tweak: rename tail_tokens -> preserve_recent_tokens (#23491)

packages/opencode/src/config/config.ts

@@ -208,8 +208,8 @@ const InfoSchema = Schema.Struct({
         description:
           "Number of recent user turns, including their following assistant/tool responses, to keep verbatim during compaction (default: 2)",
       }),
-      tail_tokens: Schema.optional(NonNegativeInt).annotate({
+      preserve_recent_tokens: Schema.optional(NonNegativeInt).annotate({
-        description: "Token budget for retained recent turn spans during compaction",
+        description: "Maximum number of tokens from recent turns to preserve verbatim after compaction",
       }),
       reserved: Schema.optional(NonNegativeInt).annotate({
         description: "Token buffer for compaction. Leaves enough window to avoid overflow during compaction.",

packages/opencode/src/session/compaction.ts

@@ -34,18 +34,21 @@ export const PRUNE_MINIMUM = 20_000
 export const PRUNE_PROTECT = 40_000
 const PRUNE_PROTECTED_TOOLS = ["skill"]
 const DEFAULT_TAIL_TURNS = 2
-const MIN_TAIL_TOKENS = 2_000
+const MIN_PRESERVE_RECENT_TOKENS = 2_000
-const MAX_TAIL_TOKENS = 8_000
+const MAX_PRESERVE_RECENT_TOKENS = 8_000
 type Turn = {
   start: number
   end: number
   id: MessageID
 }
 
-function tailBudget(input: { cfg: Config.Info; model: Provider.Model }) {
+function preserveRecentBudget(input: { cfg: Config.Info; model: Provider.Model }) {
   return (
-    input.cfg.compaction?.tail_tokens ??
+    input.cfg.compaction?.preserve_recent_tokens ??
-    Math.min(MAX_TAIL_TOKENS, Math.max(MIN_TAIL_TOKENS, Math.floor(usable(input) * 0.25)))
+    Math.min(
+      MAX_PRESERVE_RECENT_TOKENS,
+      Math.max(MIN_PRESERVE_RECENT_TOKENS, Math.floor(usable(input) * 0.25)),
+    )
   )
 }
 
@@ -134,7 +137,7 @@ export const layer: Layer.Layer<
     }) {
       const limit = input.cfg.compaction?.tail_turns ?? DEFAULT_TAIL_TURNS
       if (limit <= 0) return { head: input.messages, tail_start_id: undefined }
-      const budget = tailBudget({ cfg: input.cfg, model: input.model })
+      const budget = preserveRecentBudget({ cfg: input.cfg, model: input.model })
       const all = turns(input.messages)
       if (!all.length) return { head: input.messages, tail_start_id: undefined }
       const recent = all.slice(-limit)

packages/opencode/test/session/compaction.test.ts

@@ -925,7 +925,7 @@ describe("session.compaction.process", () => {
           auto: false,
         })
 
-        const rt = runtime("continue", Plugin.defaultLayer, wide(), cfg({ tail_turns: 2, tail_tokens: 10_000 }))
+        const rt = runtime("continue", Plugin.defaultLayer, wide(), cfg({ tail_turns: 2, preserve_recent_tokens: 10_000 }))
         try {
           const msgs = await svc.messages({ sessionID: session.id })
           const parent = msgs.at(-1)?.info.id
@@ -954,7 +954,7 @@ describe("session.compaction.process", () => {
     })
   })
 
-  test("shrinks retained tail to fit tail token budget", async () => {
+  test("shrinks retained tail to fit preserve token budget", async () => {
     await using tmp = await tmpdir()
     await Instance.provide({
       directory: tmp.path,
@@ -970,7 +970,7 @@ describe("session.compaction.process", () => {
           auto: false,
         })
 
-        const rt = runtime("continue", Plugin.defaultLayer, wide(), cfg({ tail_turns: 2, tail_tokens: 100 }))
+        const rt = runtime("continue", Plugin.defaultLayer, wide(), cfg({ tail_turns: 2, preserve_recent_tokens: 100 }))
         try {
           const msgs = await svc.messages({ sessionID: session.id })
           const parent = msgs.at(-1)?.info.id
@@ -999,7 +999,7 @@ describe("session.compaction.process", () => {
     })
   })
 
-  test("falls back to full summary when even one recent turn exceeds tail budget", async () => {
+  test("falls back to full summary when even one recent turn exceeds preserve token budget", async () => {
     await using tmp = await tmpdir({ git: true })
     const stub = llm()
     let captured = ""
@@ -1021,7 +1021,7 @@ describe("session.compaction.process", () => {
           auto: false,
         })
 
-        const rt = liveRuntime(stub.layer, wide(), cfg({ tail_turns: 1, tail_tokens: 20 }))
+        const rt = liveRuntime(stub.layer, wide(), cfg({ tail_turns: 1, preserve_recent_tokens: 20 }))
         try {
           const msgs = await svc.messages({ sessionID: session.id })
           const parent = msgs.at(-1)?.info.id
@@ -1051,7 +1051,7 @@ describe("session.compaction.process", () => {
     })
   })
 
-  test("falls back to full summary when retained tail media exceeds tail budget", async () => {
+  test("falls back to full summary when retained tail media exceeds preserve token budget", async () => {
     await using tmp = await tmpdir({ git: true })
     const stub = llm()
     let captured = ""
@@ -1082,7 +1082,7 @@ describe("session.compaction.process", () => {
           auto: false,
         })
 
-        const rt = liveRuntime(stub.layer, wide(), cfg({ tail_turns: 1, tail_tokens: 100 }))
+        const rt = liveRuntime(stub.layer, wide(), cfg({ tail_turns: 1, preserve_recent_tokens: 100 }))
         try {
           const msgs = await svc.messages({ sessionID: session.id })
           const parent = msgs.at(-1)?.info.id
@@ -1544,7 +1544,7 @@ describe("session.compaction.process", () => {
           auto: false,
         })
 
-        const rt = liveRuntime(stub.layer, wide(), cfg({ tail_turns: 2, tail_tokens: 10_000 }))
+        const rt = liveRuntime(stub.layer, wide(), cfg({ tail_turns: 2, preserve_recent_tokens: 10_000 }))
         try {
           let msgs = await svc.messages({ sessionID: session.id })
           let parent = msgs.at(-1)?.info.id

packages/sdk/js/src/v2/gen/types.gen.ts

@@ -1648,9 +1648,9 @@ export type Config = {
      */
     tail_turns?: number
     /**
-     * Token budget for retained recent turn spans during compaction
+     * Maximum number of tokens from recent turns to preserve verbatim after compaction
      */
-    tail_tokens?: number
+    preserve_recent_tokens?: number
     /**
      * Token buffer for compaction. Leaves enough window to avoid overflow during compaction.
      */

packages/sdk/openapi.json

@@ -11878,8 +11878,8 @@
                 "minimum": 0,
                 "maximum": 9007199254740991
               },
-              "tail_tokens": {
+              "preserve_recent_tokens": {
-                "description": "Token budget for retained recent turn spans during compaction",
+                "description": "Maximum number of tokens from recent turns to preserve verbatim after compaction",
                 "type": "integer",
                 "minimum": 0,
                 "maximum": 9007199254740991