@wardby/cli 0.3.0 → 0.4.1

package/dist/providers/llm/pricing-anthropic.js

@@ -1,48 +0,0 @@
-/**
- * Anthropic direct-API pricing table. Every entry is a fully literal,
- * hardcoded ModelPricing — every field, including cache read/write, is the
- * provider's own published rate for that exact model, never derived from
- * another field via a ratio (see CLAUDE.md "LLM pricing tables — STRICT": a
- * multiplier that happens to be correct today silently goes stale the
- * moment one model's real cache pricing diverges from the pattern).
- */
-import { computeCost } from "./pricing-core.js";
-function claude(inputPerMTok, outputPerMTok, cachedInputPerMTok, cacheWritePerMTok) {
-    return { encoding: "o200k_base", inputPerMTok, outputPerMTok, cachedInputPerMTok, cacheWritePerMTok };
-}
-// Rates are the exact published 5-minute cache-TTL numbers
-// (platform.claude.com/docs/en/about-claude/pricing, "Model pricing" table)
-// — wardby only ever emits the default 5-minute cache_control breakpoint,
-// so the 1-hour-TTL rates are out of scope and not stored here.
-const PRICING = {
-    "claude-opus-5": claude(5, 25, 0.5, 6.25),
-    "claude-sonnet-5": claude(2, 10, 0.2, 2.5),
-    "claude-fable-5": claude(10, 50, 1, 12.5),
-    "claude-haiku-4-5": claude(1, 5, 0.1, 1.25),
-};
-// Reasoning effort levels each model accepts (`output_config.effort`). The
-// single source of truth for direct-API Claude models; a model missing here
-// accepts no effort, so a new roster entry never gets one by accident.
-// Haiku 4.5 rejects the parameter outright.
-const ALL_EFFORTS = ["low", "medium", "high", "xhigh", "max"];
-const EFFORTS = {
-    "claude-opus-5": ALL_EFFORTS,
-    "claude-sonnet-5": ALL_EFFORTS,
-    "claude-fable-5": ALL_EFFORTS,
-    "claude-haiku-4-5": [],
-};
-export function anthropicSupportedEfforts(model) {
-    return EFFORTS[model] ?? [];
-}
-export function getAnthropicPricing(model) {
-    const p = PRICING[model];
-    if (!p)
-        throw new Error(`No pricing entry for model "${model}" — refusing to price at zero. Add it to src/providers/llm/pricing-anthropic.ts.`);
-    return p;
-}
-export function anthropicSupportedModels() {
-    return Object.keys(PRICING);
-}
-export function anthropicPriceUsd(model, usage) {
-    return computeCost(getAnthropicPricing(model), usage);
-}

package/dist/providers/llm/pricing-bedrock-claude.d.ts

@@ -1,20 +0,0 @@
-/**
- * Bedrock-Claude pricing table. Every entry is a fully literal, hardcoded
- * ModelPricing — every field, including cache read/write, is the provider's
- * own published rate for that exact model, never derived from another
- * field via a ratio (see CLAUDE.md "LLM pricing tables — STRICT": a
- * multiplier that happens to be correct today silently goes stale the
- * moment one model's real cache pricing diverges from the pattern).
- */
-import { type ModelPricing, type UsageTokens } from "./pricing-core.js";
-import type { LlmEffort } from "./types.js";
-export declare function getBedrockClaudePricing(model: string): ModelPricing;
-export declare function bedrockClaudeSupportedModels(): string[];
-/**
- * Reasoning effort is not sent through Bedrock: whether InvokeModel accepts
- * the Messages `output_config.effort` field for these models is unconfirmed,
- * and an unaccepted field fails the whole call. Every Bedrock model therefore
- * accepts no effort until that is verified per model.
- */
-export declare function bedrockClaudeSupportedEfforts(_model: string): readonly LlmEffort[];
-export declare function bedrockClaudePriceUsd(model: string, usage: UsageTokens): number;

