ak activity stats

Aggregate privacy-bounded local skill usage across AgentKit activity and Claude Code sessions without enabling analytics.

Use ak activity stats to summarize local skill use by skill, kit, runtime, source, date, status, and duration bucket. It is a local read, not remote telemetry.

Usage

ak activity stats

The command accepts no positional arguments.

Command flags

Flag	Default	Description
`--kit <id>`	Empty	Keep rows for one exact kit ID.
`--runtime <id>`	Empty	Keep rows for one exact runtime, such as `codex` or `claude-code`.
`--window <duration>`	`7d`	Set a positive lookback such as `24h`, `7d`, or `2w`.

Day and week forms use fixed 24-hour days. Standard positive Go durations are also accepted. Zero, negative, or malformed windows exit 2.

Common examples

ak activity stats --window 24h
ak activity stats --runtime codex --window 7d --json
ak activity stats --kit engineer --window 2w

Filters are exact and combine when both are present. The result counts events, not unique sessions or users.

Sources and privacy

The live collector reads:

~/.agentkit/activity/events.ndjson, with AGENTKIT_HOME changing the base;
JSONL session files under ~/.claude/projects, with AGENTKIT_CLAUDE_HOME changing the Claude home.

An existing verified analytics generation can serve the same aggregate. If the index is unavailable or invalid, the command silently falls back to the live sources. It never enables, creates, refreshes, or repairs the index.

Rows deliberately exclude prompts, paths, arguments, session IDs, trace IDs, and raw errors. Claude Code Skill-tool calls are parsed locally. There is no direct Codex or OpenCode session parser for this report; their usage can still appear when recorded in the AgentKit activity log. The coverage object names those parser gaps.

The lookback window only filters the report. It does not delete session files, the activity log, or indexed facts. The command performs no network or provider operation and requires no AgentKit authentication or entitlement.

Human and JSON output

Human output prints aggregate rows with skill, runtime, source, status, duration bucket, count, and UTC date bucket. JSON success uses:

schema_version: 1
kind: activity.stats
data.schema_version: 1
data.rows: array
data.total: integer
data.coverage.sources: source, parsed, and skipped counts
data.coverage.no_parser_runtimes: array

Each row contains skill, optional kit, runtime, source, date_bucket, status, duration_bucket, and count. total is the sum of row counts, not the number of rows.

Interaction and exits

The command never prompts and ignores stdin. --yes has no effect.

Exit	Meaning
`0`	Aggregation completed, including when no usage matched.
`1`	A local source scan/read or output write failed, or a positional argument was supplied.
`2`	A flag was invalid or `--window` was not a positive supported duration.

Command failures in --json mode can return no structured error body. Use the process status and validate the success envelope before consuming it.

On macOS, the CLI-wide first-launch guard can also write ~/.agentkit/.first-run-darwin and print a one-time hint to stderr before this command runs.