fix(cline-cli): stop billing cache tokens twice (#1675)

Cline CLI's metrics.inputTokens is the full prompt, cache reads and writes
included; Cline's own pricing bills max(0, input - cacheRead - cacheWrite)
at the input rate. We treated it as uncached input, so every cache read was
also charged at the full input rate. The split now happens in parseMetrics,
covering per-message and rollup calls.

Bumps the cline-cli parse version so cached calls re-parse and the daily
cache to v63 so settled days re-derive. Call counts are unchanged.

src/daily-cache.ts

@@ -304,7 +304,11 @@ import type { DateRange, ProjectSummary } from './types.js'
 // v62: Devin's SWE-2 prices at Cognition's list rate and swe-1-7-lightning as
 // swe-1.7-lightning instead of $0. Only cost rises; call counts are unchanged,
 // so no PENDING_REDERIVE_PROVIDER_VERSIONS entry is needed.
-export const DAILY_CACHE_VERSION = 62
+// v63: Cline CLI input tokens include cache reads and writes, which were then
+// billed again at the cache rates. Only input tokens and estimated cost fall;
+// call counts are unchanged, so no PENDING_REDERIVE_PROVIDER_VERSIONS entry is
+// needed.
+export const DAILY_CACHE_VERSION = 63
 const MIN_SUPPORTED_VERSION = 28
 
 /// Providers whose per-day CALL COUNT means something different at

src/providers/cline-cli.ts

@@ -89,11 +89,15 @@ type ClineCliMetrics = {
 
 function parseMetrics(value: unknown): ClineCliMetrics | null {
   if (!isRecord(value)) return null
+  const cacheReadTokens = safeTokenCount(value['cacheReadTokens'])
+  const cacheWriteTokens = safeTokenCount(value['cacheWriteTokens'])
   const metrics: ClineCliMetrics = {
-    inputTokens: safeTokenCount(value['inputTokens']),
+    // Cline's inputTokens is the full prompt, cache read and write included
+    // (its own pricing bills max(0, input - cacheRead - cacheWrite) at the input rate).
+    inputTokens: Math.max(0, safeTokenCount(value['inputTokens']) - cacheReadTokens - cacheWriteTokens),
     outputTokens: safeTokenCount(value['outputTokens']),
-    cacheReadTokens: safeTokenCount(value['cacheReadTokens']),
-    cacheWriteTokens: safeTokenCount(value['cacheWriteTokens']),
+    cacheReadTokens,
+    cacheWriteTokens,
     cost: safeNonNegativeNumber(value['cost']),
     // A negative cost is not a credit we can represent — treat it as absent and
     // fall back to token pricing, rather than reporting a clamped $0 as metered.

src/session-cache.ts

@@ -377,7 +377,9 @@ export const PROVIDER_PARSE_VERSIONS: Record<string, string> = {
   // reported-cost-v1: the CLI reports its own per-message cost, so entries
   // cached before cline-cli joined the reported-cost allowlist in parser.ts
   // hold costUSD: undefined and get re-priced from tokens on every read.
-  'cline-cli': 'reported-cost-v1-est-reprice-v1',
+  // cache-inclusive-input-v1: cached calls hold inputTokens with cache reads
+  // and writes still inside it, so they must re-parse.
+  'cline-cli': 'reported-cost-v1-est-reprice-v1-cache-inclusive-input-v1',
   codewhale: 'aggregate-session-v1-est-cost',
   'command-code': 'cache-inclusive-input-v1',
   // Bump when the Codex parser changes attribution so unchanged, already-cached

tests/providers/cline-cli.test.ts

@@ -4,6 +4,7 @@ import { join } from 'path'
 import { tmpdir } from 'os'
 
 import { clineCli, createClineCliProvider, getClineCliSessionsDir } from '../../src/providers/cline-cli.js'
+import { calculateCost } from '../../src/models.js'
 import type { ParsedProviderCall } from '../../src/providers/types.js'
 
 let tmpDir: string
@@ -211,7 +212,7 @@ describe('cline-cli provider - parsing', () => {
     const calls = await collect(tmpDir)
 
     expect(calls).toHaveLength(2)
-    expect(calls.map(c => c.inputTokens)).toEqual([100, 200])
+    expect(calls.map(c => c.inputTokens)).toEqual([93, 200])
     expect(calls.map(c => c.outputTokens)).toEqual([10, 20])
     expect(calls[0]?.cacheReadInputTokens).toBe(5)
     expect(calls[0]?.cacheCreationInputTokens).toBe(2)
@@ -328,6 +329,22 @@ describe('cline-cli provider - parsing', () => {
     expect(call?.costUSD).toBeGreaterThan(0)
   })
 
+  it('takes cache reads out of the cache-inclusive input before estimating', async () => {
+    await writeSession(tmpDir, 'sess-a', {
+      messages: [{
+        role: 'assistant', text: 'a', model: 'cline-pass/deepseek-v4.1-flash',
+        metrics: { inputTokens: 14689, outputTokens: 220, cacheReadTokens: 11136, cacheWriteTokens: 0 },
+      }],
+    })
+
+    const [call] = await collect(tmpDir)
+
+    expect(call?.inputTokens).toBe(3553)
+    expect(call?.cacheReadInputTokens).toBe(11136)
+    expect(call?.costUSD).toBeCloseTo(calculateCost('cline-pass/deepseek-v4.1-flash', 3553, 220, 0, 11136, 0), 12)
+    expect(call?.costUSD).toBeGreaterThan(0)
+  })
+
   it('keeps a metered $0 cost reported instead of re-estimating it', async () => {
     await writeSession(tmpDir, 'sess-a', {
       messages: [{ role: 'assistant', text: 'a', metrics: { inputTokens: 1000, outputTokens: 100, cost: 0 } }],
@@ -428,7 +445,7 @@ describe('cline-cli provider - rollup fallback', () => {
     const calls = await collect(tmpDir)
 
     expect(calls).toHaveLength(1)
-    expect(calls[0]?.inputTokens).toBe(5483)
+    expect(calls[0]?.inputTokens).toBe(5433)
     expect(calls[0]?.outputTokens).toBe(133)
     expect(calls[0]?.cacheReadInputTokens).toBe(50)
     expect(calls[0]?.costUSD).toBeCloseTo(0.0081984, 7)