pi-mega-compact 0.6.0 → 0.6.2

package/README.md

@@ -6,7 +6,7 @@ sessions into a **local SQLite store** and offers **deduped inline recall** —
 running **locally inside the extension**, with **no remote MCP server** and
 **zero network calls at runtime** (PREVENT-PI-004).
 
-> **Current version:** `v0.6.0` — storage backend is **`node:sqlite`**
+> **Current version:** `v0.6.1` — storage backend is **`node:sqlite`**
 > (`DatabaseSync`, a Node ≥22.13 built-in), replacing the old `better-sqlite3`
 > native addon and the per-session gzipped JSON checkpoint files. **Zero native
 > build step, fully local, zero network at runtime.** Legacy

package/dist/extensions/mega-compact.test.js

@@ -408,6 +408,54 @@ test("explicit MEGACOMPACT_THRESHOLD_TOKENS overrides the tier", async () => {
     delete process.env.MEGACOMPACT_TIER;
     assert.ok(h.notifies.some((n) => n.includes("preset=custom") && n.includes("threshold=777")), "explicit threshold wins over tier (preset=custom)");
 });
+// ---- S24: memory review tied to pressure / compaction -----------------------
+// Build a decision-bearing session large enough to guarantee a real (non-skipped,
+// non-deduped) compaction. Each user turn contains a decision phrase
+// (/\bactually\b/i, /\bwe (?:use|decided)\b/i) so reviewConversation yields ops.
+function decisionSession() {
+    const out = [];
+    for (let i = 0; i < 14; i++) {
+        out.push({ role: "user", content: `actually we decided to use approach ${i} for module ${i}`, timestamp: i });
+        out.push({ role: "assistant", content: [{ type: "toolCall", name: "Edit", id: `c${i}`, arguments: {} }], api: "anthropic-messages", provider: "anthropic", model: "m", usage: { inputTokens: 1, outputTokens: 1, cacheReadTokens: 0, cacheWriteTokens: 0 }, stopReason: "tool_use", timestamp: i });
+        out.push({ role: "toolResult", content: [{ type: "text", text: `edited module ${i}` }], toolCallId: `c${i}`, toolName: "Edit", isError: false, timestamp: i });
+    }
+    return out;
+}
+test("S24: high pressure triggers a memory review on compaction", async () => {
+    const h = harness();
+    // Force a real (non-legacy) compaction at full pressure → pressureBand "mega",
+    // which must fire the shared runMemoryReview on compact (review-on-compact).
+    process.env.MEGACOMPACT_LEGACY_DURABLE_TRIM = "false";
+    try {
+        const messages = decisionSession();
+        const ctx = h.ctx({ getContextUsage: () => ({ tokens: 200000, contextWindow: 200000, percent: 100 }) });
+        await h.fire("context", { type: "context", messages }, ctx);
+        // review-on-compact runs as a fire-and-forget async (doCompact is sync), so
+        // let the microtask/macrotask queue drain before asserting the side effect.
+        await new Promise((r) => setTimeout(r, 20));
+        const { listMemories, listCheckpoints } = await import("../src/store/sqlite.js");
+        // A checkpoint must have been persisted (proves compaction ran, not skipped).
+        assert.ok(listCheckpoints("sess_ext_001", h.stateDir).length > 0, "checkpoint persisted to local vector db");
+        // The just-compacted region is worth remembering, so durable memories must
+        // have been written to the SQLite store (review-on-compact path).
+        const mem = listMemories(null, 50, h.stateDir);
+        assert.ok(mem.length > 0, "memory review wrote durable memories on compact");
+    }
+    finally {
+        delete process.env.MEGACOMPACT_LEGACY_DURABLE_TRIM;
+    }
+});
+test("S24: /mega-status reports the live pressure band + %", async () => {
+    const h = harness();
+    // Populate the runtime's live context first (a context event sets
+    // lastCtxTokens/lastCtxPercent), then read /mega-status. At 100% usage the live
+    // band must read "mega" and pressure must report 100%.
+    const ctx = h.ctx({ getContextUsage: () => ({ tokens: 200000, contextWindow: 200000, percent: 100 }) });
+    await h.fire("context", { type: "context", messages: h.session }, ctx);
+    await h.commands["mega-status"].handler("", ctx);
+    assert.ok(h.notifies.some((n) => n.includes("tier=mega (live)")), "live band reported as mega at 100% pressure");
+    assert.ok(h.notifies.some((n) => n.includes("pressure=100%")), "live pressure % reported");
+});
 // ---- /dashboard commands ----------------------------------------------------
 test("/dashboard-status reports no server when pid file missing", async () => {
     // Private base so this asserts "no server" on a range nothing else uses,

