CLI recipes

Every page of Log Book is also a logbook command, and every command only reads the warehouse unless it says otherwise. These are the common jobs; CLI reference has every option.

Find what keeps failing

logbook report failures --max-rows 20

Failed tool calls by tool, failed model requests, shell calls by purpose with how many failed and why, calls repeated after failing, and interruptions. One report of the topic at a time:

logbook report tool-failures
logbook report repeats

Find what is slow

logbook report slow
logbook report longest-turns
logbook report turn-time

The slowest tool calls, the longest turns by active time, and where turn time went, split into model time, tool time and idle. logbook report performance prints all of them, with tool output sizes and the heaviest sessions.

Find a conversation you remember

logbook search '"connection refused"'
logbook search 'invoice AND rounding' --limit 5

Search reads prompts, replies, reasoning and tool output, with SQLite's full-text syntax: quoted phrases, AND, OR, NOT and prefix*. To list sessions instead:

logbook sessions --project shop --since 2026-09-01
logbook sessions --harness opencode --limit 10

Read one session

logbook timeline claude-code:de30da7a-0000-4000-8000-000000000001
logbook tree claude-code:de30da7a-0000-4000-8000-000000000001

The timeline prints the prompts, replies and tool calls in order, with their durations; the tree prints the subagent sessions it started. A session is its warehouse id, as above, or the id your agent shows.

Ask your own question

logbook sql "SELECT bare_name, COUNT(*) AS calls FROM tool_call GROUP BY bare_name ORDER BY calls DESC LIMIT 10"

The query is read-only. Database schema lists the tables, and logbook sql --help prints them, so a coding agent can write its own queries.

Let your coding agent ask

Ask Claude Code or OpenCode something like "what kept failing in this project last week?" and it can answer from your history with these commands. What a command prints becomes part of that agent's conversation, so it reaches the agent's model provider like anything else the agent reads.

Every report

Topic	Reports
failures	tool-failures, errors, shell, tool-causes, repeats, interrupts, recovery
performance	tools, slow, latency, turn-time, longest-turns, heavy, output, context
usage	skills, commands, families, models, daily
sessions	calls, chains, work, goals, outcomes, links
interaction	reactions, replies

logbook report <topic> prints a topic's reports, and logbook report with no name prints them all. goals, outcomes, reactions and replies read model labels.