From ce54d707258d4e90680efb07006a1128d05ebea1 Mon Sep 17 00:00:00 2001
From: James Lal <james@littlebearlabs.io>
Date: Tue, 10 Feb 2026 15:38:54 -0700
Subject: [PATCH] test(compaction): add failing tests for missing headroom when
 limit.input is set

isOverflow() uses `limit.input` directly without subtracting output
token headroom, unlike the `context - output` fallback path. This means
models with an explicit input limit (e.g. Claude with prompt caching)
don't trigger compaction until the full input budget is consumed,
leaving zero room for the next response.

Three tests added:
- Near-boundary case: 198K/200K used, should compact but doesn't
- Parity test: same tokens correctly trigger compaction without limit.input
- Asymmetry test: identical 200K models diverge by exactly 32K (the missing output headroom)

Related: #10634, #8089, #11086, #12621
See also: #6875, #12924
---
 .../opencode/test/session/compaction.test.ts  | 80 +++++++++++++++++++
 1 file changed, 80 insertions(+)

diff --git a/packages/opencode/test/session/compaction.test.ts b/packages/opencode/test/session/compaction.test.ts
index 2e9c091870ed..76852c3f5359 100644
--- a/packages/opencode/test/session/compaction.test.ts
+++ b/packages/opencode/test/session/compaction.test.ts
@@ -112,6 +112,86 @@ describe("session.compaction.isOverflow", () => {
     })
   })
 
+  // ─── Bug reproduction tests ───────────────────────────────────────────
+  // These tests demonstrate that when limit.input is set, isOverflow()
+  // does not subtract any headroom for the next model response. This means
+  // compaction only triggers AFTER we've already consumed the full input
+  // budget, leaving zero room for the next API call's output tokens.
+  //
+  // Compare: without limit.input, usable = context - output (reserves space).
+  // With limit.input, usable = limit.input (reserves nothing).
+  //
+  // Related issues: #10634, #8089, #11086, #12621
+  // Open PRs: #6875, #12924
+
+  test("BUG: no headroom when limit.input is set — compaction should trigger near boundary but does not", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        // Simulate Claude with prompt caching: input limit = 200K, output limit = 32K
+        const model = createModel({ context: 200_000, input: 200_000, output: 32_000 })
+
+        // We've used 198K tokens total. Only 2K under the input limit.
+        // On the next turn, the full conversation (198K) becomes input,
+        // plus the model needs room to generate output — this WILL overflow.
+        const tokens = { input: 180_000, output: 15_000, reasoning: 0, cache: { read: 3_000, write: 0 } }
+        // count = 180K + 3K + 15K = 198K
+        // usable = limit.input = 200K (no output subtracted!)
+        // 198K > 200K = false → no compaction triggered
+
+        // WITHOUT limit.input: usable = 200K - 32K = 168K, and 198K > 168K = true ✓
+        // WITH limit.input: usable = 200K, and 198K > 200K = false ✗
+
+        // With 198K used and only 2K headroom, the next turn will overflow.
+        // Compaction MUST trigger here.
+        expect(await SessionCompaction.isOverflow({ tokens, model })).toBe(true)
+      },
+    })
+  })
+
+  test("BUG: without limit.input, same token count correctly triggers compaction", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        // Same model but without limit.input — uses context - output instead
+        const model = createModel({ context: 200_000, output: 32_000 })
+
+        // Same token usage as above
+        const tokens = { input: 180_000, output: 15_000, reasoning: 0, cache: { read: 3_000, write: 0 } }
+        // count = 198K
+        // usable = context - output = 200K - 32K = 168K
+        // 198K > 168K = true → compaction correctly triggered
+
+        const result = await SessionCompaction.isOverflow({ tokens, model })
+        expect(result).toBe(true) // ← Correct: headroom is reserved
+      },
+    })
+  })
+
+  test("BUG: asymmetry — limit.input model allows 30K more usage before compaction than equivalent model without it", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        // Two models with identical context/output limits, differing only in limit.input
+        const withInputLimit = createModel({ context: 200_000, input: 200_000, output: 32_000 })
+        const withoutInputLimit = createModel({ context: 200_000, output: 32_000 })
+
+        // 170K total tokens — well above context-output (168K) but below input limit (200K)
+        const tokens = { input: 155_000, output: 10_000, reasoning: 0, cache: { read: 5_000, write: 0 } }
+
+        const withLimit = await SessionCompaction.isOverflow({ tokens, model: withInputLimit })
+        const withoutLimit = await SessionCompaction.isOverflow({ tokens, model: withoutInputLimit })
+
+        // Both models have identical real capacity — they should agree:
+        expect(withLimit).toBe(true)    // should compact (170K leaves no room for 32K output)
+        expect(withoutLimit).toBe(true) // correctly compacts (170K > 168K)
+      },
+    })
+  })
+
   test("returns false when model context limit is 0", async () => {
     await using tmp = await tmpdir()
     await Instance.provide({