package/dist/extensions/mega-conflict-cmds.js

@@ -9,6 +9,8 @@
 import { detectConflicts } from "./conflict-scan.js";
 import { addMemory, listMemories, searchMemories, recallMemory } from "../src/store/sqlite.js";
 import { resolveRepoRoot } from "./mega-config.js";
+import { defaultEmbedder } from "../src/embedder.js";
+import { upsertMemoryEmbedding } from "../src/store/memoryIndex.js";
 /** Run the conflict scan and format a human-readable report. */
 export function validateExtensions() {
     const report = detectConflicts();
@@ -74,6 +76,14 @@ export function registerConflictCommands(pi, runtime) {
                 const tagMatches = [...text.matchAll(/#([\w-]+)/g)].map((m) => m[1]);
                 const content = text.replace(/#[\w-]+/g, "").trim();
                 const id = addMemory({ content, tags: tagMatches }, repo, runtime.currentStateDir);
+                // S24: mirror into the cross-repo memory index (fire-and-forget).
+                try {
+                    const vec = defaultEmbedder().embed(content);
+                    void upsertMemoryEmbedding(repo, id, content, vec);
+                }
+                catch {
+                    /* non-fatal */
+                }
                 ctx.ui.notify(`[mega-memory] saved #${id} to ${repo.split(/[\\/]/).pop()}`);
                 return;
             }
@@ -142,6 +152,13 @@ export function registerConflictCommands(pi, runtime) {
                 const tagMatches = [...text.matchAll(/#([\w-]+)/g)].map((m) => m[1]);
                 const content = text.replace(/#[\w-]+/g, "").trim();
                 const id = addMemory({ content, tags: tagMatches }, repo, runtime.currentStateDir);
+                try {
+                    const vec = defaultEmbedder().embed(content);
+                    void upsertMemoryEmbedding(repo, id, content, vec);
+                }
+                catch {
+                    /* non-fatal */
+                }
                 ctx.ui.notify(`[/m] saved #${id} to ${repo.split(/[\\/]/).pop()}`);
                 return;
             }

