fix(architect): run Sol at xhigh

[carlcayton]

Why

Architect is the only default pstack role that should use GPT-5.6 Sol at xhigh. The other Sol routes should remain at max.

Closes #28.

Scope

• Change the Architect runner descriptor from codex:gpt-5.6-sol@max to codex:gpt-5.6-sol@xhigh.
• Preserve Architect xhigh as a fixed exception when setup parses, probes, and renders configurable per-family efforts.
• Update TypeScript and shell routing invariants so Architect must use xhigh while every other default panel retains the matrix effort.

Blast Radius

The change is limited to Architect routing, setup-pstack handling, and routing contract tests. Provider dispatch and every other role retain their existing model effort.

Verification

• quick_validate.py passed for architect and setup-pstack.
• bun run typecheck passed.
• PSTACK_STATIC_ONLY=1 bash tests/skill-collision-repro.sh passed.
• bun run test passed 154 of 155 tests. The remaining runner/run.test.ts failure, terminalizes catchable failures after reserving output paths, is unrelated and reproduces outside this patch.
• The exact Architect candidate is installed in the active pstack 1.0.5 cache.
• Native Codex Architect smoke returned ARCHITECT_SOL_XHIGH_NATIVE_OK.
• Claude-parent external Codex smoke returned ARCHITECT_SOL_XHIGH_EXTERNAL_OK, with the receipt recording gpt-5.6-sol and xhigh.

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request

[greptile-apps]

Greptile Summary

This PR changes Architect's default GPT-5.6 Sol effort from max to xhigh while retaining max for the other Sol routes.

• Updates Architect and setup skill descriptors and confirmation text.
• Expands setup probing and smoke guidance to cover both distinct Sol effort descriptors.
• Revises the static routing invariant to check the Architect-specific exception separately.

Confidence Score: 4/5

The PR appears safe to merge, with one non-blocking maintainability issue in the model-swap robustness of the static routing invariant.

The current descriptors and expected quads are consistent, but the new test couples Architect's effort override to the exact current Sol model slug and may stop enforcing the intended invariant correctly after a frontier-model update.

Files Needing Attention: tests/skill-collision-repro.sh

Important Files Changed

Filename	Overview
plugins/pstack/skills/architect/SKILL.md	Correctly changes only Architect's Sol runner descriptor from max to xhigh.
plugins/pstack/skills/setup-pstack/SKILL.md	Consistently documents, probes, and presents Architect's distinct xhigh Sol route.
tests/skill-collision-repro.sh	Enforces the new routing split, but derives the Architect expectation through a model-slug-specific substitution that will become stale on a future Sol model swap.

_{Reviews (1): Last reviewed commit: "fix(architect): run Sol at xhigh" | Re-trigger Greptile}

[greptile-apps]

Model-specific invariant derivation

The Architect expectation replaces only the exact gpt-5.6-sol@max descriptor, so a future Codex model swap makes the substitution a no-op and causes the invariant to reject a correct xhigh route or accept an incorrect canonical max route.

Suggested change

	architect_quad="$(printf '%s\n' "$canon_quad" | sed 's/codex:gpt-5\.6-sol@max/codex:gpt-5.6-sol@xhigh/')"
	architect_quad="$(printf '%s\n' "$canon_quad" | sed -E 's/(codex:[a-z0-9.-]+)@max/\1@xhigh/')"

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[carlcayton]

Closing because this was scoped incorrectly. The requested change is local to Herdr-launched Codex CLIs, not upstream pstack.