Struct ClaudeTranscript 

pub struct ClaudeTranscript { /* private fields */ }

A Claude Code session: the main transcript plus every subagent transcript under its subagents/ directory, folded into one summary.

Claude Code bills a subagent’s API calls to the session that spawned it and shows them in its own cost display, but writes them to a separate file, so a session that used the Agent tool reads low if only the main transcript is counted. Each subagent file is tailed like the main one and its tokens, cost, turns, tool calls and spans are added to the parent’s. A subagent may run a different model from its parent; each line is priced by the model it names, so that is handled without special casing.

Implementations

impl ClaudeTranscript

pub fn new(path: impl Into<PathBuf>) -> Self

pub fn with_prices(self, prices: &'static Table) -> Self

Price with this table instead of the process-wide one. Lets a test assert a cost without the developer’s own price file changing it.

pub fn with_spans(self, retention: SpanRetention) -> Self

Keep every span instead of the newest MAX_SPANS. See SpanRetention.

pub fn set_registry_hints(&mut self, ps: &PidSession)

Trait Implementations

impl SessionTracker for ClaudeTranscript

fn refresh(&mut self) -> Result<bool>

Ingest whatever was appended since the last call. Returns true when there is still unread data (the byte budget was exhausted).

fn summary(&self) -> &SessionSummary

fn path(&self) -> &Path

fn refresh_all(&mut self) -> Result<()>

Ingest the whole file, however many refreshes that takes. For a one-shot read such as an export; the live collector spreads a large transcript over several ticks instead.