package/dist/extensions/mega-events.js

@@ -10,13 +10,22 @@ import { normalizeSessionId } from "../src/store.js";
 import { autoCompactCheck } from "../src/compact.js";
 import { estimateSessionTokens } from "../src/tokens.js";
 import { recentUserQuery, WIDGET_KEY } from "./mega-runtime.js";
-import { runCompact, doRecall, doRecallAsync, piCompactWouldNoop } from "./mega-pipeline.js";
+import { runCompact, doRecall, doRecallAsync, piCompactWouldNoop, runMemoryReview } from "./mega-pipeline.js";
 import { recallMemoriesAndInline } from "../src/recall.js";
 import { driveNativeCompaction } from "./mega-compact-driver.js";
 import { computeLiveTrimCut, liveTrimSummaryMessage } from "./mega-trim.js";
 import { pressureFromPct, memoryReviewCadence } from "./mega-config.js";
+/**
+ * DIAG accessor for the headless test harness: the most recently constructed
+ * MegaRuntime, so a test that loads the compiled extension via its default
+ * export can read diag counters (diagLiveTrimFires / diagBeforeCompactFires /
+ * diagBeforeCompactSupplied / diagAgentEndIdle) after firing synthetic events.
+ * No-op in production — nothing reads this outside tests.
+ */
+export let lastRuntime;
 /** Register all pi lifecycle event handlers. */
 export function registerEventHandlers(pi, runtime, config) {
+    lastRuntime = runtime;
     // ---- Session lifecycle (state reset points) -------------------------------
     // Capture model/provider whenever it changes (drives real cost estimation).
     pi.on("model_select", async (_event, ctx) => {
@@ -56,6 +65,8 @@ export function registerEventHandlers(pi, runtime, config) {
             try {
                 const mr = await recallMemoriesAndInline({
                     query, stateDir: runtime.getStateDir(), limit: 5,
+                    crossRepo: config.crossRepoEnabled,
+                    crossRepoCosine: config.crossRepoCosine,
                 });
                 if (!mr.empty)
                     runtime.pendingMemoryRecallBlock = mr.block;
@@ -82,7 +93,7 @@ export function registerEventHandlers(pi, runtime, config) {
                 }
                 // S21: parallel memory recall. Trigram embedder is sub-ms; await is fine.
                 try {
-                    const mr = await recallMemoriesAndInline({ query, stateDir: runtime.getStateDir(), limit: 5 });
+                    const mr = await recallMemoriesAndInline({ query, stateDir: runtime.getStateDir(), limit: 5, crossRepo: config.crossRepoEnabled, crossRepoCosine: config.crossRepoCosine });
                     if (!mr.empty)
                         runtime.pendingMemoryRecallBlock = mr.block;
                 }
@@ -139,6 +150,46 @@ export function registerEventHandlers(pi, runtime, config) {
                 const idle = ctx.isIdle?.() ?? true;
                 const queued = ctx.hasPendingMessages?.() ?? false;
                 const now = Date.now();
+                // DIAG (team-run relief): surface whether the agent is idle + over
+                // threshold at agent_end so we can see if a mid-run durable-trim trigger
+                // *should* have fired but didn't.
+                const overThreshold = (runtime.lastCtxTokens ?? 0) >= config.thresholdTokens;
+                runtime.diagAgentEndIdle++;
+                runtime.logger.info("agent-end-idle", {
+                    sessionId: runtime.rt.sessionId,
+                    idle,
+                    queued,
+                    overThreshold,
+                    ctxPct: runtime.lastCtxPercent,
+                    ctxTokens: runtime.lastCtxTokens,
+                    thresholdTokens: config.thresholdTokens,
+                    wouldNudge: idle && queued && now >= runtime.resumeNudgeUntil,
+                });
+                // S16+S24: MID-RUN DURABLE TRIM. During a long team run (sub-agents),
+                // pi's native durable compaction only fires from _checkCompaction at
+                // PARENT settle (agent-session.js:760/844), so the on-disk transcript +
+                // context meter balloon to ~150k and never relieve until the very end
+                // ("compacts but doesn't resume"). agent_end with activeAgents===0 is a
+                // SAFE, settled point: calling ctx.compact() here does NOT abort an
+                // in-flight turn (the S16 danger is only mid-turn). ctx.compact() runs
+                // pi's flow, which fires our session_before_compact handler to supply
+                // the durable trim (truncates the transcript from firstKeptEntryId).
+                // Guarded three ways: only when truly idle + over threshold, only when
+                // pi would actually compact (piCompactWouldNoop skips the user-facing
+                // no-op throw), and debounced (one durable trim per 2s) to avoid
+                // thrashing the transcript while sub-agents keep settling.
+                if (idle && overThreshold && now >= runtime.debounceUntil) {
+                    if (!piCompactWouldNoop(ctx)) {
+                        runtime.debounceUntil = now + 2000;
+                        runtime.diagAgentEndDurable++;
+                        runtime.logger.info("agent-end-durable-trigger", {
+                            sessionId: runtime.rt.sessionId,
+                            ctxTokens: runtime.lastCtxTokens,
+                            thresholdTokens: config.thresholdTokens,
+                        });
+                        ctx.compact({ customInstructions: undefined }); // guardrails-allow PREVENT-PI-004: local ctx.compact() — no network; agent settled so no in-flight abort
+                    }
+                }
                 if (idle && queued && now >= runtime.resumeNudgeUntil) {
                     runtime.resumeNudgeUntil = now + 30_000;
                     pi.sendUserMessage("[mega-compact] continue from the compacted context above.");
@@ -166,23 +217,13 @@ export function registerEventHandlers(pi, runtime, config) {
         if (config.memoryAutoReview && runtime.currentTurn > 0) {
             const cadence = memoryReviewCadence(runtime.pressureBand, config.memoryReviewInterval);
             if (runtime.currentTurn % cadence === 0) {
-                try {
-                    const { reviewConversation } = await import("../src/memory.js");
-                    const { applyMemoryOps } = await import("../src/memoryOps.js");
-                    const entries = ctx.sessionManager.getEntries();
-                    const view = runtime.engineView(entries.flatMap((e) => (e.message ? [e.message] : [])));
