Install

macOS users can brew install:

  brew install agent-ecosystem/tap/agentminutes

For other install options, including npm, PyPI, go install, and prebuilt binaries, refer to Installation.

Convert a transcript

To parse a native transcript into a normalized session record, use the convert command (the harness is auto-detected):

  agentminutes convert ~/.claude/projects/my-project/some-session.jsonl

The default output is one JSON document: session metadata, the ordered events, token totals, and a report accounting for everything that did not become an event.

Find your transcripts

Harnesses write transcripts to global, harness-owned locations. To find the sessions you care about without parsing everything, use the sessions command:

  agentminutes sessions --cwd ~/runs/exp-42 | xargs -n1 agentminutes convert

This scans every supported harness’s default root, reads each transcript’s identity cheaply, and filters by the keys harnesses actually record. See Session Discovery.

Pick up subagent transcripts

A session that delegated work to subagents is more than one file: Claude Code and Codex both write each subagent conversation as its own transcript, with its own events and its own token totals. Discovery treats them as one session, so a session-ID lookup lists every file the task touched, ready to convert together:

  agentminutes sessions --harness claude-code --session-id "$SESSION" |
  xargs -n1 agentminutes convert

A parent transcript’s totals cover only its own context, so skipping the subagent files undercounts the task. See Subagents for how the files tie together and how to aggregate a whole task.

Summarize a session

To see a session’s behavior at a glance (tool mix, bytes retrieved, latency, tokens, observed models, final answer), use the stats command:

  agentminutes stats session.jsonl

For the full command reference, refer to CLI.