ferrumctl

Use when you need the host codex-goalctl command to inspect or change another Codex CLI thread/subagent's persisted goal: assign, update objective/status/token budget, clear, verify state, or set a spawned subagent's goal by thread id. Do not use for your own task goal, ordinary worker-thread tasks, or built-in goal tools unless the user explicitly asks for codex-goalctl.

Use when you need the host codex-limitctl command to inspect current Codex subscription limits, daily account token activity, or local rollout-derived limit and thread activity history, or to gate work on remaining account capacity. Do not use for per-thread context usage, goal token budgets, API billing, exact account-to-thread attribution, paid or reset credits, or reset consumption.

Use when you need the host codex-readcov command to inspect transcript-recorded file-read actions for a Codex CLI thread or rollout JSONL file: read counts, snapshot/delta windows, path filtering, paths-only output, or Unix composition of read path lists. Do not use for test coverage, file search, summarizing file contents, operating-system audit logs, or proving that commands succeeded or every file access was observed.

Use when the useful handle is a Codex thread id or persisted thread state is needed: use the host codex-threadctl command through app-server to discover or search stored sessions and spawned threads unavailable through native handles, inspect activity, context, retained messages, or running terminal processes, start input on an idle thread, steer one expected active turn, resume a persisted thread, interrupt one exact turn, or terminate one exact terminal process. Do not use for ordinary native subagent messaging, waiting, or result retrieval while this session owns the live handle; future or conditional wakes; goal editing; file-read coverage; terminal keystroke injection; or agent spawning.

Use when input must be delivered to a Codex thread after a later time, goal, turn-completion, or host condition, or when durable wake jobs and runners must be managed. Also covers synchronous goal or turn waits for scripts and thread-id-only controllers. Do not use to monitor a terminal process this turn already owns; use native process polling. Do not use for immediate input, thread inspection or interruption, goal editing, read coverage, terminal injection, agent spawning, or jobs you do not own.