@sunerpy/kiro-provider 3.0.0 → 3.1.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/README.md +17 -6
  2. package/dist/cli.js +105 -92
  3. package/package.json +2 -1
package/README.md CHANGED
@@ -41,9 +41,10 @@
41
41
  - Live per-account model discovery and account-aware routing through Kiro management, with bounded stale/static fallback. Production calls use the live-probe-confirmed `runtime.<region>.kiro.dev` dialect. Token-usage metadata is an immediate completion witness; the current runtime's valid terminal metering event is accepted only when followed by clean EOF.
42
42
  - Default `v3-auto` transport selection: ordinary requests use KiroRuntime's
43
43
  native OpenAI Responses operation and request shapes requiring `store:false`,
44
- max effort, encrypted reasoning replay, custom grammar, namespace tools, or
45
- Codex collaboration use the canonical stateless fallback.
46
- - Encrypted reasoning replay for complete native Kiro envelopes: opaque `kr1_...` tokens, AES-256-GCM storage, tenant/model/account/conversation/output binding, TTL/LRU cleanup, and account-locked replay.
44
+ max effort, provider `kr1_` replay, custom grammar, or Codex collaboration
45
+ use the canonical stateless fallback. Verified namespace/free-form tools can
46
+ stay on native Responses through a persistent tool bridge.
47
+ - Complete signed Kiro envelopes use provider `kr1_...` replay tokens, AES-256-GCM storage, tenant/model/account/conversation/output binding, TTL/LRU cleanup, and account-locked replay. Native opaque tokens stay on CreateResponse and recover their owner from durable response records.
47
48
  - Multi-account rotation with automatic token refresh and failover. Exhausted accounts are hard-excluded from model attempts, then automatically rejoin only after a bounded, deduplicated Kiro usage probe confirms a new quota window. A provider-owned maintenance loop also refreshes near-expiry tokens and stale usage while the service is idle.
48
49
  - `kiro-provider login` and `accounts import` write directly to the provider-owned local authentication store. The former `auth_source: "opencode-shared"` compatibility mode was removed in 0.7.0; a configuration that still selects it fails at startup with migration instructions (import once, then use `local`).
49
50
  - A single global `proxy_url` that, when set, routes all upstream egress (model requests, token refresh, quota probes, device-code login) through one HTTP(S) proxy.
@@ -55,9 +56,9 @@ V3 implements the core OpenAI Responses resource and makes every upstream
55
56
  difference explicit:
56
57
 
57
58
  - native JSON/SSE creation, instructions, function tools, supported effort and
58
- token controls, and native `previous_response_id`;
59
- - automatic stateless fallback for `store:false`, max effort, encrypted
60
- reasoning, custom grammar, namespace tools, and Codex multi-agent items;
59
+ token controls, and `previous_response_id` (exact native replay for affected Claude and Sol reasoning histories);
60
+ - automatic stateless fallback for `store:false`, max effort, provider-token
61
+ replay, custom grammar, unverified tool-bridge combinations, and Codex multi-agent items;
61
62
  - tenant-isolated local response mirrors for retrieve, delete, input-items
62
63
  pagination, and continuation;
63
64
  - field-level OpenAI error envelopes for capabilities Kiro cannot preserve,
@@ -71,6 +72,16 @@ account does not advertise the private `systemPrompt` feature. The default
71
72
  `v3-auto` path instead uses KiroRuntime CreateResponse's native
72
73
  `instructions` field.
73
74
 
75
+ `responses_fidelity_mode` defaults to `compatible` and reports known losses in
76
+ `X-Kiro-Compatibility`; `strict` rejects those semantics before generation.
77
+ `X-Kiro-Transport` distinguishes native, native-adapted, and stateless calls.
78
+ Native tool bridges are enabled only for verified model/region cells. Instruction
79
+ lifting stays experimental until its complete continuation gate passes. See the
80
+ [Responses fidelity validation](docs/audits/kiro-provider-responses-fidelity-2026-09-10.zh.md)
81
+ for history, reasoning, instruction-priority boundaries, and storage migration.
82
+ The [before/after report](docs/audits/kiro-provider-responses-before-after-2026-09-10.zh.md)
83
+ includes real OpenAI SDK, Codex, and Zuno results.
84
+
74
85
  For the transport decision table, stored-response contract, data-retention
75
86
  boundary, verified model controls, and current client evidence, see
76
87
  [`docs/PROTOCOL_COMPATIBILITY.md`](docs/PROTOCOL_COMPATIBILITY.md) and the