What they are & how to use them

TL;DR: Agent Cards are the one-page spec for each digital worker in your stack. They tell you when to use it, what it needs, how it behaves, what it must never do, and how you prove it works (HIL gates, budgets, run-logs, ROI). Think of them as “role sheets” for governed AI workers.

These are provided in your paths and don’t require much of you other than copy pasting them into your workbench. 80% of the thinking is done for you.

Where Agent Cards fit (at a glance)

• SOPs: your company standard (the “official way we do X”).
• Run Cards: your step-by-step play (how to execute today).
• Agent Cards (this page): your digital worker profile (mission, tools, guardrails, approvals, proof).

Use them together: SOP ⇒ Agent Card ⇒ Run Card. SOP sets the policy, Agent Card defines the worker, Run Card is the daily checklist.

What’s inside every Agent Card (the anatomy)

1. Agent ID & Mission (1 sentence) — What outcome this worker owns.
2. When to Use / Not to Use — Clear triggers and exclusions.
3. Inputs & Context Pack — Required fields, files, and links the agent will read.
4. Tools & Access — Connected apps, scopes, rate limits; link to Access & Policy Matrix.
5. System Prompt Summary — 5–10 bullets of rules (tone, steps, formatting).
6. Capabilities & Boundaries — What it can do; Never-Do list it must avoid.
7. HIL (Human-in-the-Loop) Gates — Approval points, thresholds, and approver.
8. Budgets & Limits — Max runs/day, tokens/run, timeouts, spend caps.
9. Fallbacks & Edge Cases — What to do on 429s, CAPTCHAs, empty states, auth fails.