Expand description
HTTP contract for the Agent Evaluate domain application.
Evaluate owns exactly one immutable registration unit — the Snapshot — which carries its cases, each case’s evaluators and the execution spec. Everything else it stores is a reference (trace, sandbox, model) or a judgement (metric, verdict, report).
§Layering
agent-evaluate-contract paths + DTOs (this crate)
agent-evaluate-domain pure domain logic
agent-evaluate-adapter-* SQL / Infra SDK
agent-evaluate-server router + auth + binRe-exports§
pub use admin::*;pub use analysis::*;pub use binding::*;pub use catalog::*;pub use driver::*;pub use evaluator::*;pub use execution::*;pub use report::*;pub use run::*;
Modules§
- admin
- Administration surface.
- analysis
- Asynchronous analysis: model-graded judges, plugins and re-analysis.
- binding
- Bindings: which snapshots an agent or runtime is expected to pass.
- catalog
- The snapshot catalogue: the one immutable registration unit.
- driver
- Driver protocol shared by every executor: the in-sandbox runner, a cloud runtime and the CLI’s local driver all speak exactly this.
- evaluator
- Evaluators: the domain rules attached to a case.
- execution
- How a case is executed: which sandbox it gets and where the agent runs.
- report
- Verdicts, reports and the release gate.
- run
- Runs and case runs: the durable side of an evaluation.
Structs§
- ApiError
- Error body shared by every non-2xx response.
- Eval
Scope - Tenant/project scope derived from a verified identity. Callers never get to assert their own scope in a request body.
Constants§
- ADMIN_
BINDINGS_ PATH - ADMIN_
JUDGE_ ALLOWLIST_ PATH - ADMIN_
OVERVIEW_ PATH - ADMIN_
RUNS_ PATH - ADMIN_
SNAPSHOTS_ DEPRECATE_ PATH - ADMIN_
SNAPSHOTS_ EXPORT_ PATH - ADMIN_
SNAPSHOTS_ GET_ PATH - ADMIN_
SNAPSHOTS_ IMPORT_ PATH - ADMIN_
SNAPSHOTS_ LIST_ PATH - ADMIN_
SNAPSHOTS_ PATH - ADMIN_
SNAPSHOTS_ PUBLISH_ PATH - ANALYSIS_
CLAIM_ PATH - ANALYSIS_
RERUN_ PATH - ANALYSIS_
SUBMIT_ PATH - AUDIENCE
- BINDINGS_
DELETE_ PATH - BINDINGS_
GET_ PATH - BINDINGS_
LIST_ PATH - BINDINGS_
PATH - CANCEL_
RUN_ PATH - DEFAULT_
EPHEMERAL_ TTL_ MS - Ephemeral snapshot policy (design §15.1). Temporary assets must never enter a long-lived dependency graph, so they are also barred from bindings and includes.
- DEFAULT_
PAGE_ SIZE - DEFAULT_
PORT - DRIVER_
CLAIM_ PATH - DRIVER_
RENEW_ PATH - DRIVER_
SUBMIT_ PATH - HEALTH_
PATH - MAX_
CASES_ PER_ SNAPSHOT - MAX_
EPHEMERAL_ CASES_ PER_ PROJECT - MAX_
EPHEMERAL_ SNAPSHOTS_ PER_ PROJECT - MAX_
EPHEMERAL_ TTL_ MS - MAX_
EVALUATORS_ PER_ CASE - MAX_
FIXTURES_ PER_ CASE - MAX_
FIXTURE_ BYTES - MAX_
INCLUDES_ PER_ SNAPSHOT - MAX_
JUDGES_ PER_ EVALUATOR - MAX_
PAGE_ SIZE - MAX_
RESOURCE_ ID_ CHARS - Identifier length cap shared by every product identifier. Matches the
indexable prefix of a
VARCHAR(191)column underutf8mb4. - MAX_
SNAPSHOTS_ PER_ BINDING - MAX_
TURNS_ PER_ CASE - METRIC_
CANDIDATES_ PATH - METRIC_
HEALTH_ PATH - OPERATIONS_
PATH - READY_
PATH - REPORT_
COMPARE_ PATH - REPORT_
GET_ PATH - REVIEWS_
CLAIM_ PATH - REVIEWS_
COMPLETE_ PATH - REVIEWS_
PATH - RUNS_
PATH - RUN_
CASES_ PATH - SERVICE_
NAME - SNAPSHOTS_
GET_ PATH - SNAPSHOTS_
LIST_ PATH - SNAPSHOTS_
PATH