Skip to content
Back to skills

Admin Ai Usage

ASecurity

Answer questions about AI spend and adoption on this platform - "what did we spend on AI this week", "top users by cost", "which skills / models / plugins are used most", "were there failed or denied requests", "how does this week compare to last". For what a person or team has been working on, use `admin_person_activity`. All tools are on the admin `systemprompt` server and read this platform's own audit tables, so figures are exact; quote them as returned.

  • 29 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added October 1, 2026
ai-agentsdebugging

Works with

  • claude code
  • cursor
  • cli

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned October 1, 2026

npx -y skills add systempromptio/systemprompt-template --skill admin_ai_usage --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Admin Ai Usage?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Admin Ai Usage
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/systempromptio-admin-ai-usage/badge)](https://www.skillsdirectory.com/skills/systempromptio-admin-ai-usage)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
# AI spend and adoption

Answer questions about AI spend and adoption on this platform - "what did we spend on AI this
week", "top users by cost", "which skills / models / plugins are used most", "were there failed
or denied requests", "how does this week compare to last". For what a person or team has been
working on, use `admin_person_activity`.

All tools are on the admin `systemprompt` server and read this platform's own audit tables,
so figures are exact; quote them as returned.

## Which tool answers what

| Question | Tool |
|---|---|
| Totals, trend, provider/model split | `admin_report {"report":"costs","days":7}` — call it fresh every run |
| Spend per person | `usage_by_user {"since":"7d","limit":100}` then `users` for names |
| What people did, active days, skills per person | `user_activity {"since":"30d"}` (empty `user` = everyone) |
| Which skills are used, by whom | `conversation_list {"since":"30d","skill":"<name>"}`; `user_activity` `skills` column |
| Failed or denied requests | `conversation_list` rows with `errors`/`denied` > 0, then `conversation_audit` on one or two |
| One request's detail | `request_log` then `conversation_audit` — debugging only |
| Anything else | the `systemprompt` CLI tool (`analytics costs breakdown --by model --since 7d`, `analytics requests stats`); flags are tabled in `systemprompt_cli` |

For a change, run the same tool again with `since`/`until` set to the previous period of
equal length and report both.

`user_activity` and `conversation_list` read the conversation rollup. `facts_as_of` is when it
last ran; activity after it is not counted yet, and an instance whose rollup has never run
returns no rows from them — fall back to `usage_by_user` and `admin_report` and say so.

## Rules

- **Rank people by conversations and spend, never by requests.** One Claude Code session is
  hundreds of requests.
- **Never conclude from one page.** Totals, "top", "busiest" and trends come from the
  aggregate tools (`admin_report`, `usage_by_user`, `user_activity`) or from paging a list to
  its end. A page that returned `next_cursor` is not the whole window; say so if you stop.
- **When two sources disagree, report both figures with their sources.** Do not invent a
  reason. `usage_by_user` excludes deleted users; `request_log` does not.
- **Say what was missing.** State the window, and name any source that was denied, failed or
  empty. A missing source is not a zero.
- Money: quote `*_usd`, never convert `*_microdollars` yourself. A model's request share is not
  its spend share.
- Do not paste full prompts; quote at most a trimmed line when it supports a judgement.

## Changes

Reporting authorises no writes. When a change is explicitly requested: inspect the current
state, run the exact command, read the result back, and report it. Never retry a write after a
timeout before checking whether it applied.

Files in this skill

  • SKILL.md2.8 KB
  • config.yaml349 B

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.