Use when a Codex thread or local session is slow because prior turns contain heavy image, screenshot, or base64 payloads. Helps locate the session JSONL, back it up, remove only embedded image blobs, preserve conversation text, and validate the cleaned file.
Use when Codex is adding, editing, selecting, reviewing, or explaining tests in any repository, including mocks, fixtures, snapshots, CI validation, regression coverage, TDD/red-green proof, or readiness claims. Enforces behavior-first test design so tests catch real regressions instead of merely passing the current implementation.
Produce a compact, complete context dump of the current session directly in chat. Use when the user wants the current conversation summarized for reference, asks for a context dump, says the chat is bloated or slow, or wants a concise session state without creating files or migrating to a new chat.
Produce a copyable handoff prompt so a fresh agent can continue the current conversation in a new chat. Use when the user wants to continue elsewhere, migrate sessions, recover performance from a bloated chat, or prepare a complete next-session prompt with goals, completed work, remaining work, current state, files, repos, URLs, branches, commands, blockers, and suggested skills.
Run Codex code review as a closeout check for local dirty changes, branch/PR diffs, or a single commit. Use when the user asks for Codex review, autoreview, a second-model review, review after fixes, or when non-trivial code edits need a final advisory pass before commit, push, ship, or PR update.
Consult the AI SDK Playground as a frontier-model oracle for hard engineering, code review, architecture, PR, debugging, product, thesis, proof-check, and adversarial reasoning decisions. Use when Codex is uncertain, the problem is high-impact or subtle, a second frontier opinion would materially reduce risk, the user asks to proof-check or stress-test a plan or implementation with the playground, or competing approaches need adversarial analysis before action.
Autonomous Codex run-to-completion workflow for low quota/cota situations. Use when the user asks for an uninterrupted agent run, says "nao me peca confirmacao", "nao trave", "nao pare", "rode ate o final", or wants Codex to self-manage terminal waits, sleep, polling, long-running commands, tests, builds, and verification instead of stopping for avoidable questions.