hook-dsh-normalize-fullwidth

Family position (the @-sentence Hook @ DSH @ Normalize @ Fullwidth): a hook child of the plugin-dsh-factory service and the hook-dsh-core machinery; the sixth of the six stream normalizer siblings:

A non-manifest factory consumer: it injects ["pluginFactory"] and uses only State (cell unwrap + shared Ledger/Enabled mappings + its own fields) and Append; the config is composed by the factory's standalone Schema helper with shared: false - the minimal block, no fs/observed dead fields. It touches no files, so fs/write-intent and fs/observed never see it; it wraps the downstream result and always calls next(), so it composes with other llm/stream listeners regardless of registration order.

CJK input methods emit full-width punctuation, digits and letters - and a full-width comma or digit in a command, code block or path is invisible on screen and wrong for every parser, shell and diff. This flavor maps the whole fullwidth range to the unambiguous ASCII half-width forms before it reaches the transcript.

The transform - exactly the core's Fullwidth map (@playform/hook-dsh-core's Normalize/Fullwidth), applied per text segment through the core's ReplaceMap - the standard fullwidth-to-halfwidth mapping over the whole range:

No context rules - one character in, its half-width counterpart out. Chunk-boundary-safe: single-character substitution, no lookahead - per-chunk application can never disagree with whole-text application. The mapping values are returned through a function replacer, so they are inserted literally (no $-pattern interpretation - the table's backslash, circumflex, vertical bar and tilde values are exactly the characters a string replacement would corrupt). Replaced characters are counted per stream for the ledger line.

There is no replacement field (MAP flavor). Volatile cells commit without remounting the plugin; the factory's State builder unwraps them defensively. Example cordis.patch.yml row:

One stream, one transformation. The model emits CJK-typing artifacts - a full-width word, a full-width path punctuation run, a full-width digit; the live UI and the transcript receive the half-width ASCII forms:

Before → after - the model stream, then what reaches the transcript

Every full-width code point in the incoming text - the solidus U+FF0F, the letters U+FF55-U+FF4E, the digits U+FF11-U+FF13 and the exclamation U+FF01 - becomes its U+0021-U+007E half-width counterpart; ASCII characters that were already half-width are untouched. This is what keeps a pasted path openable and a pasted number parseable. The backslash, circumflex, vertical bar and tilde entries are exactly the characters a naive string replacement would corrupt - hence the function replacer. When a stream finishes normally with replacements made, the ledger gets the count line shown below. The same pass runs over reasoning deltas when normalizeReasoning is on, and over tool-call arguments when the example patch enables normalizeToolArguments - with the edit name exempt and raw-marker calls passing through unnormalized.

Two lines, both written through the factory's Append (the hook-dsh-normalize-fullwidth: prefix is the logger's <State.Module>:; the durable file line is [<ISO>] <message>):

The activation line is written by apply(); the count line only follows a normal stream completion and only when N > 0 (a thrown-away stream writes no ledger line).

License: MIT.