- Add REVIEW_FILTER_TASK template and LLM conversation config
- Implement executeReviewFilter to post-process comments per file
- Add CommentsForPath and RemoveByPathAndIndices to CommentCollector
- Apply timeout from template config to review filter LLM call
- Await CommentWorkerPool before running filter to prevent race condition
- Include raw response preview in filter JSON parse error logs
- Propagate json.Unmarshal error in buildAnthropicParams when parsing
tool call arguments, preventing silent empty-map fallback on malformed JSON
- Return explicit error message in ParseComments when comments JSON
string fails to unmarshal, instead of silently ignoring
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Replace os.ReadFile + strings.Split with bufio.Reader streaming in
FileReadProvider, so only the requested window (up to 500 lines) is kept
in memory instead of the entire file. Adds Runner.Stream for semaphore-
aware stdout piping, FileReader.ReadLines for line-windowed reading, and
validates invalid line ranges. Reduces per-goroutine memory from ~8MB to
~44KB for large files under concurrent workloads.
Introduce gitcmd.Runner with channel-based semaphore to cap concurrent
git subprocesses (default 16, configurable via --max-git-procs). Route
all tool-layer and diff-layer git calls through the shared runner.
Also fix diff.Provider.runGit lacking context.Context — now the full
chain (GetDiff → MergeBase → ParseDiffText → finalizeDiff) propagates
the caller's context for proper cancellation and timeout support.
Registry was a plain map alias and DiffMap a shared map[string]string,
both relying on implicit write-before-read ordering with no compile-time
safety. Refactor Registry into an encapsulated struct with Freeze() that
panics on post-freeze writes, and wrap DiffMap as an immutable value type
with defensive copy on construction. Move DiffMap injection to before
filterDiffs to preserve the original behavior of exposing all file diffs.
Add ctx parameter to Provider.Execute interface and propagate it through
FileReader.Read, code_search.gitGrep/runGitGrep, and file_find.listGitFiles.
This ensures parent task cancellation/timeout properly terminates running
git subprocesses (grep, show, ls-files, ls-tree). Also adds 10s timeout
to file_find and fixes context.Canceled handling in code_search.
The code_search tool used -E (extended regex) by default, causing
unmatched parentheses in search patterns (e.g. "func Hello()") to
trigger "Unmatched ( or \(" fatal errors on Linux (GNU regex is strict),
while macOS (BSD regex) silently tolerated them. Since the tool
definition states the default mode is "literal text", switch to -F
(fixed string). Also capture stderr separately so real errors are
surfaced instead of being silently reported as "No matches found".
The searchText was placed after '--', causing git to interpret it as a
pathspec instead of a pattern. Use '-e' to explicitly mark the search
pattern and place the ref after it. Also replace CombinedOutput with
Output to prevent stderr from polluting result parsing.
code_search used SplitN(":", 3) to parse git grep output, but in
range/commit mode git grep prefixes each line with the ref
(ref:file:line:content). All matches silently failed Atoi on the
filepath field, returning an empty string that caused the review
session to terminate immediately.
Fix code_search to detect the ref prefix and adjust split/index
accordingly. Also replace the hard stop on empty tool results with
a retry loop (up to 3 consecutive empty rounds) so the LLM can
recover or call task_done gracefully.
file_find and finalizeDiff always read from the working tree, even in
range/commit mode where the review targets a specific git ref. This
caused inconsistent file versions compared to file_read and code_search
which correctly used git show. Fix file_find to use git ls-tree and
finalizeDiff to use git show when a ref is specified.
Re-location LLM calls were invisible in session history and their token
consumption was unaccounted for. Refactor the code_comment handling path
to record re-location requests/responses into session JSONL, accumulate
token usage into global counters, and display them in the viewer.
Also fixes: async context cancellation risk (WithoutCancel), duplicate
telemetry recording, missing timeout enforcement, missing {existing_code}
placeholder in the re-location prompt, and renames Parse to ParseComments
for clarity.
MuoDoo