ai-agents-metrics analyzes local Claude Code and Codex history.
ai-agents-metrics history-update
ai-agents-metrics show
ai-agents-metrics render-htmlRuns ingest, normalization, classification, and derivation for local agent history. Use --source codex, --source claude, or --source all; --source-root overrides a source directory and --warehouse-path overrides the SQLite warehouse.
Run this command when another command reports an outdated warehouse schema. The warehouse is
rebuilt from source history; see the lifecycle policy in Architecture.
Prints a warehouse-native summary of threads, sessions, sessions per thread, messages,
token usage, coverage, and the observed time window. Pass --json for the stable,
versioned machine-readable output or --warehouse-path to select a warehouse.
Pass --by model, --by project, or --by token-type to print a token breakdown instead of
the aggregate summary. Model and project breakdowns accept --top N; omitted rows are combined
into an explicit other row. The JSON form preserves the same breakdown and scope information.
Generates a self-contained HTML report from the warehouse. --output selects the destination,
--days N limits the time window, and --cwd PATH selects a project in a shared warehouse.
The project scope includes agent worktrees attributed to that parent checkout, matching show.
The generated report also has in-page selectors for one warehouse project or all projects together
and for the trend period. The period can cover the full available history, a recent preset, or a
custom date range. When --days is used, that CLI window remains the maximum range available to
the period selector.
The individual history-ingest, history-normalize, history-classify, and history-derive commands expose each pipeline stage for debugging and automation.
completion {bash,zsh}— shell completion output.install-self— install the executable in a stable local path.verify-public-boundary— check a public repository tree against boundary rules.security— scan staged changes for sensitive data.