dispatchSubtasks groups items into batches per the configured strategy, then processes batches sequentially while running files within each batch conc

--- helpers ---

newQuietHandle silences stdout when outputFormat=="json" or audience=="agent"; otherwise the returned handle is a no-op restorer.

sanitizeReason is the single, best-effort redaction+truncation floor applied to every reason the builder stores, so no caller path can write an unreda

AnyToAttr converts an arbitrary value to an OTel attribute.KeyValue.

CommentsForPath returns a copy of comments whose Path matches the given path.

CompletionsWithCtx sends a chat completion request with context support.

DiscoverRepos walks the sessions root and returns one entry per subdirectory.

ErrorEvent emits an error event with error status.

Finalize performs the hard-validated close of the run and returns the immutable manifest. In order it: sweeps any item still selected into failed (col

IsPathExcluded returns true when relPath matches any of the supplied gitignore-style patterns or any default excluded directory prefix (see ExcludedDi

RecordLLMResult sets the outcome of an LLM request on the span.

RecordUsage adds the prompt/completion/cache tokens reported by an LLM response to the runner's aggregate counters. Used by callers (plan phase in age

ResolveInput freezes this run's commit endpoints by asking git, following the input-mode matrix: - range: base = merge-base(from,to); head = the

SetPendingFailureCause records the classification Finalize must use when it sweeps items that never received a Mark*. It records no run_failure and tr

ValidateProtocol accepts the three canonical protocol names and rejects everything else.

mapResponsesResponse converts the SDK Response into the shared ChatResponse. Text output is read via the SDK's OutputText() helper (it walks all outpu

peekSession reads the first record, all review_item records (for comment count), and the last record of a JSONL file.

recordWarning adds a non-fatal warning to the agent's warning list.

retryCodesMiddleware returns an HTTP middleware that forces the SDK to retry responses whose status code is in the given set, by injecting the x-shoul

runCompression performs three-zone memory compression on the given messages, summarizing the compress zone while preserving the active zone intact. Re

validateReviewRefs rejects ref-option injection (#112): any --from/--to/ --commit value must be a real commit ref and must not start with '-'.

CanonicalPath returns an absolute path with symlinks resolved.

IsUserExcluded reports whether the given path matches any user exclude pattern.

LoadDetail returns the summary plus per-file item records for one session.

MarkReused records that the item was reused from a parent session checkpoint.

MarkWaived records that the user explicitly waived this selected item, which must still be in the pre-terminal (selected) state. reason is required an

--- Factory --- NewLLMClient creates the appropriate client based on the resolved endpoint protocol. protocol dispatch (canonical names from protocol.

NewProvider creates a Provider for range mode: from..to (via merge-base).

ParseComments extracts LlmComment entries from tool call arguments without writing to the Collector. Returns parsed comments and an error message (emp

ParseHunks parses raw unified diff text for a single file into a slice of Hunks. Lines before the first @@ header (file-level headers like "diff --git

RecordReviewItemReused records that this run reused a checkpoint from another session.

RunManifest returns the frozen v1 coverage result for review runs. Scan remains legacy and returns nil.

StripMarkdownFences removes ```json and ``` wrappers some models add around structured outputs. Exposed so callers (e.g. agent's review-filter post-st

applyCLIExcludes appends user-supplied --exclude patterns (already split into a []string) onto cc.FileFilter.Exclude. Creates the FileFilter if none w

executeReviewFilter runs the REVIEW_FILTER_TASK to remove comments that are provably incorrect based solely on the diff. Errors are logged and silentl

extractSideLines extracts one side of the diff from a hunk. When newSide is true, returns context+added lines with new-file line numbers. When newSide

groupBatches partitions items according to strategy, then slices each natural group into BatchSize-sized chunks (when size > 0). Within a batch the in

initBareRepo creates an empty (unborn) git repository with identity configured but no commits, returning the repo dir.

initRepoWithChange creates a real git repository with one committed file and an uncommitted working-tree modification, returning the repo dir. There i

loadLLMRuntime loads tool defs from toolConfigPath, reads the app config from the user's default config path (applying the configured language to tpl

looksLikeFilePath returns true when s is likely a file path (not inline content). Heuristic: multi-line text is always inline; single-line text withou

newCustomModelTUI builds a modelTUIModel backed by a custom provider with a writable config path, so add/delete/persist operations exercise the save p

parseBpeData parses the base64-encoded BPE data format. Each line: <base64-token> <rank>

runConfig provides test compatibility: dispatches config subcommands through a fresh cobra command tree that mirrors the production config command, so

运行任意参数，流式回调日志，结束返回 stdout 全文。退出码非 0 时 reject，并带上 CLI 报错文本。

tryApplyPendingCompression checks whether a background compression has completed and swaps the rebuilt messages into place. Returns true if applied.