-                    const ops = reviewConversation(view, []);
-                    if (ops.length) {
-                        await applyMemoryOps(ops, runtime.currentStateDir);
-                        // S21.2: a memory op landed in this turn window. The pipeline reads
-                        // this counter after a successful compaction and fires
-                        // `consolidateMemories` only when it's > 0.
-                        runtime.memoriesTouchedThisCompaction += ops.length;
-                    }
-                }
-                catch {
-                    /* non-fatal — auto-review must not break the turn loop */
-                }
+                // S20+S24: review the conversation and persist durable memories. The
+                // cadence scales with pressure (memoryReviewCadence): as context fills,
+                // the conversation is reviewed more often so memories keep pace with
+                // faster churn. Shared runMemoryReview body (also used on compact).
+                const entries = ctx.sessionManager.getEntries();
+                const view = runtime.engineView(entries.flatMap((e) => (e.message ? [e.message] : [])));
+                await runMemoryReview(runtime, view, "turn");
             }
         }
     });
@@ -217,22 +258,30 @@ export function registerEventHandlers(pi, runtime, config) {
         const currentTokens = usage?.tokens ?? estimateSessionTokens(view) ??
             Math.round((pct / 100) * (usage?.contextWindow ?? 0));
         // FAST GATE: token-based (tier threshold), not percentage-based.
-        if (currentTokens < config.thresholdTokens)
+        if (currentTokens < config.thresholdTokens) {
+            runtime.diagCtxFastGate++;
             return;
+        }
         const check = autoCompactCheck(currentTokens, config.thresholdTokens); // SERVER-STYLE CONFIRM (local)
-        if (!check.shouldCompact)
+        if (!check.shouldCompact) {
+            runtime.diagCtxNoCompact++;
             return;
+        }
         // Debounce so we don't fire on every context event past threshold.
         const now = Date.now();
-        if (now < runtime.debounceUntil)
+        if (now < runtime.debounceUntil) {
+            runtime.diagCtxDebounce++;
             return;
+        }
         runtime.debounceUntil = now + 2000;
         // Adaptive compression (Fix E): scale compression strength + keepFrom depth
         // with how close we are to the model context limit.
         const pressure = pressureFromPct(pct);
         const ran = runCompact(pi, runtime, config, ctx, messages, { compressionPressure: pressure });
-        if (ran.skipped)
+        if (ran.skipped) {
+            runtime.diagCtxRunSkipped++;
             return;
+        }
         // LEGACY path (rollback): v0.4.28 ctx.compact() + the no-op gate. The
         // manual compact path aborts the in-flight turn — only used behind the flag.
         // Read live from env (in addition to the load-time config) so the flag can be
@@ -264,8 +313,16 @@ export function registerEventHandlers(pi, runtime, config) {
                 summary: ran.result.summary,
                 anchorUserMessages,
             });
-            if (cut === null)
+            if (cut === null) {
+                runtime.diagCtxCutNull++;
+                runtime.logger.info("live-trim-skip", {
+                    sessionId: runtime.rt.sessionId,
+                    compactedFrom: ran.result.compactedFrom,
+                    viewLen: view.length,
+                    anchorUserMessages,
+                });
                 return; // unsafe / below anchor floor — no trim this call
+            }
             const summaryMsg = liveTrimSummaryMessage({
                 compactedFrom: ran.result.compactedFrom,
                 summary: ran.result.summary,
@@ -279,9 +336,23 @@ export function registerEventHandlers(pi, runtime, config) {
             };
             const recent = messages.slice(cut); // guardrails-allow PREVENT-PI-002: `cut` is the pre-sanitized `compactedFrom` produced by src/boundary.ts computeDropRange, so the preserved run begins on a toolPair-safe index.
             runtime.snapshot(ctx);
+            // DIAG (team-run relief): confirm the live trim actually fires + how big
+            // the window still is. The return is non-durable (per-LLM-call only), so
+            // this is the signal that the model is being fed a compacted view while
+            // the on-disk transcript + context meter keep growing.
+            runtime.diagLiveTrimFires++;
+            runtime.logger.info("live-trim", {
+                sessionId: runtime.rt.sessionId,
+                inputMsgs: messages.length,
+                outputMsgs: recent.length + 1,
+                compactedFrom: cut,
+                ctxPct: pct,
+                ctxTokens: usage?.tokens ?? null,
+            });
             return { messages: [summaryAgentMsg, ...recent] };
         }
         catch {
+            runtime.diagCtxThrown++;
             return; // non-fatal: no trim this call; the next context event retries
         }
     });
@@ -293,11 +364,26 @@ export function registerEventHandlers(pi, runtime, config) {
     // there is no full-reload + additive recall inflation.
     pi.on("session_before_compact", async (event, ctx) => {
         runtime.resetRuntime(ctx.sessionManager.getSessionId());
+        // DIAG (team-run relief): this is the ONLY durable-trim entry point. Log
+        // every fire + whether we supplied a compaction (truncates transcript) or
+        // fell through to {} (pi runs its own). If this is sparse during a team
+        // run, the durable trim is firing too late (only at parent settle).
+        const prep = event.preparation;
+        runtime.diagBeforeCompactFires++;
+        runtime.logger.info("before-compact-entry", {
+            sessionId: runtime.rt.sessionId,
+            reason: event.reason,
+            hasPrep: !!prep,
+            msgsToSummarize: prep?.messagesToSummarize?.length ?? 0,
+            firstKeptEntryId: prep?.firstKeptEntryId ?? null,
+            activeAgents: runtime.activeAgents,
+        });
         if (!config.auto)
             return {}; // let pi run its own native compaction
         try {
             const result = driveNativeCompaction(event, runtime, config);
             if (result) {
+                runtime.diagBeforeCompactSupplied++;
                 runtime.logger.info("native-compact", {
                     sessionId: runtime.rt.sessionId,
                     firstKeptEntryId: result.compaction.firstKeptEntryId,

