Add Anthropic platform: wire-served metadata via listing dialect (provider 2)

The 5th registry row: id "anthropic", ANTHROPIC_API_KEY > auth.json
"anthropic" scope, api.anthropic.com/v1 with KIGI_ANTHROPIC_BASE_URL
override, Messages dialect. Two new spec dimensions most future rows
reuse: ListingDialect (Anthropic's /v1/models wants x-api-key +
anthropic-version headers, ?limit=1000, and its own response shape) and
PlatformKeyHeader (Bearer vs x-api-key across listing/validation/
inference, with auth_scheme stamped onto entries).

The 2026 Anthropic listing serves real metadata: the adapter maps
max_input_tokens, per-level effort capabilities (low..max as the menu,
xhigh/max distinct), thinking/image flags — and enrichment fills only
genuine wire gaps (e2e pins wire-1M beating enrichment, and a zero
context filled to 200k).

Two review-confirmed defects fixed red-green:
- Output caps were dropped at three layers, so every sub-128K-output
  model (64k Haiku, legacy models) would 400 on EVERY request against
  the sampler's 128K max_tokens default. Wire max_tokens and enrichment
  limit.output now flow to entry.max_completion_tokens.
- An explicit wire effort-decline was indistinguishable from wire
  silence, letting enrichment inject effort menus pre-4.6 models reject
  (adaptive thinking 400). The adapter now emits a decline sentinel
  (support:false) that enrichment respects — proven end to end.

