[codex-validation] Help validate FH under OpenAI Codex — open call (beta-removal gate)

[chrono-meta]

What this is

forge-harness (FH) ships a Codex-compatible (beta) path: you can run FH governance, skills, and agents through OpenAI Codex (codex exec) instead of Claude Code, via the runtime adapters in @chrono-meta/fh-gate.

The beta label stays until external (non-author) Codex users confirm it works. Author runs don't count by design. This issue is the open call to close that gate.

Why your run matters

Beta-removal conditions (from docs/codex-compat.md):

• Known-limitation list published
• 5+ externally validated M1 skill runs (not the FH author)
• At least 1 external Codex user confirms the methodology reproduces
• README badge → Codex-compatible (un-beta), blocked on the two above

How to participate (~5 min)

# 1. Have the Codex CLI installed + authenticated (npm i -g @openai/codex)
# 2. Run an M1-tier skill through the Codex backend:
FH_BACKEND=codex npx --package @chrono-meta/fh-gate fh-run \
  --skill asset-placement-gate --file path/to/some/SKILL.md

# or the governance gate on a diff:
FH_BACKEND=codex npx --package @chrono-meta/fh-gate fh-gate

M1-tier skills (text-reasoning, no Claude-only tooling) are the best validation targets — e.g. asset-placement-gate, source-grounding-audit, steel-quench. See docs/codex-compat.md for the tier list and known limitations.

What to report (comment below)

1. Codex CLI version + model used
2. Which skill/command you ran
3. Did the output match the skill's documented contract? (paste the verdict block)
4. Anything that broke, was confusing, or diverged from the docs

A short "ran X, got Y, matched/didn't" is enough. Five of these + one "yes it reproduces" and the beta badge comes off. Thanks for kicking the tires. 🛠️