NewToolResultMessage creates a tool-role message with the given result. Uses the OpenAI Chat Completions format: role="tool" with tool_call_id and pla

ParseReviewMode returns the correct ReviewMode based on provided flag values.

PhaseEvent records a review phase completion with duration and optional error.

PrintToolCallError prints a line when a tool fails. Example: [ocr] ✘ file_read "internal/config/rules/loader.go" failed: permission denied

PrintToolCallFinished prints a line when a tool finishes successfully. Example: [ocr] ✔ file_read "internal/config/rules/loader.go" (12ms)

ProjectSummary returns the markdown project-level summary produced after all batches finish. Empty when SkipSummary is set, PROJECT_SUMMARY_TASK is ab

Quiet replaces stdout with io.Discard and returns a cleanup function. Usage: defer stdout.Quiet()() WARNING: Quiet must ONLY be called from the mai

RefValue returns the git ref that should be used for reading file contents in range or commit mode. Returns ("", false) for workspace mode.

RemoteIdentity returns a stable, credential-free identity string for the repository's "origin" remote, suitable for hashing into the run manifest's re

RuleGroupsMarkdown renders rule groups into a markdown section.

RunSplit executes a git command and returns stdout and stderr separately.

SessionsRoot returns the root directory where session JSONL files are stored.

SetExecution sets the non-secret execution provenance.

SetParentRunID links this run to the session it resumed from. It only records the direct parent; the parent's input is never copied into this run.

SetRepository sets the redacted repository identity.

StartToolSpan creates a span for a tool execution with standard attributes.

Stream acquires the semaphore, starts a git command, and passes its stdout as an io.Reader to consume. The semaphore is held for the full duration. co

ToolCalls extracts tool calls from the first choice.

ToolDefsByPhase returns the parsed tool definitions filtered by phase. planOnly=true returns only tools with plan_task:true. planOnly=false returns on

Warnings returns the warnings recorded by the LLM runner.

WriteSessionStart writes the initial session_start record.

Make a GitLab API request with retry on rate-limit and transient errors. Returns a dict:: {'success': bool, 'data': response or None,

Return the first present, non-empty field among several possible key names.

Case-insensitive header lookup. urllib normalizes response header keys to title-case (e.g. 'Retry-After'), but this defensive check also hand

addNextMessage extends the conversation with the assistant message and tool responses, applying three-zone compression at the soft (60%) and warning (

applyLanguage appends instruction to all system-role messages in conv.

beginManualTokenReplace is the manual-tab equivalent of beginAPIKeyReplace.

buildBadge renders a compact "[category · severity]" tag for a finding. It returns an empty string when neither structured field is present, so text o

buildChangeFilesExcept returns a formatted list of changed files except the given path.

buildDedupCommentsJSON renders the batch comments as a JSON list with stable c-N ids that the LLM groups by. Only fields the LLM needs to judge simila

buildFilterCommentsJSON serializes comments into a JSON array with generated IDs.

buildMessageXML serializes msgs into the <message><content> form expected by the MEMORY_COMPRESSION_TASK prompt template.

buildSummaryCommentsList renders comments as a compact path-anchored markdown list suitable for embedding in the PROJECT_SUMMARY_TASK prompt. Format:

Build the category/severity badge for a comment.

canonicalRemote reduces a git remote URL to a stable, credential-free identity string for hashing into repository.identity_sha256, so the same reposit

checkMetricErr intentionally ignores metric registration errors: telemetry is best-effort and must not interrupt the main flow.

classifyItemError maps a subtask error to a stable item failure class and a safe, generic reason. It never returns the raw error text (which may embed

copyMessages creates a shallow copy of a message slice.

countLines returns the number of lines in content. A file without a trailing newline still counts its final line. Empty input → 0.

countReviewable counts diffs that will survive all filters and are not pure deletions.

cursorOnDeletableModel reports whether the model-step cursor is on a row that can be deleted (not on "Enter custom model name...").

customAPIKeyForSave reports the API key to persist and whether the user edited the field (vs left the masked placeholder untouched).

estimateDiffCost projects token usage for reviewing the given diffs. It is the diff-path analogue of scan.estimateCost and is used for the pre-review