⏳ This skill is pending AI review.

Scores will appear once the review pipeline completes.

version unknown

dratamer

@5o1⭐ 0 stars

Hand an execution task to OpenAI Codex CLI running in the background and supervise it. Claude designs the work and writes the task; Codex does the execution and, when delegated, the research. Claude watches Codex's log with an event-driven monitor (each agent message, failure-specific alerts with a 30 s cooldown, a 15-minute idle check, process exit) and steps in by stopping Codex and resuming it with a written ruling. Use when the user asks to have Codex do the work, to "control/supervise codex", or to run a task through Codex in the background.

Choose how to use this skill

You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.

1. Native installer

This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.

Do not guess an installer command or replace an existing version without reviewing the diff.

2. Complete package recommended

Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.

Download the complete ZIP

3. Prompt-only

Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.

Need only the instruction file?

Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.

No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.

—/10

// RATINGS

⭐GitHub Stars
⭐ 0 on GitHubGitHub ↗

New / niche

🟢ProSkills Score
—
📍

Not yet listed on ClawHub or SkillsMP

// README

DraTamer

The white-haired dragon girl should be whipped hard by the orange-haired young lady until she gets to work.

An orange-haired young lady with a riding crop supervises a flustered white-haired dragon girl buried in paperwork

DraTamer is a Claude Code skill for having Claude orchestrate the OpenAI Codex CLI. Claude writes the task, Codex runs in the background, and Claude supervises from Codex's log. When Codex goes off track, Claude stops it and resumes the same session with a written ruling.

Why

Left alone, Codex tends to:

  • simplify the task: hardcoding values, skipping checks, swapping in weaker methods;
  • drift off scope: extra tooling, starting later phases, editing files it should not touch.

Watching it turn by turn is expensive, though. DraTamer keeps the human out of the loop for routine progress and wakes Claude only when something happens.

How it works

  1. Brief. Claude writes a task file with:

    • verified context;
    • a precise method;
    • critical gates, which stop the run, and non-critical gates, which are recorded and the run continues;
    • a bounded self-fix budget;
    • the files Codex may touch;
    • the required deliverables.

  2. Launch. codex exec runs in the background with an explicit model, sandbox and network setting.

  3. Watch. scripts/codex_watch.sh tails the Codex log under Claude Code's Monitor tool. Each output line is an event that wakes Claude:

    EventMeaningThrottled
    CODEX_MSGCodex wrote an agent messageno
    CODEX_MSG [after context compaction]the first message after Codex compacted its context; re-check it against the briefno
    CODEX_IDLEnothing happened for the idle interval (default 15 min)no
    CODEX_EXITEDthe Codex process is goneno
    CODEX_NET_RETRY / CODEX_NET_FAILCodex is reconnecting to its API / gave upyes
    CODEX_ERRORanother Codex CLI erroryes
    REMOTE_SSH_FAILan ssh connect or key-auth failure in command outputyes
    NET_DOWN / NET_UPthe Codex backend became unreachable / reachableyes
    HOST_DOWN / HOST_UPWATCH_HOST became unreachable / reachableyes

    Each known failure has its own hook, so a real failure does not wait for the idle check. Throttled alerts share a cooldown (WATCH_COOLDOWN, default 30 s): at most one wakes Claude per cooldown. The rest are summarised in a single ALERT_SUPPRESSED line.

  4. Intervene. Claude stops Codex, writes a ruling (root cause, corrected method, what stays unchanged), and resumes the same Codex session.

  5. Review. Claude reads the report, spot-checks the key numbers and citations, checks git status, and reports to the user.

The full procedure is in skills/dratamer/SKILL.md.

Install

Requirements:

  • Claude Code with the Monitor tool;
  • the Codex CLI, logged in;
  • bash, awk, curl, flock (util-linux) and Python 3.
git clone https://github.com/5o1/DraTamer.git
ln -s "$PWD/DraTamer/skills/dratamer" ~/.claude/skills/dratamer

Claude Code then lists dratamer among its skills. Ask Claude to "have Codex do it" or to "supervise Codex", or invoke the skill by name.

Watcher usage

WATCH_PID=<codex pid> WATCH_HOST=<host>:<port> WATCH_COOLDOWN=30 \
  bash ~/.claude/skills/dratamer/scripts/codex_watch.sh <codex-log> [idle-seconds]
  • WATCH_HOST is optional. Set it when Codex works on a remote machine, so losing that machine raises HOST_DOWN.
  • WATCH_PROBE_URL overrides the backend reachability probe (default https://chatgpt.com/).
  • WATCH_PID tracks the Codex process you launched. Without it, any running codex exec (including other sessions' runs) counts as alive.

scripts/codex_digest.py <codex-log> prints the log content added since its previous call: agent messages in full, commands truncated.

License

MIT © 2026 5o1

// HOW IT'S BUILT

KEY FILES

skills/dratamer/SKILL.mdREADME.md

// REPO STATS

0 stars

// SHARE

Share on XShare on LinkedIn