fix: prevent codex exec stdin deadlock with </dev/null redirect

[loning]

Summary

• Fixes the codex exec / codex review deadlock that hangs every gstack outside-voice / adversarial / plan-review skill on codex CLI 0.120.0+ when invoked from a non-TTY shell (Claude Code Bash tool, background bash, CI, OpenClaw).
• Adds < /dev/null to every invocation in the source-of-truth files (`scripts/resolvers/.ts` and `.md.tmpl`), then regenerates all skill docs via `bun run gen:skill-docs --host all`.
• Fixes codex exec hangs in skill bash blocks: missing </dev/null causes 'Reading additional input from stdin' deadlock (codex CLI 0.120.0+) #971.

Why

`codex exec --help` (0.120.0):

[PROMPT]
Initial instructions for the agent. If not provided as an argument (or if - is used), instructions are read from stdin. If stdin is piped and a prompt is also provided, stdin is appended as a <stdin> block.

When skill bash blocks invoke `codex exec "$PROMPT" ...`, the parent shell's stdin is inherited. In Claude Code's Bash tool (and any background or non-TTY context), stdin is not a TTY but also not closed. Codex sees this as "stdin is piped, append it as `` block" and blocks waiting for an EOF that never comes. The positional `$PROMPT` arg is silently followed by an indefinite stdin wait.

Symptom: stderr shows `Reading additional input from stdin...`, process sits at 0% CPU, stdout empty. Skill workflow times out or hangs.

For users running gstack from a real interactive terminal, stdin is a TTY and codex 0.120.0 takes the fast path. The bug only surfaces in non-TTY contexts, which is why nobody caught it earlier in interactive testing.

Repro (against current main, codex CLI 0.120.0)

```bash
codex --version

codex-cli 0.120.0

Hangs forever (no stdin redirect):

codex exec "say hi" -s read-only > out 2> err &
sleep 10
cat err

Reading additional input from stdin...

ps -o pcpu -p $!

0.0

Works:

codex exec "say hi" -s read-only < /dev/null > out 2> err
echo $?

0

```

I hit this today running `/plan-eng-review` on a Voice Avatar design doc — first invocation sat at 0% CPU for 13 minutes, no stdout. After adding `< /dev/null` the same prompt completed in 2:25.

Fix scope (minimal, line-level only)

Source-of-truth files (the only files I hand-edited):

File	Invocations fixed
`scripts/resolvers/review.ts`	4 (office-hours, ship/review adversarial, codex review fallback, plan review)
`scripts/resolvers/design.ts`	3 (design-review-lite, design-consultation sketch, plan-design-review)
`codex/SKILL.md.tmpl`	5 (review modes + resume)
`autoplan/SKILL.md.tmpl`	4 (CEO/design/eng/devex phases)

Regenerated SKILL.md (via `bun run gen:skill-docs --host all`):
`autoplan`, `codex`, `design-consultation`, `design-review`, `office-hours`, `plan-ceo-review`, `plan-design-review`, `plan-devex-review`, `plan-eng-review`, `review`, `ship`.

Diff is symmetric: 38 insertions / 38 deletions, every change is just `+ < /dev/null ` inserted right before the existing stderr redirect (or at end-of-line for autoplan markdown blocks that had no redirect).

What I deliberately did NOT change

• `--enable web_search_cached` is left untouched. I observed that flag also seems suspicious in 0.120.0 (the first hang I hit in my repro included it, and removing it didn't stop the deadlock — only `< /dev/null` did). But since the verified bug is the stdin one, I kept this PR scoped to that single fix. If you want, a follow-up can audit web_search_cached behavior in current codex CLI versions.
• The Cursor mirror skills under `.cursor/skills/` were not regenerated (gen:skill-docs --host all does not appear to touch them, and they were already not in sync with the .tmpl files in the upstream main branch).

Verification

```bash
bun run skill:check

✅ All Claude Code generated files are fresh

✅ All OpenAI Codex CLI generated files are fresh

✅ All Factory Droid generated files are fresh

✅ All Kiro / OpenCode / Slate / Cursor / OpenClaw generated files are fresh

Total: 36 skills, 0 missing

```

End-to-end verified by re-running `/plan-eng-review` against my own design doc with the patched skills — Codex returned a clean plan review in ~2.5 minutes, full output as expected.

Test plan

• `bun run skill:check` passes (freshness for all 8 hosts)
• Manual `codex exec ... < /dev/null` repro returns 0
• End-to-end `/plan-eng-review` outside voice completes
• CI green
• Maintainer reviews diff symmetry (38/38)

🤖 Generated with Claude Code