Shared agent skills for Codex and Claude Code. One tree, both agents.

Codex and Claude Code use the same skill format: a directory containing a SKILL.md whose frontmatter carries name and description. Nothing in this repo needs reformatting for either agent.

So skills/ is shared verbatim and symlinked into both agent directories. There is deliberately no codex/ or claude/ copy — a second copy is how skills drift apart, and it is what this layout exists to prevent.

codex_skills/
  install.sh
  skills/
    robot-ros/
      SKILL.md            <- read by BOTH agents
      agents/openai.yaml  <- Codex-only metadata; Claude ignores it
      references/         <- optional bundled docs
      scripts/            <- optional bundled scripts

~/.codex/skills/robot-ros   -> <repo>/skills/robot-ros
~/.claude/skills/robot-ros  -> <repo>/skills/robot-ros

agents/openai.yaml is Codex interface metadata. Claude Code ignores unknown subdirectories, so it costs nothing to keep it alongside the shared SKILL.md.

git clone [email protected]:Idate96/codex_skills.git ~/git/codex_skills
cd ~/git/codex_skills
./install.sh

cd ~/git/codex_skills && git pull && ./install.sh

install.sh is idempotent. Because the agent directories hold symlinks rather than copies, a plain git pull already updates every existing skill — re-running install.sh is only needed to pick up newly added skills and to prune ones that were deleted upstream. Running it always is harmless.

Restart Codex / Claude Code afterwards so they re-scan the skill directories.

Useful flags:

mkdir -p skills/my-skill
cat > skills/my-skill/SKILL.md <<'MD'
---
name: my-skill
description: What it does, and when an agent should reach for it.
---

# My Skill
...
MD
./install.sh          # links it into both agents
git add -A && git commit -m "Add my-skill" && git push

description is what each agent matches against to decide whether to load the skill, so write it as what it does + when to use it, not just a title.

install.sh skips any directory without a SKILL.md, and never overwrites a real directory sitting in an agent's skills folder — if it reports KEEP, move that directory into this repo first, then re-run.

About 23 skills hard-code absolute paths like /home/lorenzo/codex_skills/skills/chrome-cdp/scripts/cdp inside their SKILL.md and scripts. install.sh therefore also creates:

~/codex_skills -> <repo>

so those paths keep resolving wherever the repo is checked out. There is still exactly one real copy of every skill.

Portability caveat: those paths embed /home/lorenzo, so they only work on a machine with that home directory. On a host with a different username they need rewriting to a relative or $HOME-based form.

• Only directories containing SKILL.md are linked.
• Broken links left behind by deleted skills are pruned automatically.
• Real (non-symlink) directories in the agent folders are never touched.
• install.sh will not replace a real ~/codex_skills directory; move an old clone aside first, then re-run.