dcdpr / jp / 28930272646

69%

Build Type

push

github

Committed by web-flow

Commit Message

feat(cli, md, config): Shade tool chrome inside reasoning regions (#848)

When the assistant interleaves reasoning with tool calls, the tool
chrome (the `Calling tool …` header, arguments, progress line, and
result) now carries the reasoning background so the whole span reads
as one continuous reasoning region instead of the shading flipping off
for each tool call and back on for the next reasoning block. Any
background the tool's own output sets (a syntax theme, hyperlink text)
is preserved; the region fill only backs spans left at the terminal
default.

A tool call continues the region when the chat response immediately
before it was reasoning; it ends at the first message content. The
boundary is transparent for chrome that isn't shown at all (per-tool
`hidden`, global `style.tool_call.show = false`, or JSON output), so an
invisible tool between two reasoning blocks doesn't break the region.
`jp conversation print` now applies the same visibility predicate as
the live path, which also fixes a pre-existing bug where replay
ignored `style.tool_call.show` and JSON format.

A new `style.reasoning.extend_across_tool_calls` setting (default
`true`) gates the extension; setting it to `false` restores the
previous per-block behavior. It has no effect when
`style.reasoning.background` is unset.

`jp_md` gains a public `ShadedWriter`/`shade` that maintains a
default-background invariant across a byte stream, including
carriage-return rewrites and `\x1b[K` erases, and `AnsiState` now
parses compound SGR parameters and OSC (hyperlink) sequences as whole
units so shading doesn't corrupt combined attributes or split a URL.

See RFD 088 for the full design.

Signed-off-by: Jean Mertz <git@jeanmertz.com>

Coverage Stats

320 of 345 new or added lines in 10 files covered. (92.75%)

1 existing line in 1 file now uncovered.

39052 of 57144 relevant lines covered (68.34%)

667.87 hits per line