cost

name	cost
description	Report what your AI coding sessions have cost — total spend, cache savings, and a tokens-per-day timeline — from your own OpenStory store. Use when the user asks "what did this cost", "my agent spend", "how much have I spent", "tokens per day", or anything about token usage or cost of coding sessions.

cost

Report agent spend from the user's own OpenStory data. Two phases: pull the numbers (deterministic, via MCP), then render + narrate. Never invent figures — report only what the tools return.

Phase 1 — pull the numbers

Call the OpenStory MCP tools (fetch their schemas with ToolSearch first if they aren't loaded — they're named mcp__openstory__*):

• mcp__openstory__token_usage — totals: input / output / cache tokens + estimated cost.
• mcp__openstory__daily_token_usage — the per-day series for the timeline.

If the user names a window ("this week", "last 30 days"), pass it through (most tools take a days argument). Default to the last 7 days.

Scope to one session. token_usage also takes session_id. If the user passes a session id — or says "this session", "latest", or "current" — report just that one session instead of a window:

• a specific id → token_usage(session_id=<id>).
• "this session" / "latest" / "current" → call mcp__openstory__list_sessions, take the most recent ongoing session for this host/user, then pass its id.
• skip daily_token_usage in this mode (there's no multi-day timeline for one session).

If those tools aren't available, OpenStory isn't connected. Say so and tell the user to start it (openstory serve) and connect the MCP — don't guess numbers.

Phase 2 — render the report

Emit a compact spend report:

Agent spend — last <N> days
Total $<total> · <sessions> sessions · cache saved $<saved> (<pct>% off retail)

Tokens / day
<date>  $<cost>  <bar>        # ASCII bar; longest bar = the priciest day
...

Round money to whole dollars. Build the bar lengths proportional to each day's cost. Close with one line of narration: the spendiest day, and the single biggest call if the data exposes it.

Single-session mode (when scoped to one session_id): skip the daily timeline and emit one tight block instead —

Agent spend — this session (<model>)
$<total> · <messages> messages · <total_tokens> tokens · cache saved $<saved> (<pct>% off)
<first_event date> → <last_event date> · <project>

When NOT to use this skill

• For one specific session's detail, use /openstory:recall or a session synopsis.
• For team-wide spend across people, that's a future /openstory:team skill.