\S+ was greedily consuming trailing punctuation (. ! , ;) into the URL
match, causing new URL() to mutate clean ASCII URLs — e.g.
"see https://example.com." became "see https://example.com./".
Fix: strip trailing sentence punctuation before parsing, extract and
replace only the hostname (not href) to avoid adding trailing slashes to
bare origins, and restore the stripped tail afterward.
Move Kilo-specific URL normalization logic out of shared upstream files
and into a mirror file under src/kilocode/, with kilocode_change markers
at each call site per the fork merge minimization convention.
Prevent homograph attacks where visually identical Unicode characters
(e.g. Cyrillic а vs Latin a) could trick users into approving malicious
URLs in bash and webfetch permission prompts.
* fix(tool): clarify semantic_search returns snippets not file paths
The first description bullet read "Find files most relevant to the
search query using semantic search.", which led an AI agent to tell a
user that the tool returns file paths only. The tool actually returns
matching content with file paths, line ranges, and relevance scores.
Updates the first two description bullets to name the output shape
directly. No runtime change.
Closes#9767.
* test(cli): cover semantic_search description
---------
Co-authored-by: marius-kilocode <marius@kilocode.ai>
This flag currently does nothing. Presumably, it was recently added by
OpenCode upstream, but since Kilo already has the `--auto` flag, the
code that would handle `--dangerously-skip-permissions` is wired into
`--auto` instead.
Removing this useless flag avoids confusion for users.
When the consumer (readLines) hits the line/byte cap and destroys the
PassThrough, the underlying createReadStream had no link back and would
keep reading chunks to EOF in the background, defeating the early-exit
optimisation for large files.
Optimistically stream the file as UTF-8 -- the common case -- using a
fatal-mode TextDecoder so the read tool can stop pulling bytes from
disk once the line / 50KB byte cap is hit. Only fall back to a
full-buffer iconv decode when the bytes turn out not to be valid UTF-8.
The streaming + retry logic lives in a new kilo helper
(packages/opencode/src/kilocode/text-stream.ts) so the read tool's
`lines` function stays close to upstream OpenCode shape.
The existing apply_patch encoding test only checked final file bytes,
which were correct because Patch.deriveNewContentsFromChunks does its
own encoding-aware read. The diff and additions/deletions counts
returned to the user (and shown in the permission prompt) were
nonetheless garbled because the surrounding tool reused a hard-coded
UTF-8 decoder. Pin the diff metadata to catch that regression.
Bom.readFile uses a hard-coded UTF-8 TextDecoder, so the oldContent
captured for the diff, additions/deletions counts, and permission-prompt
metadata was mojibake for any non-UTF-8 file. The actual bytes written
were correct because deriveNewContentsFromChunks already does
encoding-aware reads, but the user-visible diff did not reflect the real
file contents. Switch to EncodedIO.read to match.
Kirill Kalishev