feat(routing): align guided setup architecture and skill

[ss-o]

Summary

• add ADR-0029 for the guided setup configuration topology and short setup.zsh entrypoint
• narrowly supersede ADR-0025 only for configuration topology and startup integration while retaining its planner-first decision
• route both ADRs to their implementation surfaces and update the canonical zi-install skill
• regenerate the decision index and extend routing validation

Instruction impact review

1. Classification: architecture decision, scoped guidance routing, and canonical skill guidance.
2. Consumers: Codex, Claude Code, Copilot, Gemini CLI, humans implementing the guided setup planner, and users following the canonical installation skill.
3. Canonical owners: ADR-0025 remains canonical for planner-first execution. ADR-0029 owns configuration topology and the generated setup.zsh entrypoint. .github/skills/zi-install/SKILL.md owns operational installation guidance.
4. Duplication: ADR-0029 references rather than rewrites ADR-0025. Selector routes point to canonical decision files, and the delivered skill is generated from the organization copy.
5. Manifests: public routes retain ADR-0025 and add ADR-0029 for architecture and planner implementation paths. The decision index is regenerated.
6. Delivery: routes are required for supported runtimes; the installer skill documents standalone same-ref asset retrieval and the short managed .zshrc block.
7. Validation: decision validation, the targeted route test, all 17 ADR tests, Trunk, and diff checks passed.

Traceability

Refs z-shell/src#208 and z-shell/src#221.

[ss-o]

Fallback review under ADR-0026: Copilot request not registered on a4613c7

No findings.

Reviewed the complete three-file diff against AGENTS.md, the instruction-update runbook, ADR-0025, and the canonical code-review guidance. The implementation keeps one canonical decision file, constrains the new route to the four planner paths, rejects exact duplicate selector sets, and requires shared-file kind, authority, and review-owner metadata to agree.

Evidence: clean-snapshot policy validation passed; all 92 policy tests and 17 decision-record tests passed; Trunk passed; hosted Agent Instruction Validation, commit policy, and CodeQL checks passed.

Residual evidence gap: the new route was not independently exercised through each supported runtime discovery mechanism. Runtime-generated workspace carriers were excluded from the standalone-clone validation because they are not repository content.