fix: correct context window percentage and provider token mapping (#306)

The context fullness gauge was driven by output token count, causing
erratic jumps between turns (e.g. 84% -> 64%) with no compaction.

Status bar and estimateContextTokens now use calculatePromptTokens()
which returns input + cacheRead + cacheWrite — the actual input context
size. Previously both used a formula that included the final output token
count, which fluctuates with response length and is not part of the
context window for the current request.

isContextOverflow's usage-based fallback (z.ai silent overflow) was
also missing cacheWrite (cache_creation_input_tokens). Per Anthropic
docs the threshold is input + cache_read + cache_creation — all three.
Ref: https://platform.claude.com/docs/en/about-claude/pricing#long-context-pricing

google.ts and google-vertex.ts were double-counting cached tokens.
Gemini's promptTokenCount already includes cachedContentTokenCount, so
assigning input = promptTokenCount and cacheRead = cachedContentTokenCount
overcounted by cachedContentTokenCount on every cached request. Fixed
by subtracting first, matching the OpenAI convention:
  input = promptTokenCount - cachedContentTokenCount
  cacheRead = cachedContentTokenCount
  => input + cacheRead = promptTokenCount (total prompt, no double-count)
Ref: https://ai.google.dev/api/generate-content#v1beta.GenerateContentResponse.UsageMetadata

All other providers validated: amazon-bedrock (inputTokens is uncached
by API contract), openai-completions/responses/azure (already subtract
cached), kimi/gitlab-duo (delegate to correct implementations), cursor
(API exposes output tokens only — input stays 0 by design).

Co-authored-by: Miroslav Drbal <miroslav.drbal@gendigital.com>

packages/ai/src/providers/google-vertex.ts

@@ -234,11 +234,17 @@ export const streamGoogleVertex: StreamFunction<"google-vertex"> = (
 				}
 
 				if (chunk.usageMetadata) {
+					// promptTokenCount includes cachedContentTokenCount when cached content is used.
+					// Subtract to get non-cached input, matching the OpenAI convention where
+					// input = uncached prompt tokens and cacheRead = cached tokens so that
+					// input + cacheRead = total prompt tokens (no double-counting).
+					// Ref: https://ai.google.dev/api/generate-content#v1beta.GenerateContentResponse.UsageMetadata
+					const cachedTokens = chunk.usageMetadata.cachedContentTokenCount || 0;
 					output.usage = {
-						input: chunk.usageMetadata.promptTokenCount || 0,
+						input: (chunk.usageMetadata.promptTokenCount || 0) - cachedTokens,
 						output:
 							(chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),
-						cacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,
+						cacheRead: cachedTokens,
 						cacheWrite: 0,
 						totalTokens: chunk.usageMetadata.totalTokenCount || 0,
 						cost: {

packages/ai/src/providers/google.ts

@@ -217,11 +217,17 @@ export const streamGoogle: StreamFunction<"google-generative-ai"> = (
 				}
 
 				if (chunk.usageMetadata) {
+					// promptTokenCount includes cachedContentTokenCount when cached content is used.
+					// Subtract to get non-cached input, matching the OpenAI convention where
+					// input = uncached prompt tokens and cacheRead = cached tokens so that
+					// input + cacheRead = total prompt tokens (no double-counting).
+					// Ref: https://ai.google.dev/api/generate-content#v1beta.GenerateContentResponse.UsageMetadata
+					const cachedTokens = chunk.usageMetadata.cachedContentTokenCount || 0;
 					output.usage = {
-						input: chunk.usageMetadata.promptTokenCount || 0,
+						input: (chunk.usageMetadata.promptTokenCount || 0) - cachedTokens,
 						output:
 							(chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),
-						cacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,
+						cacheRead: cachedTokens,
 						cacheWrite: 0,
 						totalTokens: chunk.usageMetadata.totalTokenCount || 0,
 						cost: {

packages/ai/src/utils/overflow.ts

@@ -115,7 +115,7 @@ export function isContextOverflow(message: AssistantMessage, contextWindow?: num
 
 	// Case 2: Usage-based overflow (silent or provider-specific)
 	if (contextWindow) {
-		const inputTokens = message.usage.input + message.usage.cacheRead;
+		const inputTokens = message.usage.input + message.usage.cacheRead + message.usage.cacheWrite;
 		if (inputTokens > contextWindow) {
 			return true;
 		}

packages/coding-agent/src/modes/components/status-line.ts

@@ -7,6 +7,7 @@ import { settings } from "../../config/settings";
 import type { StatusLinePreset, StatusLineSegmentId, StatusLineSeparatorStyle } from "../../config/settings-schema";
 import { theme } from "../../modes/theme/theme";
 import type { AgentSession } from "../../session/agent-session";
+import { calculatePromptTokens } from "../../session/compaction/compaction";
 import { findGitHeadPathSync, sanitizeStatusText } from "../shared";
 import {
 	canReuseCachedPr,
@@ -365,12 +366,7 @@ export class StatusLineComponent implements Component {
 			.reverse()
 			.find(m => m.role === "assistant" && m.stopReason !== "aborted") as AssistantMessage | undefined;
 
-		const contextTokens = lastAssistantMessage
-			? lastAssistantMessage.usage.input +
-				lastAssistantMessage.usage.output +
-				lastAssistantMessage.usage.cacheRead +
-				lastAssistantMessage.usage.cacheWrite
-			: 0;
+		const contextTokens = lastAssistantMessage ? calculatePromptTokens(lastAssistantMessage.usage) : 0;
 		const contextWindow = state.model?.contextWindow || 0;
 		const contextPercent = contextWindow > 0 ? (contextTokens / contextWindow) * 100 : 0;
 

packages/coding-agent/src/session/agent-session.ts

@@ -108,6 +108,7 @@ import { extractFileMentions, generateFileMentionMessages } from "../utils/file-
 import {
 	type CompactionResult,
 	calculateContextTokens,
+	calculatePromptTokens,
 	collectEntriesForBranchSummary,
 	compact,
 	estimateTokens,
@@ -5076,7 +5077,7 @@ Be thorough - include exact file paths, function names, error messages, and tech
 			};
 		}
 
-		const usageTokens = calculateContextTokens(lastUsage);
+		const usageTokens = calculatePromptTokens(lastUsage);
 		let trailingTokens = 0;
 		for (let i = lastUsageIndex + 1; i < messages.length; i++) {
 			trailingTokens += estimateTokens(messages[i]);