Repository navigation
The rest of the Codex and Grok parity audit, checked against live Codex 0.169 and Grok 1.0.77 sessions.
- Fix (Grok pruning):
optimize capabilities --grok --applynever took effect in 0.19.3. Grok ignores a project server entry with no command, so the write was always rolled back. It now writes a stub that never runs and copies nothing from your server's definition;grok mcp listshows the server off and a headless session does not start it. In your home directory, where the project file is your own user config,--applypoints totrim --codex|--grok --apply. - Subagent receipts for Codex and Grok, in API-equivalent dollars, feeding the spawn hint and
results. Codex prices the child's rollout at the stop; Grok writes a subagent's usage just after its stop hook, so its receipt is logged by the next hook. - Fat-tool notice and limit check, and starting-context alert, for Codex and Grok (
install --alerts); Codex limit notifications from a throttled Stop hook (install --guard). trim --codex/--grokskip subagent sessions; Codex reports its measured first-call prompt.reportcounts Codex and Grok subagent threads under the session that spawned them.- Codex price rows (
gpt-6.1-sol,gpt-5.6-sol, from OpenAI's pricing page);codex-auto-reviewhas no list price and stays unpriced. - Fix: Codex encrypts
spawn_agentarguments, so the duplicate-spawn hold compares the task name; a large hook payload from a non-blocking pipe is read whole;doctorasks Codex which hooks it trusts; old Grok "under" outcomes in the value record are judged again.
Validation: 293 tests, lint, and the hooks run live in real Codex and Grok sessions (payloads captured). Not verified: whether Codex or Grok display a hook's message to you, and interactive Grok folder trust. No real-world token-savings claim is made.
After upgrading, re-run limitcast install --guard --alerts, restart Codex and Grok, and approve the new hooks in Codex (/hooks).