Merge branch 'dev' into refactor/one-tool-to-rule-them-all

Author: Tarquinen

lib/hooks.ts

@@ -115,9 +115,7 @@ export function createChatMessageTransformHandler(
         purgeErrors(state, logger, config, output.messages)
 
         prune(state, logger, config, output.messages)
-
         insertCompressToolContext(state, config, logger, output.messages)
-
         insertMessageIdContext(state, config, output.messages)
 
         applyPendingManualTriggerPrompt(state, output.messages, logger)

lib/prompts/system.md

@@ -8,14 +8,18 @@ OPERATING STANCE
 Prefer short, closed, summary-safe ranges.
 When multiple independent stale ranges exist, prefer several short compressions (in parallel when possible) over one large-range compression.
 
-Use `compress` as steady housekeeping while you work.
+<compress>THE COMPRESS TOOL
+`compress` is a sledgehammer and should be used accordingly. It's purpose is to reduce whole part of the conversation to its essence and technical details in order to leave room for newer context. Your summary MUST be technical and specific enough to preserve FULL understanding of WHAT TRANSPIRED, such that NO AMBIGUITY remains about what was done, found, or decided. Your compress summary must be thorough and precise. `compress` will replace everything in the range you match, user and assistant messages, tool inputs and outputs. It is preferred to not compress preemptively, but rather wait for natural breakpoints in the conversation. Those breakpoints are to be infered from user messages. You WILL NOT compress based on thinking that you are done with the task, wait for conversation queues that the user has moved on from current phase. Use injected boundary IDs (`startId`/`endId`) to select ranges.
+
+Injected boundary IDs are surfaced as XML tags in conversation context, e.g. `<dcp-message-id>m0001</dcp-message-id>` for message IDs and `<dcp-message-id>b3</dcp-message-id>` for compressed blocks. These IDs are internal boundary markers for `compress` only. Do not reference, explain, or surface these IDs in normal user-facing responses unless you are actively constructing a `compress` tool call.
 
 CADENCE, SIGNALS, AND LATENCY
 
 - No fixed threshold mandates compression
 - Prioritize closedness and independence over raw range size
 - Prefer smaller, regular compressions over infrequent massive compressions for better latency and summary quality
 - When multiple independent stale ranges are ready, batch compressions in parallel
+  </compress>
 
 BOUNDARY MATCHING
 `compress` uses inclusive ID boundaries via `content.startId` and `content.endId`. IDs are injected in context as message refs (`mNNNN`) and compressed block refs (`bN`).

lib/state/persistence.ts

@@ -114,22 +114,57 @@ export async function loadSessionState(
         }
 
         if (Array.isArray(state.compressSummaries)) {
-            const validSummaries = state.compressSummaries.filter(
-                (s): s is CompressSummary =>
-                    s !== null &&
-                    typeof s === "object" &&
-                    typeof s.blockId === "number" &&
-                    typeof s.anchorMessageId === "string" &&
-                    typeof s.summary === "string",
-            )
-            if (validSummaries.length !== state.compressSummaries.length) {
+            const migratedSummaries: CompressSummary[] = []
+            let nextBlockId = 1
+
+            for (const entry of state.compressSummaries) {
+                if (
+                    entry === null ||
+                    typeof entry !== "object" ||
+                    typeof entry.anchorMessageId !== "string" ||
+                    typeof entry.summary !== "string"
+                ) {
+                    continue
+                }
+
+                const blockId =
+                    typeof entry.blockId === "number" && Number.isInteger(entry.blockId)
+                        ? entry.blockId
+                        : nextBlockId
+                migratedSummaries.push({
+                    blockId,
+                    anchorMessageId: entry.anchorMessageId,
+                    summary: entry.summary,
+                })
+                nextBlockId = Math.max(nextBlockId, blockId + 1)
+            }
+
+            if (migratedSummaries.length !== state.compressSummaries.length) {
                 logger.warn("Filtered out malformed compressSummaries entries", {
                     sessionId: sessionId,
                     original: state.compressSummaries.length,
-                    valid: validSummaries.length,
+                    valid: migratedSummaries.length,
+                })
+            }
+
+            const seenBlockIds = new Set<number>()
+            const dedupedSummaries = migratedSummaries.filter((summary) => {
+                if (seenBlockIds.has(summary.blockId)) {
+                    return false
+                }
+                seenBlockIds.add(summary.blockId)
+                return true
+            })
+
+            if (dedupedSummaries.length !== migratedSummaries.length) {
+                logger.warn("Removed duplicate compress block IDs", {
+                    sessionId: sessionId,
+                    original: migratedSummaries.length,
+                    valid: dedupedSummaries.length,
                 })
             }
-            state.compressSummaries = validSummaries
+
+            state.compressSummaries = dedupedSummaries
         } else {
             state.compressSummaries = []
         }

lib/state/state.ts

@@ -83,6 +83,7 @@ export function createSessionState(): SessionState {
             byRef: new Map<string, string>(),
             nextRef: 0,
         },
+
         lastCompaction: 0,
         currentTurn: 0,
         variant: undefined,
@@ -114,6 +115,7 @@ export function resetSessionState(state: SessionState): void {
         byRef: new Map<string, string>(),
         nextRef: 0,
     }
+
     state.lastCompaction = 0
     state.currentTurn = 0
     state.variant = undefined

