Expose the read-only memory Skill to Claude in Step 06

[miniLV]

This was generated by AI during triage.

What to build

Finish Step 06 so users can install and manage the read-only engineering memory skill for Codex, Claude, or both independently.

The local linking and status implementation already understands Codex and Claude targets. Complete the user-facing flow and behavioral coverage so Claude is a real selectable target rather than a hidden backend capability.

• Expose separate Codex and Claude installation targets in the setup interface.
• Support Codex-only, Claude-only, and combined installations.
• Show installed, missing, stale, and invalid-link status for each target.
• Keep repeated application idempotent.
• Remove only repository-managed links for targets the user explicitly deselects.
• Preserve unrelated user-managed skills and files.
• Keep query-skill installation independent from scheduled-writer selection.

Acceptance criteria

• Codex-only, Claude-only, and combined selections are visible and usable from Step 06.
• Each selected target reports missing, valid, stale, or incorrect-link status independently.
• Reapplying a valid selection leaves existing links unchanged.
• Deselecting one target does not damage the other target.
• Cleanup preserves unrelated and user-managed skills.
• Installing the Claude target does not change or enable the scheduled writer.
• Isolated-home tests cover both targets, idempotency, selective pruning, and unrelated-file preservation.
• User-facing copy explains that this skill is read-only.

Blocked by

None — can start immediately.

[miniLV]

This was generated by AI during triage.

Agent Brief

Category: enhancement
Summary: Finish the existing multi-target Skill backend by exposing Claude in Step 06 and locking behavior with tests.

Current behavior:
The local linking and status implementation already understands Codex and Claude targets, including selective pruning and preservation of unrelated files. The setup interface only presents Codex, filters status to Codex, and resets the Claude selection; automated coverage primarily exercises Codex.

Desired behavior:
Step 06 visibly supports Codex-only, Claude-only, and combined read-only Skill targets. Each target has independent status and safe lifecycle behavior. Selecting Claude never enables or changes the scheduled writer.

Key interfaces:

• Memory Skill target selection — Codex and Claude are independent choices.
• Target status mapping — missing, valid, stale, or incorrect link per agent.
• Managed-link lifecycle — idempotent installation and selective pruning without deleting unrelated files.

Acceptance criteria:

• Step 06 exposes both target choices and supports all three non-empty combinations.
• UI status and action details include every selected target.
• Reapplying valid links is idempotent.
• Deselecting one target preserves the other target.
• Unrelated or user-managed files are never deleted.
• Claude target selection does not alter scheduled-writer state.
• Isolated-home tests cover both agents and selective pruning.

Out of scope:

• Claude scheduled writer support.
• Writer ownership or leases.
• Replacing or redesigning the query behavior.