Shared-core pre-edit.py branches on Claude-native tool names, bypassing the seam

Codex multi-host support (ADR-0011). Code under review is on branch feat/codex-support-m0, not main. ca-codex ships BETA; these are blocking-severity for the affected code but nothing is merged to the default branch yet.

Severity: low | Confidence: 0.75 | Group: pre-edit-native-names

Where:

• core/pysrc/pre-edit.py:53-56
• core/pysrc/pre-edit.py:86-92
• core/pysrc/pre-edit.py:119-122
• core/pysrc/hostapi.py:108-111

Evidence / impact:

• (architecture-008) hostapi.py establishes canonical tool categories ('EXEC|WRITE|EDIT|READ|OTHER') as the seam every shared entry should reason through, and pre-edit.py imports get_host()/normalize_tool_input — yet its guard logic branches on raw native names: if tool == "MultiEdit" (lines 53, 119), if tool == "NotebookEdit" (line 86). The escape is rationalized in a comment ('Codex does not register this entry
• (architecture-008 impact) Latent third-host bug and pattern drift: any future host that registers this entry (or a Codex change routing edits through it) gets guard semantics keyed to another host's tool names — MultiEdit-batch and NotebookEdit refusals silently never fire, weakening H-05/H-18 on that host with no signal. It

Recommendation:

Route pre-edit.py's MultiEdit/NotebookEdit branches through normalize_tool categories so a future host that registers this entry cannot silently lose H-05/H-18.

Acceptance criteria:

• No shared-core file branches on a raw native tool name; guard decisions flow through the seam's canonical categories.