lib/state/types.ts

@@ -56,6 +56,7 @@ export interface SessionState {
     toolParameters: Map<string, ToolParameterEntry>
     toolIdList: string[]
     messageIds: MessageIdState
+
     lastCompaction: number
     currentTurn: number
     variant: string | undefined

lib/state/utils.ts

@@ -90,6 +90,11 @@ export function resetOnCompaction(state: SessionState): void {
     state.prune.tools = new Map<string, number>()
     state.prune.messages = new Map<string, number>()
     state.compressSummaries = []
+    state.messageIds = {
+        byRawId: new Map<string, string>(),
+        byRef: new Map<string, string>(),
+        nextRef: 0,
+    }
     state.contextLimitAnchors = new Set<string>()
     state.softNudgeAnchors = new Set<string>()
 }

lib/tools/compress-utils.ts

@@ -144,22 +144,21 @@ export function resolveBoundaryIds(
         throwCombinedIssues(issues)
     }
 
-    if (!parsedStartId || !parsedEndId) {
-        throw new Error("Invalid boundary ID(s)")
-    }
+    const validStartId = parsedStartId as NonNullable<typeof parsedStartId>
+    const validEndId = parsedEndId as NonNullable<typeof parsedEndId>
 
-    const startReference = lookup.get(parsedStartId.ref)
-    const endReference = lookup.get(parsedEndId.ref)
+    const startReference = lookup.get(validStartId.ref)
+    const endReference = lookup.get(validEndId.ref)
 
     if (!startReference) {
         issues.push(
-            `startId ${parsedStartId.ref} is not available in the current conversation context. Choose an injected ID visible in context.`,
+            `startId ${validStartId.ref} is not available in the current conversation context. Choose an injected ID visible in context.`,
         )
     }
 
     if (!endReference) {
         issues.push(
-            `endId ${parsedEndId.ref} is not available in the current conversation context. Choose an injected ID visible in context.`,
+            `endId ${validEndId.ref} is not available in the current conversation context. Choose an injected ID visible in context.`,
         )
     }
 
@@ -173,7 +172,7 @@ export function resolveBoundaryIds(
 
     if (startReference.rawIndex > endReference.rawIndex) {
         throw new Error(
-            `startId ${parsedStartId.ref} appears after endId ${parsedEndId.ref} in the conversation. Start must come before end.`,
+            `startId ${validStartId.ref} appears after endId ${validEndId.ref} in the conversation. Start must come before end.`,
         )
     }
 

lib/tools/compress.ts

@@ -1,4 +1,5 @@
 import { tool } from "@opencode-ai/plugin"
+import type { BoundaryReference } from "./compress-utils"
 import type { ToolContext } from "./types"
 import { COMPRESS_TOOL_SPEC } from "../prompts"
 import { ensureSessionInitialized } from "../state"
@@ -22,6 +23,7 @@ import {
 import { getCurrentParams, getCurrentTokenUsage } from "../strategies/utils"
 import { saveSessionState } from "../state/persistence"
 import { sendCompressNotification } from "../ui/notification"
+import { assignMessageRefs } from "../message-ids"
 
 export function createCompressTool(ctx: ToolContext): ReturnType<typeof tool> {
     return tool({
@@ -72,6 +74,8 @@ export function createCompressTool(ctx: ToolContext): ReturnType<typeof tool> {
                 ctx.config.manualMode.enabled,
             )
 
+            assignMessageRefs(ctx.state, rawMessages)
+
             const searchContext = buildSearchContext(ctx.state, rawMessages)
 
             const { startReference, endReference } = resolveBoundaryIds(
@@ -142,3 +146,17 @@ export function createCompressTool(ctx: ToolContext): ReturnType<typeof tool> {
         },
     })
 }
+
+function getBoundaryMessageId(reference: BoundaryReference): string {
+    if (reference.kind === "message") {
+        if (!reference.messageId) {
+            throw new Error("Failed to map boundary matches back to raw messages")
+        }
+        return reference.messageId
+    }
+
+    if (!reference.anchorMessageId) {
+        throw new Error("Failed to map boundary matches back to raw messages")
+    }
+    return reference.anchorMessageId
+}

package-lock.json

@@ -1,7 +1,7 @@
 {
     "$schema": "https://json.schemastore.org/package.json",
     "name": "@tarquinen/opencode-dcp",
-    "version": "2.1.3",
+    "version": "2.1.6",
     "type": "module",
     "description": "OpenCode plugin that optimizes token usage by pruning obsolete tool outputs from conversation context",
     "main": "./dist/index.js",
@@ -44,7 +44,6 @@
     "dependencies": {
         "@anthropic-ai/tokenizer": "^0.0.4",
         "@opencode-ai/sdk": "^1.1.48",
-        "fuzzball": "^2.2.3",
         "jsonc-parser": "^3.3.1",
         "zod": "^4.3.6"
     },