package/dist/extensions/mega-pipeline.js

@@ -18,6 +18,35 @@ import { resolveRepoRoot, preserveRecentForPressure } from "./mega-config.js";
 import { runRaptor } from "../src/dedup/raptor/index.js";
 import { loadDedupConfig } from "../src/config/dedup.js";
 import { upsertEmbedding as indexUpsertEmbedding } from "../src/store/vectorIndex.js";
+/**
+ * Review the live conversation and persist durable memories (S20+S24). Shared by
+ * the pressure-scaled turn-end cadence (mega-events.ts) AND review-on-compact
+ * (below) so both paths run the identical review body. Best-effort + non-fatal:
+ * a review failure is swallowed and never breaks the caller. On success, the
+ * number of applied ops is returned so callers can feed the consolidation gate.
+ *
+ * @param view the engine message view to review (caller builds it)
+ * @param label a short source tag for the ticker line (e.g. "pressure" / "turn")
+ */
+export async function runMemoryReview(runtime, view, label) {
+    try {
+        const { reviewConversation } = await import("../src/memory.js");
+        const { applyMemoryOps } = await import("../src/memoryOps.js");
+        const ops = reviewConversation(view, []);
+        if (ops.length) {
+            await applyMemoryOps(ops, runtime.currentStateDir);
+            // S21.2: ops landed — the compaction path reads this counter and fires
+            // `consolidateMemories` only when > 0.
+            runtime.memoriesTouchedThisCompaction += ops.length;
+            runtime.pushTicker(`${C.green}🧠${C.reset} reviewed ${ops.length} memory op${ops.length === 1 ? "" : "s"} (${label})`);
+        }
+        return ops.length;
+    }
+    catch {
+        /* non-fatal — auto-review must never break the turn loop / compaction */
+        return 0;
+    }
+}
 /** Run the full compaction pipeline and persist a checkpoint. Returns the result. */
 export function runCompact(pi, runtime, config, ctx, messages, opts = {}) {
     runtime.bindRepo(ctx.cwd);
@@ -135,25 +164,11 @@ function doCompact(view, keepFrom, opts, sid, config, pi, ctx, runtime) {
     }
     // S24 review-on-compact: when pressure is high, the just-compacted region is
     // exactly the context worth remembering, so review it immediately rather than
-    // waiting for the next turn-cadence tick. Fire-and-forget (doCompact is sync):
-    // best-effort + non-fatal, paralleling the consolidate pass above. Only fires
-    // above the `high` band so low-pressure compactions don't pay the review cost.
+    // waiting for the next turn-cadence tick. Uses the shared runMemoryReview
+    // helper (fire-and-forget; doCompact is sync). Best-effort + non-fatal. Only
+    // fires above the `high` band so low-pressure compactions don't pay the cost.
     if (!result.deduped && config.memoryAutoReview && runtime.pressureBand !== "low" && runtime.pressureBand !== "medium") {
-        void (async () => {
-            try {
-                const { reviewConversation } = await import("../src/memory.js");
-                const { applyMemoryOps } = await import("../src/memoryOps.js");
-                const ops = reviewConversation(view, []);
-                if (ops.length) {
-                    await applyMemoryOps(ops, runtime.currentStateDir);
-                    runtime.memoriesTouchedThisCompaction += ops.length;
-                    runtime.pushTicker(`${C.green}🧠${C.reset} reviewed ${ops.length} memory op${ops.length === 1 ? "" : "s"} (pressure)`);
-                }
-            }
-            catch {
-                /* non-fatal — review-on-compact must never break the compaction */
-            }
-        })();
+        void runMemoryReview(runtime, view, "pressure");
     }
     // Sentinel marker: a non-LLM bookkeeping entry so subsequent triggers can
     // skip re-vectorizing an already-compacted region (zero token cost).

