v1.59.1.0 feat: add /grok skill for xAI second opinion from Claude Code

[ljodea]

Summary

Adds /grok — the symmetric counterpart to /codex for getting an independent second opinion from Grok Build CLI while working in Claude Code.

Three modes, same contract as /codex:

• review — diff-scoped review with [P1]/[P2] findings and pass/fail gate
• challenge — adversarial failure-mode review
• consult — open Q&A with session resume via -r / -c

Relationship to #1631

#1631 (stale, open since 2025-05-21) tackles a different but related problem: when Codex is broken/unavailable during automatic outside-voice steps in /plan-*-review and /office-hours, the chain today falls through to a Claude subagent — losing cross-model independence.

#1631 proposed Codex → llm CLI (datasette/llm, e.g. llm-xai) → Claude subagent.

This PR does not supersede #1631 entirely. It adds an explicit /grok skill (like /codex) but does not wire Grok into the plan-review fallback chain. After merge, a natural follow-up is to revive #1631's intent using Grok Build CLI (grok --prompt-file, gstack-grok-probe) instead of datasette llm, reusing the probe/auth patterns from this PR.

Implementation

• grok/SKILL.md.tmpl — skill workflow (read-only via --permission-mode plan, --prompt-file, --max-turns, --effort)
• bin/gstack-grok-probe — auth (~/.grok/auth.json, $XAI_API_KEY), timeout wrapper, telemetry
• test/grok-hardening.test.ts — probe unit tests + shell invocation invariants (no --reasoning-effort)
• scripts/resolvers/review.ts — grok-review row in ## GSTACK REVIEW REPORT
• Docs: README power tools, docs/skills.md, CLAUDE.md

Usage (after merge + ./setup)

/grok review
/grok challenge security
/grok Is this migration ordering safe?

Requires grok on PATH and auth via grok login or $XAI_API_KEY.

Test plan

• bun test test/grok-hardening.test.ts (10/10 pass)
• bun run gen:skill-docs + bun run gen:skill-docs --host all
• Manual smoke: grok --prompt-file ... --permission-mode plan diff review in temp repo

[trunk-io]

Merging to main in this repository is managed by Trunk.

• To merge this pull request, check the box to the left or comment /trunk merge below.

After your PR is submitted to the merge queue, this comment will be automatically updated with its status. If the PR fails, failure details will also be posted here

[ljodea]

Follow-up: CONTRIBUTING.md compliance pass

I opened this before reading CONTRIBUTING.md and circled back to close the gaps. Pushed one corrective commit (c1fada2) — no history rewrite, branch was already up to date with main:

• AGENTS.md — registered /grok in the skill table (the skill-validation doc-inventory cross-check requires every skill with a SKILL.md.tmpl to appear in both AGENTS.md and docs/skills.md; docs/skills.md already had it, AGENTS.md didn't).
• Golden snapshots — refreshed test/fixtures/golden/{claude,codex,factory}-ship-SKILL.md. The grok-review dashboard line added to the shared ship source regenerated ship/SKILL.md, so the host-config golden-file regression tests needed the snapshots updated (a clean 2-line change per host).
• VERSION + CHANGELOG — bumped VERSION to 1.59.1.0 (next free slot per gstack-next-version; v1.59.0.0 feat: /land skill + trunk.io merge-queue integration #1815 holds 1.59.0.0) and added a CHANGELOG entry, to match the per-PR version convention. Happy to re-bump if the queue moves before merge, or to drop it if you'd rather assign at merge.

Verification (local):

• bun run gen:skill-docs --host all → clean git diff across Claude / Codex / Factory hosts, so skill-docs.yml should pass.
• bun test → green except 5 pre-existing gstack-gbrain-detect failures that fail identically on origin/main (my local box has no gbrain binary; status 127). They're unrelated to this PR.

No checks have run yet on the branch — looks like it's waiting on a first-time-contributor "Approve and run workflows." Could you kick that off when you get a chance? Thanks, and sorry for the process miss on the first pass.

[ljodea]

Superseded by #2241

Closing this PR in favor of a clean re-land.

Why: The original was useful product-wise (and I've been using the fork skill daily), but it was opened before reading CONTRIBUTING fully, then drifted — VERSION/title behind the release queue, merge conflicts against main, deferred host-exclusion issues. Not a rejection of Grok support; maintainers already noted on #2028 that a /grok skill (this PR) is orthogonal to Grok-as-host.

What #2241 keeps: /grok three-mode skill, gstack-grok-probe, hardening tests, review-report row, docs inventory, non-Claude host skip parity with /codex, VERSION 1.62.0.0 on current main.

What stays out of both PRs: plan-review fallback chain (#1631), host registration (#2028).

Thanks for any time spent looking at this one — the better packaging is in #2241.

[ljodea]

Superseded by #2241 (clean re-land on current main).