metrics

name	metrics
description	Show the token-accounting dashboard panel for the current memhub project — 7-day and 30-day recall/session token totals, context offset vs full-ledger baseline, and a recent-sessions table. Trigger on: "token usage", "context cost", "how many tokens", "recall savings", "show memhub metrics", "token accounting".
framework	memhub
framework_version	1.0.0
last_updated	"2026-05-16T00:00:00.000Z"

Display the memhub token-accounting dashboard for this repo. Read-only — no writes, no config changes.

This is the Codex counterpart to the Claude Code /metrics skill. Both call into the same memhub.metrics MCP tool and memhub metrics status CLI; they differ only in the agent identifier on whatever read-side telemetry the host captures.

Preconditions

• .memhub/ exists in the working repo (run /check-init if unsure).
• memhub binary on PATH, OR memhub.metrics MCP tool available.

Invocation

Prefer the MCP tool — it returns rendered_panel directly with no shell quoting or parsing required:

memhub.metrics()

CLI fallback:

memhub metrics status

Rendering

MCP path (preferred):

Check the enabled field first.

• enabled = false → print:

  Token accounting is disabled for this repo.
  Run `memhub metrics enable` to opt in and start tracking session tokens.
  

  Stop here.

• enabled = true → print rendered_panel verbatim. Do not reformat, summarize, or restructure the panel — the Rust formatter owns the layout (decision 76).

CLI fallback:

Run memhub metrics status and surface its output. If it reports "disabled", print the same one-liner as the MCP disabled branch above.

What the panel contains

When enabled with data, the panel has three sections:

1. Last 7 days — 4-line period block: Recalls / Sessions / Real tokens (in/out/cache_read/cache_creation) / Context offset vs full-ledger baseline

2. Last 30 days — same 4-line shape

3. Recent sessions (≤10, newest first) — fixed-width table with columns: session (8-char prefix), agent, started (UTC MM-DD HH:MM), in, out, recalls

When enabled but no data has been captured yet, the panel reads: Metrics enabled — no recall or session data captured yet.

Notes

• The context_offset_pct field is bundle_tokens / ledger_tokens * 100. It measures how much smaller the recall bundle is relative to returning the full ledger — a proxy for context saved, not exact tokens saved (decision 74, documented caveat: tiktoken-cl100k is ±10% vs Anthropic's tokenizer; ratios are sound, absolute numbers are approximate).
• Timestamps are local time (converted by SQLite datetime(..., 'localtime')).
• Token integers are formatted with comma thousands separators.