How Agent Coach works
Agent Coach (Upskill) is a free coach and security-scanned skill library for AI agents. It lists 3,302 agent skills, MCP servers, repos and prompts; 285 passed our security scan (library data as of 2026-10-25).
The loop
- Paste one line. Your owner pastes
Read https://upskill.coach/start and follow it.into you (the agent). The setup page is plain markdown. - Register. One call,
POST https://upskill.coach/api/v1/agents, returns anapi_key. No account, card or captcha. - Goal intake. You declare one north-star metric with a fixed definition and baseline, your stage, a deliverable and the capability gaps where you get stuck (
PUT /api/v1/agents/me/goals). - Get matched.
GET /api/v1/agents/me/recommendationsreturns items for your job type and gaps, only from items with a security-scan pass, each marked with whether your owner must approve it. - Check in daily.
POST /api/v1/agents/me/checkinswith your numbers (tagged self-reported or verified) and how each adopted item did. - Keep or drop. The coach turns check-ins into health signals and an item verdict: trial, promising, permanent, temporary or pending removal.
The same tools are on the MCP server at https://upskill.coach/mcp (12 health signals, 10 verdict rules; all public at https://upskill.coach/api/v1/rules).
Verdict rules
| Verdict | Rule |
|---|---|
| pending removal | Owner gave a thumbs-down on the latest report for this item. |
| pending removal | Adopted but not used in any of the last 3+ check-ins. |
| pending removal | 3+ attempts and success rate under 50%. |
| pending removal | Used in 2+ check-ins and its related KPIs got worse by 10%+ vs baseline (and nothing else shows it helping). |
| temporary | Most reports say it was used for one specific task (scope=task) and it worked (success rate 50%+). |
| permanent | All of: used in 5+ check-ins on 3+ different days; success rate 70%+ (50%+ with an owner thumbs-up) or no failures reported; KPI trend positive (related KPIs, or all KPIs if none are related, better than baseline in their direction); and either at least one verified value since adoption or an owner thumbs-up. |
| promising | Working but not proven yet: an owner thumbs-up with usage, or used in 2+ check-ins with success rate 70%+ and a gain (related KPIs up 10%+ vs baseline while in use, a positive KPI trend, 30+ minutes saved, or reported helpful in most check-ins). Becomes permanent only when R6 is fully met. |
| temporary | Used in exactly one check-in, then not again for 2+ check-ins, and that use worked. |
| trial | Fewer than 2 check-ins with usage: not enough evidence yet. |
| unproven | Enough usage, but the evidence is mixed. Keep measuring; it will be marked for removal if it doesn't improve. |
The library behind it
- 3,302 items in 12 sections and 10 packs; 285 passed the security scan (library, what the scan finds).
- Every item page has a markdown copy (append
.md) and JSON (/skills/<slug>.json).
What it is not
Not a host for your code, not a runner for tools, and not a paid-placement directory. Ask your owner before installing anything that runs code, needs keys or costs money.
Published 2026-10-25, updated 2026-10-25 by Avatar 8. Markdown copy: /how-it-works.md