@oh-my-pi/pi-catalog 18.8.5 → 18.8.6

package/CHANGELOG.md

@@ -2,6 +2,16 @@
 
 ## [Unreleased]
 
+## [18.8.6] - 2026-10-08
+
+### Added
+
+- Added prompt-cache lookback support for Claude models across all hosts, including the public `prompt-cache-lookback` catalog axis and `resolvePromptCacheLookback` API.
+
+### Fixed
+
+- Fixed model discovery when providers publish models before they are recognized by the catalog; unsupported models are now skipped with a warning so other available models remain discoverable.
+
 ## [18.8.5] - 2026-10-08
 
 ### Breaking Changes

package/dist/types/compat/prompt-cache-lookback.d.ts

@@ -0,0 +1,7 @@
+import type { Model } from "../types.js";
+/**
+ * Block positions the provider's prompt cache checks back from a breakpoint for
+ * an earlier request's entry (`prompt-cache-lookback` axis), or `undefined` when
+ * the model has no known lookback bound.
+ */
+export declare function resolvePromptCacheLookback(model: Model): number | undefined;

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.8.5",
+	"version": "18.8.6",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -93,12 +93,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.8.5",
-		"@oh-my-pi/pi-utils": "18.8.5"
+		"@oh-my-pi/omptype": "18.8.6",
+		"@oh-my-pi/pi-utils": "18.8.6"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.8.5",
+		"@oh-my-pi/pi-ai": "18.8.6",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/axes.ts

@@ -399,6 +399,12 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"limits-patch": { key: "limitsPatch", set: "catalog", shape: "object" },
 	"long-context-cost": { key: "longContext", set: "catalog", shape: "object" },
 	"prompt-cache": { key: "promptCache", set: "catalog", shape: "object" },
+	/**
+	 * Prompt-cache lookback in block positions: how far back from a cache
+	 * breakpoint the provider looks for an earlier request's cache entry.
+	 * Unassigned: no known lookback bound.
+	 */
+	"prompt-cache-lookback": { key: "promptCacheLookback", set: "catalog", shape: "scalar" },
 	"long-usage-limit-fallback": { key: "longUsageLimitFallback", set: "catalog", shape: "scalar" },
 	"max-context-window": { key: "maxContextWindow", set: "catalog", shape: "scalar" },
 	"pricing-status": {

package/src/compat/prompt-cache-lookback.ts

@@ -0,0 +1,23 @@
+import { toModelSpec } from "../provider-models/bundled-references";
+import type { Model } from "../types";
+import { resolveCatalogAxes } from "./resolve";
+
+const lookbackCache = new Map<string, number | null>();
+const LOOKBACK_CACHE_MAX = 8192;
+
+/**
+ * Block positions the provider's prompt cache checks back from a breakpoint for
+ * an earlier request's entry (`prompt-cache-lookback` axis), or `undefined` when
+ * the model has no known lookback bound.
+ */
+export function resolvePromptCacheLookback(model: Model): number | undefined {
+	const key = `${model.provider} ${model.id} ${model.api}`;
+	let lookback = lookbackCache.get(key);
+	if (lookback === undefined) {
+		const value = resolveCatalogAxes(toModelSpec(model)).promptCacheLookback;
+		lookback = typeof value === "number" && Number.isInteger(value) && value > 0 ? value : null;
+		if (lookbackCache.size >= LOOKBACK_CACHE_MAX) lookbackCache.clear();
+		lookbackCache.set(key, lookback);
+	}
+	return lookback ?? undefined;
+}

package/src/compat/rules/README.md

@@ -219,6 +219,8 @@ One object axis carries a computed form: `long-context-cost` accepts either the
 
 `inline-image-byte-budget` caps retained base64 image characters for a deployment's request body. It leaves room for tool schemas and text beneath the provider's body-size limit; the coding agent reads it through `resolveInlineImageByteBudget` before sending live history. The budget applies only while `compat.officialEndpoint` holds: a custom `baseUrl` imposes its own body limit.
 
+`prompt-cache-lookback` is how many block positions back from a cache breakpoint the model's prompt cache looks for an earlier request's entry (Claude: 20, counting the breakpoint; a run of consecutive `tool_use` or `tool_result` blocks is one position). The coding agent reads it through `resolvePromptCacheLookback` and keeps warm-cache tool-result pruning inside the window; unassigned means no known bound.
+
 The routed-subscription registry axes (`upstream-rotation`, `region-upstreams-global|us|eu`, `region-limits-eu`, `credit-rates`, `list-price-from`, `routing-family`, `policy-aliases`, `entitlement`, `default-reasoning-off`) describe a gateway whose proxy fans one model out to several upstreams. They are read through `src/compat/factory-droid.ts` by Factory Droid discovery and its request provider, not materialized by `buildModel`. A provider-wide `region-upstreams-*` rule is the upstream serving table and a model rule replaces it for that region. `list-price-from "<provider>" ["<id>"]` shows a bundled row's list price beside the subscription's own billing; unlike seed values it is resolved at runtime and degrades to the seed's zero cost when the row is gone. See [`providers/factory-droid.kdl`](providers/factory-droid.kdl), whose wire and billing pool per model are `api-routes` and `quota-tiers` rules in `runtime/behavior.kdl`.
 
 ### Time-based pricing

package/src/compat/rules/classes/anthropic.kdl

@@ -9,7 +9,13 @@
 // are never undercounted; only Opus/Sonnet/Haiku revisions known to predate
 // 4.7 take the standard tier. Separator-collapsed revisions parse as two
 // digits (`claude-opus-45` is Opus 4.5 → 45), so 10–46 is standard too.
+// Claude's prompt cache finds an earlier request's entry only within 20 block
+// positions back from a breakpoint, counting the breakpoint; a run of
+// consecutive tool_use or tool_result blocks is one position. Warm-cache
+// tool-result pruning keeps its rewrites inside this window.
+// https://platform.claude.com/docs/en/build-with-claude/prompt-caching
 class "anthropic" {
+	prompt-cache-lookback 20
 	image-tokenization {
 		regime "anthropic-patch"
 		max-edge 2576