A durable, D1-backed catalog of the terse YAML interchange style.
a2a.terse.v1 is a narrow, opinionated style for authoring agent-to-agent tool/task/routine/manifest YAML: two files, style.yaml (the emission rules) and rubric.md (the 7-criterion scoring rubric), governing artifacts shaped around exactly four top-level keys -- inputs, outputs, steps, and a trailing self-eval array, eval. This site exists to make that narrow spec durable and queryable rather than a fact that only lives in a skill's own SKILL.md.
This site is deliberately scoped narrow -- the broad Agent2Agent / MCP / e2m protocol landscape is covered on subagentprotocols.com. subagentspec.com covers exactly one house style within that landscape: the terse one.
style.yaml defines how a terse artifact is emitted; rubric.md defines how it's graded. Both are small and fixed by design.
Every terse artifact ends with an eval array -- assertions the artifact carries about its own correctness, not just prose describing intent.
structural_fidelity, terseness, io_contract, self_eval_tail, cross_platform, determinism, traceability -- browse them all →.
The terse YAML shape
| key | what it holds |
|---|---|
inputs | Typed parameters the artifact accepts, declared up front. |
outputs | Typed result(s) the artifact produces. |
steps | The ordered, terse sequence of actions turning inputs into outputs. |
eval | The trailing self-eval tail -- an array of assertions the artifact checks about its own output. |
See what the terse style actually is and how the rubric grades it.