hook-dsh-normalize-spaces

Family position (the @-sentence Hook @ DSH @ Normalize @ Spaces): a hook child of the plugin-dsh-factory service and the hook-dsh-core machinery; the fourth 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.

Model output picks up exotic spaces - the no-break space from copy-pasted prose, the thin space from typographic text, the ideographic space from CJK input. Markdown renderers and shell quoting do not treat them as spaces; a command with a U+00A0 in it fails for no visible reason. This flavor makes the plain ASCII space the one that reaches the transcript.

The transform - exactly the core's Spaces class (@playform/hook-dsh-core's Normalize/Spaces), applied per text segment through the core's generic class-to-string Replace:

Exactly Unicode's Zs category minus the ASCII space - becomes the config replacement (default " ", the plain ASCII space U+0020):

No context rules - hermes-style whole-chunk replacement, no lookahead; chunk-boundary-safe. The replacement is applied with a function replacer, so a custom replacement containing $ patterns is inserted literally. Replaced characters are counted per stream for the ledger line.

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 copy-pasted typographic spacing; the live UI and the transcript receive plain ASCII spaces. In the incoming line below the three gaps are the no-break space U+00A0, the em quad U+2003 and the ideographic space U+3000, in that order:

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

Each non-ASCII space in the Zs-minus-ASCII class - U+00A0 between deploy and the, U+2003 between agent and to, U+3000 between ring and buffer in the sample above - becomes exactly one ASCII space (U+0020); ASCII spaces that were already there are untouched. This is what makes a pasted command executable again: echo a - b with a hidden no-break space between the arguments runs after the pass, where before the pass the shell saw one mangled word. When a stream finishes normally with replacements made, the ledger gets the count line shown below (note the replacement= field renders the default as a bare space). The replacement is hot-editable (say to _ for debugging), and 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-spaces: prefix is the logger's <State.Module>:; the durable file line is [<ISO>] <message>). In the activation line, replacement= is followed by the ASCII space itself - the default renders as a trailing blank:

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.