Agent Cards
A standardized format for documenting AI agent systems -- what they are, how they work, and how they perform.
Model cards describe models. Agent cards describe agents: the full system including planning strategy, tool use, memory, error recovery, and the models inside them.
Why agent cards?
Evaluating an agent isn't just about what score it gets. It's about understanding the system that produced that score. Two agents can achieve identical benchmark results while differing dramatically in cost, failure modes, and architecture. Agent cards make these differences visible.
See Rethinking Agent Evaluation Reporting for the motivation behind this format.
Quick start
Copy the template and fill it in for your agent. See examples/ for reference.
Template structure
An agent card is a markdown file with YAML frontmatter:
Frontmatter (machine-readable):
- Identity: name, version, developers, license
- Framework and repository
- Models used
- Tags for discoverability
Sections (human-readable):
- Agent Details -- what it is
- Architecture -- how it works (planning, memory, tools, error recovery)
- Models -- which LLMs and how they're used
- Supported Environments -- what domains and benchmarks it handles
- Evaluation Results -- performance, cost, and failure data
- Limitations -- known failure modes and unsupported scenarios
- How to Run -- installation and usage