load-persona-memorylisted

# Load Persona Memory ## Purpose Read repo-backed persona memory (the claude-personas pattern) from any of the three supported tools. The memory repo is the source of truth. Do not mirror memory into tool-native stores (Codex `~/.codex/memories/`, etc.) and do not treat tool-generated memories as persona guidance. ## Discovery Resolve the memory repo and role before reading memory, in this order - first hit wins. 1. Read the workspace's `.agents/memory` symlink target: the target directory name is the role, the target path is the memory repo. In the memory repo itself this is the Memory Manager's self-mount (`.agents/memory -> ../memory_manager`, wired by `init-clone.sh --self`; today the tool lives at `framework/tools/init-clone.sh` in the framework repo or a memory repo created from it, and will land at `.agents/tools/init-clone.sh` once the installer ships), so `memory_manager` resolves here exactly like any other role. 2. Else read `.claude/memory` the same way (v3.1 legacy direct symlink). 3. Else enumerate candidates: every sibling directory of the workspace (same parent dir) containing a `.claude-personas/project.txt` marker. Verify each by comparing the workspace's `origin` remote against that `project.txt`. Normalize both sides to a canonical `owner/repo` before comparing: strip a leading `git@<host>:`, `ssh://git@<host>/`, or `https://<host>/` prefix and any trailing `.git`, so the scp-like SSH, scheme SSH, and HTTPS forms of the same remote compare eq