| name | o9k-stats |
| description | Token usage report for the current project from Claude Code session transcripts. Use when the user asks how many tokens were used/saved, invokes /o9k-stats, or wants to verify the o9k pillars are paying off. |
o9k-stats — Measure It
Run the bundled analyzer against the current project:
node "${CLAUDE_PLUGIN_ROOT}/scripts/o9k-stats.mjs"
Present the output compactly (caveman rules apply). Interpretation guide:
- output share of total traffic — the lever
caveman pulls (companion
upstream caveman; skip this line if it is not installed). Uncompressed
agents typically sit noticeably higher; falling share across sessions =
compression working.
- avg output/turn — watch the trend, not the absolute: it should drop after
installing
caveman (upstream reports ~65% measured output reduction).
- cache read vs fresh input — high cache-read share is healthy (stable
context prefix); lots of fresh input suggests context churn — check
scout
discipline and whether something regenerates the repo map every turn.
Limits (be honest about them): transcripts only cover Claude Code sessions on
this machine; subagent traffic is attributed to its own transcript; there is no
counterfactual baseline — for a before/after comparison, compare sessions from
before the o9k install date with sessions after.