Also: the Messages client now sends anthropic-version (previously never
sent — real api.anthropic.com rejects such requests; pinned across all
three scheme/backend quadrants), key validation builds per-key-header
requests, missing listing data fails fast, empty-id ghosts drop with a
warning, kimi-import recognizes api.anthropic.com as built-in
automatically.
This commit is contained in:
2026-07-21 06:12:53 -04:00
parent 23e94939c0
commit b86722f508
7 changed files with 579 additions and 27 deletions
@@ -266,12 +266,25 @@ fn fetch_one_platform_models(
enrichment: &kigi_models::enrichment::EnrichmentCatalog,
) -> Result<(Vec<crate::agent::config::ModelEntryConfig>, Option<String>), BackendError> {
let client = crate::http::shared_blocking_client();
let url = platform_models_url(platform, endpoints);
let url = match platform.listing() {
kigi_models::ListingDialect::OpenAi => platform_models_url(platform, endpoints),
// Anthropic paginates (default 20); limit=1000 is the documented max
// and far above the catalog size (the adapter warns on has_more).
kigi_models::ListingDialect::Anthropic => {
format!("{}?limit=1000", platform_models_url(platform, endpoints))
}
};
tracing::info!(platform = platform.as_str(), url = %url, "fetching platform models");
let response = client
.get(&url)
.header("Authorization", format!("Bearer {}", bearer))
.send()?;
let request = match platform.key_header() {
kigi_models::PlatformKeyHeader::Bearer => client
.get(&url)
.header("Authorization", format!("Bearer {}", bearer)),
kigi_models::PlatformKeyHeader::XApiKey => client
.get(&url)
.header("x-api-key", bearer)
.header("anthropic-version", kigi_sampling_types::ANTHROPIC_VERSION),
};
let response = request.send()?;
if !response.status().is_success() {
let status = response.status().as_u16();
let body = response.text().unwrap_or_default();
@@ -282,9 +295,22 @@ fn fetch_one_platform_models(
.get("etag")
.and_then(|v| v.to_str().ok())
.map(|s| s.to_string());
let listing: kigi_models::WireModelsResponse = response.json()?;
let total = listing.data.len();
let mut filtered = kigi_models::filter_allowed_models(platform, listing.data);
let data = match platform.listing() {
kigi_models::ListingDialect::OpenAi => {
response.json::<kigi_models::WireModelsResponse>()?.data
}
kigi_models::ListingDialect::Anthropic => {
let body = response.text()?;
kigi_models::parse_anthropic_listing(&body).map_err(|e| {
BackendError::RequestFailed {
status: 200,
body: format!("anthropic listing parse failed: {e}"),
}
})?
}
};
let total = data.len();
let mut filtered = kigi_models::filter_allowed_models(platform, data);
if filtered.len() != total {
tracing::info!(
platform = platform.as_str(),
@@ -432,19 +458,27 @@ pub(crate) fn platform_wire_model_to_entry(
kigi_models::PlatformWireApi::Responses => crate::sampling::ApiBackend::Responses,
kigi_models::PlatformWireApi::Messages => crate::sampling::ApiBackend::Messages,
};
let auth_scheme = match platform.key_header() {
kigi_models::PlatformKeyHeader::Bearer => None,
kigi_models::PlatformKeyHeader::XApiKey => Some(kigi_sampler::AuthScheme::XApiKey),
};
crate::agent::config::ModelEntryConfig {
id: Some(platform.managed_model_key(&wire.id)),
name: Some(wire.display_name.clone().unwrap_or_else(|| wire.id.clone())),
model: wire.id,
base_url: base_url.to_owned(),
description: None,
max_completion_tokens: None,
// The wire/enrichment output cap; the sampler otherwise defaults to
// 128K, which Anthropic rejects on smaller-cap models (400 on every
// request for e.g. a 64K haiku).
max_completion_tokens: (wire.max_output_tokens > 0)
.then(|| u32::try_from(wire.max_output_tokens).unwrap_or(u32::MAX)),
temperature: None,
top_p: None,
api_key: None,
env_key,
api_backend,
auth_scheme: None,
auth_scheme,
reasoning_effort: think_efforts
.and_then(|t| t.default_effort.as_deref())
.and_then(|s| s.parse().ok()),
@@ -836,6 +870,158 @@ mod tests {
);
}
/// Anthropic-cycle e2e (mock wire): the Anthropic listing dialect —
/// x-api-key + anthropic-version headers, ?limit=1000 — maps
/// wire-served metadata (max_input_tokens, per-level effort
/// capabilities) onto Messages-backed XApiKey entries, and enrichment
/// fills a zero max_input_tokens without touching wire-served values.
#[tokio::test(flavor = "multi_thread")]
#[serial_test::serial]
async fn anthropic_listing_maps_wire_metadata_and_enrichment_fills_gaps() {
let platform_server = wiremock::MockServer::start().await;
wiremock::Mock::given(wiremock::matchers::method("GET"))
.and(wiremock::matchers::path("/models"))
.and(wiremock::matchers::query_param("limit", "1000"))
.and(wiremock::matchers::header("x-api-key", "sk-ant"))
.and(wiremock::matchers::header(
"anthropic-version",
"2023-06-01",
))
.respond_with(wiremock::ResponseTemplate::new(200).set_body_json(
serde_json::json!({ "data": [
{
"id": "claude-opus-4-8",
"display_name": "Claude Opus 4.8",
"type": "model",
"max_input_tokens": 1_000_000,
"capabilities": {
"effort": {
"supported": true,
"low": {"supported": true},
"medium": {"supported": true},
"high": {"supported": true},
"xhigh": {"supported": true},
"max": {"supported": true}
},
"thinking": {"supported": true},
"image_input": {"supported": true}
}
},
{
"id": "claude-gap-test",
"type": "model",
"max_input_tokens": 0,
"capabilities": {
"effort": {"supported": false},
"thinking": {"supported": true},
"image_input": {"supported": false}
}
}
], "has_more": false }),
))
.expect(1)
.mount(&platform_server)
.await;
let modelsdev_server = wiremock::MockServer::start().await;
wiremock::Mock::given(wiremock::matchers::method("GET"))
.and(wiremock::matchers::path("/api.json"))
.respond_with(wiremock::ResponseTemplate::new(200).set_body_json(
serde_json::json!({ "anthropic": { "models": {
"claude-gap-test": {
"limit": {"context": 200000, "output": 64000},
"tool_call": true,
"reasoning": true,
"reasoning_options": [
{"type": "effort", "values": ["low", "high"]}
]
},
"claude-opus-4-8": {
"limit": {"context": 555},
"tool_call": true
}
}}}),
))
.expect(1)
.mount(&modelsdev_server)
.await;
let cache_dir = tempfile::tempdir().unwrap();
let _base = kigi_test_support::EnvGuard::set(
kigi_models::ANTHROPIC_BASE_URL_ENV,
platform_server.uri(),
);
let _mdev = kigi_test_support::EnvGuard::set(
crate::agent::enrichment_fetch::MODELS_DEV_URL_ENV,
format!("{}/api.json", modelsdev_server.uri()),
);
let _mdev_cache = kigi_test_support::EnvGuard::set(
crate::agent::enrichment_fetch::MODELS_DEV_CACHE_DIR_ENV,
cache_dir.path(),
);
let endpoints = crate::agent::config::EndpointsConfig::default();
let keys = crate::agent::models::PlatformApiKeys::test_single(
kigi_models::PlatformId::Anthropic,
"sk-ant",
);
let result = tokio::task::spawn_blocking(move || {
fetch_platform_models_blocking(&endpoints, None, &keys)
})
.await
.unwrap()
.expect("fetch must succeed");
assert_eq!(result.models.len(), 2);
let opus = &result.models[0];
assert_eq!(opus.id.as_deref(), Some("anthropic/claude-opus-4-8"));
assert_eq!(
opus.context_window.get(),
1_000_000,
"wire max_input_tokens must WIN over enrichment (555)"
);
assert_eq!(opus.api_backend, crate::sampling::ApiBackend::Messages);
assert_eq!(
opus.auth_scheme,
Some(kigi_sampler::AuthScheme::XApiKey),
"anthropic entries must ride x-api-key at inference"
);
assert_eq!(
opus.reasoning_efforts
.iter()
.map(|o| o.id.as_str())
.collect::<Vec<_>>(),
vec!["low", "medium", "high", "xhigh", "max"],
"wire effort capabilities become the menu"
);
let gap = &result.models[1];
assert_eq!(
gap.context_window.get(),
200_000,
"a zero wire context must be filled by enrichment"
);
assert_eq!(
gap.max_completion_tokens,
Some(64_000),
"the enrichment output cap must reach max_completion_tokens"
);
assert!(
gap.reasoning_efforts.is_empty() && !gap.supports_reasoning_effort,
"the wire's explicit effort decline must block enrichment's menu \
(pre-4.6 models 400 on adaptive thinking); efforts={:?} supports={}",
gap.reasoning_efforts,
gap.supports_reasoning_effort,
);
let opus = &result.models[0];
assert_eq!(
opus.max_completion_tokens, None,
"no wire/enrichment cap on this fixture entry — sampler default applies"
);
assert!(
gap.capabilities
.contains(&kigi_models::ModelCapability::Thinking),
"wire thinking capability must survive"
);
}
#[test]
fn get_env_keys_parses_strings_and_rejects_non_strings() {
use crate::agent::config::EnvKeys;