package/dist/extensions/mega-runtime.js

@@ -120,6 +120,26 @@ export class MegaRuntime {
     lastCtxTokens = null;
     lastCtxPercent = null;
     lastCtxWindow = 0;
+    /**
+     * DIAG counters for the "team run doesn't relieve context" investigation.
+     * Plain integers, incremented at the three compaction decision points. They
+     * let a headless test drive the real event handlers and assert the firing
+     * cadence without scraping log files. Inert in production (the live-trim and
+     * before-compact probes also emit logger.info, but these counters are always
+     * updated and cost nothing).
+     */
+    diagLiveTrimFires = 0; // context handler returned a trimmed view
+    diagBeforeCompactFires = 0; // session_before_compact handler entered
+    diagBeforeCompactSupplied = 0; // session_before_compact supplied our trim
+    diagAgentEndIdle = 0; // agent_end with activeAgents===0
+    diagAgentEndDurable = 0; // agent_end fired ctx.compact() (mid-run durable trim)
+    // Per-skip-path counters for the team-run diagnosis.
+    diagCtxFastGate = 0; // returned at token fast-gate (below threshold)
+    diagCtxNoCompact = 0; // autoCompactCheck().shouldCompact === false
+    diagCtxDebounce = 0; // debounceUntil not yet elapsed
+    diagCtxRunSkipped = 0; // runCompact() returned skipped
+    diagCtxCutNull = 0; // computeLiveTrimCut returned null (anchor/boundary)
+    diagCtxThrown = 0; // live-trim try threw (caught)
     /**
      * Live 0–1 pressure: how full the context window is relative to the compaction
      * threshold. Computed from the most recent context event the runtime already