package/dist/providers/llm/pricing-bedrock-claude.js

@@ -1,46 +0,0 @@
-/**
- * Bedrock-Claude pricing table. Every entry is a fully literal, hardcoded
- * ModelPricing — every field, including cache read/write, is the provider's
- * own published rate for that exact model, never derived from another
- * field via a ratio (see CLAUDE.md "LLM pricing tables — STRICT": a
- * multiplier that happens to be correct today silently goes stale the
- * moment one model's real cache pricing diverges from the pattern).
- */
-import { computeCost } from "./pricing-core.js";
-function claude(inputPerMTok, outputPerMTok, cachedInputPerMTok, cacheWritePerMTok) {
-    return { encoding: "o200k_base", inputPerMTok, outputPerMTok, cachedInputPerMTok, cacheWritePerMTok };
-}
-// These are `us.` cross-region inference profiles (region us-east-1) — routability
-// is keyed by model ID only; region is a separate adapter concern
-// (BEDROCK_REGION/AWS_REGION). Rates below are the exact published 5-minute
-// cache-TTL numbers (platform.claude.com/docs/en/about-claude/pricing,
-// confirmed 2026-09-08) — this adapter only ever emits the default 5-minute
-// cache_control breakpoint (see claude-messages.ts's withCacheBreakpoints),
-// so the 1-hour-TTL rates are out of scope and not stored here.
-const PRICING = {
-    "us.anthropic.claude-sonnet-4-6": claude(3, 15, 0.3, 3.75),
-    "us.anthropic.claude-opus-4-6-v1": claude(5, 25, 0.5, 6.25),
-    "us.anthropic.claude-opus-4-8": claude(5, 25, 0.5, 6.25),
-    "us.anthropic.claude-haiku-4-5-20251001-v1:0": claude(1, 5, 0.1, 1.25),
-};
-export function getBedrockClaudePricing(model) {
-    const p = PRICING[model];
-    if (!p)
-        throw new Error(`No pricing entry for model "${model}" — refusing to price at zero. Add it to src/providers/llm/pricing-bedrock-claude.ts.`);
-    return p;
-}
-export function bedrockClaudeSupportedModels() {
-    return Object.keys(PRICING);
-}
-/**
- * Reasoning effort is not sent through Bedrock: whether InvokeModel accepts
- * the Messages `output_config.effort` field for these models is unconfirmed,
- * and an unaccepted field fails the whole call. Every Bedrock model therefore
- * accepts no effort until that is verified per model.
- */
-export function bedrockClaudeSupportedEfforts(_model) {
-    return [];
-}
-export function bedrockClaudePriceUsd(model, usage) {
-    return computeCost(getBedrockClaudePricing(model), usage);
-}

package/dist/providers/llm/pricing.d.ts

@@ -1,30 +0,0 @@
-/**
- * Static price + tokenizer table for the budget guardrail.
- *
- * Sourced from OpenAI's public pricing page at write time. Deliberately
- * static (no pricing API call) so `priceUsd` stays synchronous and
- * network-free — the budget math must work offline and in tests. An unknown
- * model throws rather than silently pricing at zero, so a misconfigured
- * agent can never bypass the budget guardrail by accident.
- *
- * `encoding` is here too (not just price) because it drives which
- * `gpt-tokenizer` BPE table `countTokens` uses — every gpt-4o-and-later
- * model uses `o200k_base`, not the older `cl100k_base` gpt-tokenizer
- * defaults to. Keeping both in one table means a model can't be priced
- * without also being told which tokenizer estimates its input cost.
- *
- * `cachedInputPerMTok` / `cacheWritePerMTok` are optional: not every model's
- * entry carries verified cache rates. When a model has cached or
- * cache-write tokens but no rate for them, `priceUsd` falls back to the
- * full (non-cached) input rate — the same fail-toward-overestimate
- * direction as the unknown-model error, never fail-toward-underestimate.
- */
-import { type ModelPricing, type UsageTokens } from "./pricing-core.js";
-export { computeCost } from "./pricing-core.js";
-export type { ModelPricing, TokenizerEncoding, UsageTokens } from "./pricing-core.js";
-/** Bump whenever a rate changes so durable reservations retain their original terms. */
-export declare const PRICING_VERSION = "2026-09-06";
-/** Looks up a model's pricing/tokenizer entry, or throws (fail closed). */
-export declare function getModelPricing(model: string): ModelPricing;
-export declare function supportedModels(): string[];
-export declare function priceUsd(model: string, usage: UsageTokens): number;

