mktrue audit
NAME
mktrue audit — measure this repository’s Claude Code sessions in one local report
SYNOPSIS
mktrue audit [--out <file>] [--json] [--target <name>] [--bench <dir>]
DESCRIPTION
audit reads this repository’s local Claude Code transcripts and writes one HTML report. It sends nothing. Run it from anywhere inside a git repository.
It reads the sessions of the repository and of its worktrees, and drops any record from another project before counting.
It reports calls by context size, the fixed reading per turn, documents read more than once, one row per subagent, resumed agents and gate runs, per slice.
The report goes under the repository’s git directory, or to --out. --json prints it to stdout and writes no file. It is a read model, not a gate: it never exits with findings.
OPTIONS
--out <file>- Write the report to this file.
--json- Print the report to stdout, and write no file.
--target <name>- The agent target to render for. The default is claude-code.
--bench <dir>- The kit to render from. The default is the installed kit.
--help- Print the usage of every command.
EXIT STATUS
| 0 | true: a report was produced. |
| 2 | usage: a bad argument, or not inside a git repository. |
| 3 | environment: no transcripts to read, or transcripts in a shape audit does not know. |
EXAMPLES
In a fresh git repository, before any Claude Code session.
mktrue auditmktrue: ✗ environment · no Claude Code transcripts for this repository why audit reads only the project dirs Claude Code wrote for this repository fix run Claude Code in this repository first, or set CLAUDE_CONFIG_DIR exit 3
Real output from mktrue 0.3.1, captured with NO_COLOR set.