package/dist/extensions/mega-teamrun.test.js

@@ -0,0 +1,143 @@
+/**
+ * mega-teamrun.test.ts — regression test for the "auto-compact runs but context
+ * never relieves during a team run (sub-agents)" bug.
+ *
+ * Loads the REAL compiled extension (extensions/mega-compact.js) through a
+ * faithful mock pi (mirrors mega-compact.test.ts's harness) and drives the
+ * exact event sequence a long team run produces:
+ *
+ *   agent_start -> context (over threshold) xN -> agent_end   (repeat x3)
+ *
+ * Asserts the TWO fixes:
+ *   1. live trim FIRES per-call (computeLiveTrimCut no longer returns null on
+ *      the anchor floor — was `cutNull`, liveTrimFires===0 before the fix).
+ *   2. the DURABLE trim fires at agent_end while idle + over threshold
+ *      (mid-run durable trigger), not only at parent settle.
+ *
+ * The mock ctx.compact() drives session_before_compact so we observe the
+ * durable truncation. Counters come from MegaRuntime.diag* (set behind the
+ * real handler code, inert in production).
+ *
+ * MEGACOMPACT_PGLITE_DISABLED keeps the run fast (no WASM index init).
+ */
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { createRequire } from "node:module";
+import { closeVectorIndex } from "../src/store/vectorIndex.js";
+const require = createRequire(import.meta.url);
+const baseTmp = mkdtempSync(join(tmpdir(), "mc-team-"));
+process.env.MEGACOMPACT_INDEX_DIR = join(baseTmp, "index");
+process.env.MEGACOMPACT_PGLITE_DISABLED = "true"; // fast: skip WASM index
+let counter = 0;
+function harness() {
+    const stateDir = join(baseTmp, `run-${counter++}`);
+    process.env.MEGACOMPACT_STATE_DIR = stateDir;
+    process.env.MEGACOMPACT_DEBUG = "true";
+    process.env.MEGACOMPACT_THRESHOLD_TOKENS = "50";
+    process.env.MEGACOMPACT_FAST_GATE_PCT = "1";
+    process.env.MEGACOMPACT_ANCHOR_USER_MESSAGES = "1";
+    process.env.MEGACOMPACT_DURABLE_TRIM_FLOOR = "0"; // piCompactWouldNoop must not skip
+    process.env.MEGACOMPACT_MEMORY_AUTO_REVIEW = "false";
+    process.env.MEGACOMPACT_RAPTOR_ENABLED = "false";
+    delete process.env.MEGACOMPACT_LEGACY_DURABLE_TRIM;
+    const handlers = {};
+    const compactCalls = [];
+    function msg(role, text, toolName) {
+        if (role === "assistant" && toolName) {
+            return { role: "assistant", content: [{ type: "toolCall", name: toolName, id: "c1", arguments: {} }], api: "anthropic-messages", provider: "anthropic", model: "m", usage: { inputTokens: 1, outputTokens: 1, cacheReadTokens: 0, cacheWriteTokens: 0 }, stopReason: "tool_use", timestamp: 0 };
+        }
+        if (role === "toolResult" && toolName) {
+            return { role: "toolResult", content: [{ type: "text", text }], toolCallId: "c1", toolName, isError: false, timestamp: 0 };
+        }
+        return { role: "user", content: text, timestamp: 0 };
+    }
+    const session = [];
+    for (let i = 0; i < 14; i++) {
+        session.push(msg("user", `actually we decided to use approach ${i} for module ${i}`));
+        session.push(msg("assistant", `edited module ${i}`, "Edit"));
+        session.push(msg("toolResult", `edited module ${i}`, "Edit"));
+    }
+    const toEntry = (m, i) => ({ type: "message", id: `e${i}`, parentId: null, timestamp: String(i), message: m });
+    const sessionManager = {
+        getSessionId: () => "sess_team_001",
+        getEntries: () => session.map(toEntry),
+        getBranch: () => session.map(toEntry),
+    };
+    function makeCtx(over = {}) {
+        return {
+            ui: { setStatus: () => { }, notify: () => { }, select: () => { }, confirm: async () => true, input: async () => "", setWidget: () => { } },
+            mode: "tui", hasUI: true, cwd: stateDir, sessionManager,
+            modelRegistry: {}, model: undefined, isIdle: () => true, isProjectTrusted: () => true,
+            signal: undefined, abort: () => { }, hasPendingMessages: () => false, shutdown: () => { },
+            getContextUsage: () => ({ tokens: 200000, contextWindow: 200000, percent: 100 }),
+            // Mock ctx.compact() runs pi's flow and fires session_before_compact.
+            compact: (opts) => {
+                compactCalls.push(opts);
+                if (handlers["session_before_compact"]) {
+                    return handlers["session_before_compact"]({ type: "session_before_compact", reason: "threshold", willRetry: false, signal: undefined, preparation: { firstKeptEntryId: "e2", messagesToSummarize: session.slice(0, 2), tokensBefore: 500 } }, makeCtx());
+                }
+                return undefined;
+            },
+            getSystemPrompt: () => "system base",
+            ...over,
+        };
+    }
+    const pi = {
+        on: (ev, h) => { handlers[ev] = h; },
+        registerCommand: () => { }, registerTool: () => { }, registerShortcut: () => { },
+        registerFlag: () => { }, getFlag: () => undefined, registerMessageRenderer: () => { },
+        registerEntryRenderer: () => { }, sendMessage: () => { }, sendUserMessage: () => { },
+        appendEntry: () => { }, setSessionName: () => { }, getSessionName: () => undefined,
+        setLabel: () => { }, exec: async () => ({ stdout: "", stderr: "", code: 0 }),
+        getActiveTools: () => [], getAllTools: () => [], setActiveTools: () => { },
+        getCommands: () => [], setModel: async () => false, getThinkingLevel: () => "off",
+        setThinkingLevel: () => { },
+    };
+    const mod = require("./mega-compact.js");
+    mod.default(pi);
+    const { lastRuntime } = require("./mega-events.js");
+    const fire = (ev, event, ctx) => handlers[ev](event, ctx);
+    return {
+        stateDir, handlers, compactCalls, fire, ctx: makeCtx, session,
+        runtime: lastRuntime, // MegaRuntime with diag* counters
+        // Advance the debounce so agent_end (same instant) can trigger durable trim.
+        clearDebounce: () => { if (lastRuntime)
+            lastRuntime.debounceUntil = 0; },
+    };
+}
+test("team run: live trim fires AND durable trim fires per sub-agent (relieves context)", async () => {
+    const h = harness();
+    const ctx = h.ctx();
+    for (let a = 0; a < 3; a++) {
+        await h.fire("agent_start", { type: "agent_start", messages: [] }, ctx);
+        for (let i = 0; i < 4; i++) {
+            await h.fire("context", { type: "context", messages: h.session }, ctx);
+        }
+        // Real team runs settle seconds after the last context event; mimic that
+        // so the 2s debounce has elapsed and the durable trigger can fire.
+        await new Promise((r) => setTimeout(r, 2100));
+        h.clearDebounce();
+        await h.fire("agent_end", { type: "agent_end", messages: [] }, ctx);
+    }
+    const rt = h.runtime;
+    // FIX 1: live trim must actually fire (was 0 — computeLiveTrimCut returned null).
+    assert.ok(rt.diagLiveTrimFires > 0, "live trim fires during the team run (anchor-floor fix)");
+    assert.equal(rt.diagCtxCutNull, 0, "no live-trim cut skipped on anchor floor");
+    // FIX 2: durable trim must fire at each agent_end (was 0 — only at parent settle).
+    assert.equal(rt.diagAgentEndDurable, 3, "mid-run durable trigger fired at each agent_end");
+    assert.equal(rt.diagBeforeCompactSupplied, 3, "our durable trim supplied 3x (context relieved)");
+    assert.ok(h.compactCalls.length >= 3, "ctx.compact() invoked for durable trim between sub-agents");
+});
+test("control: session_before_compact supplies a durable compaction (parent settles)", async () => {
+    const h = harness();
+    const res = await h.fire("session_before_compact", { type: "session_before_compact", reason: "threshold", willRetry: false, signal: undefined, preparation: { firstKeptEntryId: "e2", messagesToSummarize: h.session.slice(0, 4), tokensBefore: 500 } }, h.ctx());
+    assert.ok(res?.compaction, "compaction result returned to pi");
+    assert.equal(res.compaction.firstKeptEntryId, "e2", "reuses pi's boundary (PREVENT-PI-002)");
+});
+test("cleanup", async () => {
+    await closeVectorIndex();
+    rmSync(baseTmp, { recursive: true, force: true });
+});

