菜单
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.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Narrate the story of a project or topic from your session history — the key decisions and turning points, as a deterministic timeline grounded in real sessions and git. Use when the user asks "tell the story of <project>", "how did <X> come to be", "trace the arc of <topic>", "what's the history of this", or wants a narrative built from the record (not a vibe summary).
Prime yourself with prior OpenStory context before starting — pick up exactly where the last session on this project left off: what was done, what's open, the tools used. Use at the START of work, or when the user says "where did we leave off", "resume", "pick up where I left off", "catch up on this project", or "what was I doing here".
See who on your team has an active or recent OpenStory session and what each person is working on — live roster, per-teammate summaries, and recent focus. Requires federated sessions (multiple hosts/users reporting to one store). Use when the user asks "who's working right now", "what is <teammate> on", "team activity", "what did my team do today", or "sync me before our standup".
Show where your time actually goes across coding sessions — by project, by hour, and how focused vs scattered your work is. Distinct from cost (money) and tools (frequency). Use when the user asks "where does my time go", "how much time on X", "when am I most productive", "am I focused or scattered", or "time breakdown".
Show which tools and commands you rely on most across coding sessions — tool histogram, top shell commands, and Read:Write ratio. Frequency, not duration (that's time) or cost (that's cost). Use when the user asks "what tools do I use most", "my top commands", "what do I reach for", "how often do I edit vs read", or "tool breakdown".
Watch the work happening on a branch or project through OpenStory and summarize it live as it streams — a real-time feed of tools and edits plus a rolling summary of what the work is converging on. Use when the user says "watch this branch", "what's happening on <branch> right now", "stream the work", "follow along live", "show me activity as it happens", or "tail the agent on <project>".
| 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. |
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.
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:
token_usage(session_id=<id>).mcp__openstory__list_sessions,
take the most recent ongoing session for this host/user, then pass its id.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.
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>
/openstory:recall or a session synopsis./openstory:team skill.