douglasmonsky

Use when the user asks about Codex token usage, model/reasoning efficiency, usage dashboards, CSV exports, or per-session/per-turn Codex usage stats from local logs.

Use when the user asks about Codex token usage, model/reasoning efficiency, usage dashboards, CSV exports, or per-session/per-turn Codex usage stats from local logs.

Use when the user wants to discuss, investigate, compare, explain, or improve Codex usage with Codex Usage Tracker API or MCP tools, including token waste, cache/context problems, allowance or limit changes, pricing confidence, dashboard evidence, and local content-index investigations.

Maintainer-only release skill for preparing, verifying, and publishing Python packages to TestPyPI or PyPI. Use when the maintainer asks to configure Trusted Publishing, prepare a Python package release, update release versions, build and check distributions, smoke-test wheels, publish to TestPyPI/PyPI, or troubleshoot PyPI release failures.

Prepare, verify, document, and risk-check a software release. Use when the user says release, ship, deploy prep, cut a release, version bump, changelog, release notes, release candidate, hotfix release, rollback plan, smoke test, or pre-release checklist.

Review and rename Codex chats/threads by inspecting conversation context, scoring current title quality, proposing clearer names, and applying user-approved changes. Use when the user asks to rename the current Codex chat, rename the last N Codex chats, rename all accessible Codex chats, clean up chat/thread titles, review chat naming quality, or generate better Codex thread names from conversation history.

Delegate-first Codex orchestration for coding, debugging, review, planning, research, migration, audit, testing, documentation, or any task that benefits from routing work to subagents. The root agent acts as dispatcher, escalation controller, synthesizer, and final senior reviewer. Delegate substantive work by default, continuously reassess delegation, use explicit model routing plus reasoning effort, support custom-agent and built-in-agent runtimes, escalate stuck work narrowly, and require root-level final review before completion. Avoid only for pure conversational micro-answers or when the user forbids subagents.