package/dist/extensions/mega-trim.js

@@ -22,8 +22,39 @@ export function computeLiveTrimCut(view, opts) {
         return null; // nothing safe to cut — keep everything this call
     const recent = view.slice(cut);
     const userCount = recent.filter((m) => m.role === "user").length;
-    if (userCount < opts.anchorUserMessages)
-        return null;
+    // ANCHOR FLOOR (PREVENT-PI-001): the recent window must keep at least
+    // `anchorUserMessages` user messages. The original compactedFrom can land on a
+    // run that starts with fewer than that (e.g. the preserved region begins on a
+    // tool pair, or the session's tail is tool-heavy). Instead of bailing out and
+    // skipping the live trim entirely this call (which left the model fed a
+    // 150k-context window during long team runs), walk `cut` backward until the
+    // preserved run contains enough user messages — bounded by the boundary-safe
+    // constraint so we never split a tool pair. Falls back to null only when the
+    // whole view can't satisfy the floor (tiny sessions) — the next context event
+    // retries.
+    if (userCount < opts.anchorUserMessages) {
+        let c = cut;
+        while (c > 1) {
+            c--;
+            if (!isBoundarySafe(view, c))
+                continue;
+            const recentNow = view.slice(c);
+            const usersNow = recentNow.filter((m) => m.role === "user").length;
+            if (usersNow >= opts.anchorUserMessages) {
+                cut = c;
+                break;
+            }
+        }
+        if (cut > 1) {
+            const finalRecent = view.slice(cut);
+            if (finalRecent.filter((m) => m.role === "user").length < opts.anchorUserMessages) {
+                return null; // cannot satisfy the floor without dropping too much — retry next call
+            }
+        }
+        else {
+            return null;
+        }
+    }
     return cut;
 }
 /** The formatted compacted-region summary as a user-role engine message. */