Module splice 

The byte-level model splice.

The request body is never deserialized: the splice finds the top-level "model" key, optionally replaces its string value, and fixes Content-Length. Everything else - tool schemas, betas, context_management - passes through byte for byte, which is the whole reason the gateway forwards instead of re-encoding.

Enums

SpliceError

Why a body could not be spliced. Both are loud by design: a splice that silently forwarded the wrong model would be worse than a failure naming the session.

Functions

find_top_level_model 🔒

Locate the top-level "model" string value: its byte span (excluding the quotes) and its content. Scans with a string/escape state machine so a "model" that is only a nested key or a value elsewhere is skipped.

read_string 🔒

Read a JSON string starting at start (the opening quote). Returns the raw content bytes and the index just past the closing quote.

skip_ws 🔒

splice_model

Extract the top-level model string and, when replacement is Some, rewrite it in place. Returns the (possibly new) body and the extracted model. Nested "model" keys are never touched: only a string key at brace depth 1 counts.

unescape 🔒

Model names are plain ASCII slugs in practice; unescape the two escapes that could appear around them and reject exotic ones rather than misreporting the name.

Type Aliases

ModelSpan 🔒

The located top-level model: its byte span (excluding the quotes) and its content.