fix(post-tool): scope Bash marker clear to the command's own files (#83)

[Cannon07]

Summary

Fixes #83. When an agent issues multiple separate Bash commands that touch files (several rm deletes, shell writes), accepting the first command's PostToolUse cleared the neo-tree markers for all of them — the remaining pending commands were left unmarked.

The Bash post-tool branch cleared markers by status, globally:

changes.clear_by_statuses({ "deleted", "bash_modified", "bash_created" })

That avoided clobbering modified markers from concurrent Edit/Write, but it didn't stop concurrent Bash commands from clobbering each other.

Fix

Re-run the deterministic shell_detect.detect() on the post payload and changes.clear() only that command's rm_paths / write_paths — mirroring how the ApplyPatch branch closes specific files via patch_paths. Scoped clear, no global wipe; helps every backend.

• Concurrent pending Bash commands keep their markers until each is accepted.
• Concurrent Edit/Write modified / created markers stay untouched — those paths never appear in a shell command's detection.
• Removed the now-orphaned changes.clear_by_statuses helper (it only backed the global sweep, no caller/test remained).

Tests

Added regression specs in post_tool_handle_spec.lua:

• Two pending Bash deletes → accept the first → only its marker clears, the second survives.
• Same for shell writes (bash_created / bash_modified).
• Updated a pre-existing test that encoded the old global-clear behaviour (its command touched only a.txt but asserted b.txt cleared too).

Full plenary suite green — all 9 spec files, 0 failures / 0 errors.

🤖 Generated with Claude Code