package/dist/providers/llm/pricing.js

@@ -1,74 +0,0 @@
-/**
- * Static price + tokenizer table for the budget guardrail.
- *
- * Sourced from OpenAI's public pricing page at write time. Deliberately
- * static (no pricing API call) so `priceUsd` stays synchronous and
- * network-free — the budget math must work offline and in tests. An unknown
- * model throws rather than silently pricing at zero, so a misconfigured
- * agent can never bypass the budget guardrail by accident.
- *
- * `encoding` is here too (not just price) because it drives which
- * `gpt-tokenizer` BPE table `countTokens` uses — every gpt-4o-and-later
- * model uses `o200k_base`, not the older `cl100k_base` gpt-tokenizer
- * defaults to. Keeping both in one table means a model can't be priced
- * without also being told which tokenizer estimates its input cost.
- *
- * `cachedInputPerMTok` / `cacheWritePerMTok` are optional: not every model's
- * entry carries verified cache rates. When a model has cached or
- * cache-write tokens but no rate for them, `priceUsd` falls back to the
- * full (non-cached) input rate — the same fail-toward-overestimate
- * direction as the unknown-model error, never fail-toward-underestimate.
- */
-import { computeCost } from "./pricing-core.js";
-export { computeCost } from "./pricing-core.js";
-/** Bump whenever a rate changes so durable reservations retain their original terms. */
-export const PRICING_VERSION = "2026-09-06";
-const PRICING = {
-    "gpt-4o": { encoding: "o200k_base", inputPerMTok: 2.5, cachedInputPerMTok: 1.25, outputPerMTok: 10.0 },
-    "gpt-4o-mini": { encoding: "o200k_base", inputPerMTok: 0.15, cachedInputPerMTok: 0.075, outputPerMTok: 0.6 },
-    "gpt-4.1": { encoding: "o200k_base", inputPerMTok: 2.0, cachedInputPerMTok: 0.5, outputPerMTok: 8.0 },
-    "gpt-4.1-mini": { encoding: "o200k_base", inputPerMTok: 0.4, cachedInputPerMTok: 0.1, outputPerMTok: 1.6 },
-    "gpt-4.1-nano": { encoding: "o200k_base", inputPerMTok: 0.1, cachedInputPerMTok: 0.025, outputPerMTok: 0.4 },
-    "gpt-6-astra": {
-        encoding: "o200k_base",
-        inputPerMTok: 10.0,
-        cachedInputPerMTok: 1.0,
-        cacheWritePerMTok: 12.5,
-        outputPerMTok: 50.0,
-    },
-    "gpt-5.6-sol": {
-        encoding: "o200k_base",
-        inputPerMTok: 4.0,
-        cachedInputPerMTok: 0.4,
-        cacheWritePerMTok: 5.0,
-        outputPerMTok: 20.0,
-    },
-    "gpt-5.6-terra": {
-        encoding: "o200k_base",
-        inputPerMTok: 2.0,
-        cachedInputPerMTok: 0.2,
-        cacheWritePerMTok: 2.5,
-        outputPerMTok: 12.0,
-    },
-    "gpt-5.6-luna": {
-        encoding: "o200k_base",
-        inputPerMTok: 0.2,
-        cachedInputPerMTok: 0.02,
-        cacheWritePerMTok: 0.25,
-        outputPerMTok: 1.2,
-    },
-};
-/** Looks up a model's pricing/tokenizer entry, or throws (fail closed). */
-export function getModelPricing(model) {
-    const pricing = PRICING[model];
-    if (!pricing) {
-        throw new Error(`No pricing entry for model "${model}" — refusing to price at zero. ` + `Add it to src/providers/llm/pricing.ts.`);
-    }
-    return pricing;
-}
-export function supportedModels() {
-    return Object.keys(PRICING);
-}
-export function priceUsd(model, usage) {
-    return computeCost(getModelPricing(model), usage);
-}