installing the skill alone does not change context length; the user must invoke the skill and approve edits to their Codex config directory.

Use scripts/enable_1m_context.cjs for the deterministic path. The script:

• Locates Codex home from CODEX_HOME, then the default user .codex directory.
• Backs up config.toml and models_cache.json.
• Creates models_1m.json with the selected model set to a 1,000,000 token context window.
• Adds or updates model_context_window and model_catalog_json in config.toml.
• Restores from backups or resets the selected model to the default-like 258k display when requested.
• Prints verification details and restart instructions.

Workflow

1. Inspect the user's current Codex model and config if available.
2. Warn that the change is local, experimental, and may depend on the user's Codex version, account, and model availability.
3. Run the script from this skill folder, requesting filesystem approval when Codex home is outside the workspace.
4. Ask the user to fully quit and reopen Codex Desktop, then create a new conversation.
5. Verify by checking the newest session log for model_context_window: 1000000 or by asking the user to inspect the status bar.

Commands

Run with defaults:

node scripts/enable_1m_context.cjs

Target a specific model:

node scripts/enable_1m_context.cjs --model gpt-5.5

Use a specific Codex home:

node scripts/enable_1m_context.cjs --codex-home "C:/Users/USER/.codex"

Preview without writing:

node scripts/enable_1m_context.cjs --dry-run

Restore the latest backup:

node scripts/enable_1m_context.cjs --restore-latest

Force the selected model back to the 258k-style default if backups are unavailable or the user wants to undo the workaround:

node scripts/enable_1m_context.cjs --reset-258k

Verification

After the user restarts Codex and opens a new conversation, inspect the newest session log under .codex/sessions and search for model_context_window. Success looks like:

model_context_window: 1000000

If the UI still shows 258k, check whether the user is viewing an old conversation. Existing conversations may keep their original context window; new conversations should use the new value.

Rollback

Prefer --restore-latest because it restores the exact config.toml and models_cache.json that existed before the script changed anything.

Use --reset-258k when no backup is available or the user explicitly wants the visible status bar to return to the old 258k-style window. This mode removes model_context_window and model_catalog_json from config.toml, sets the selected model catalog entry to context_window = 272000, max_context_window = 272000, and effective_context_window_percent = 95, then asks the user to restart Codex and open a new conversation.

Safety Notes

• Do not delete existing Codex files.
• Do not edit auth.json.
• Always keep backups.
• Use --restore-latest for exact rollback; use --reset-258k as a fallback.
• If models_cache.json is missing, tell the user to open Codex once so the model catalog is created.
• If the server rejects long-context requests despite the UI showing 1M, explain that account/model entitlement may be the limiting factor.