omnius 1.0.494 → 1.0.496

package/dist/index.js

@@ -584033,7 +584033,7 @@ Your hypotheses MUST address this specific error, not generic causes.
        * system prompt. Mutates `messages` in place.
        */
       gcSystemMessages(messages2) {
-        const ctxTokens = this.options.contextWindowSize ?? 0;
+        const ctxTokens = this.effectiveContextWindow();
         const systemMaxChars = ctxTokens > 0 ? Math.max(24e3, Math.floor(ctxTokens * 4 * 0.45)) : 0;
         const beforeChars = messages2.reduce((sum2, m2) => sum2 + (m2.role === "system" && typeof m2.content === "string" ? m2.content.length : 0), 0);
         const result = this._signalExtractor.distillSystemMessages(messages2, {
@@ -586146,6 +586146,39 @@ ${body}`;
        * - Large models:  compact at ~75% (plenty of room, let attention work)
        * - Deep context:  compact at ~85% (trust the model's full attention window)
        */
+      /**
+       * Absolute working-context ceiling in TOKENS. Local models advertise a huge
+       * n_ctx_train (qwen3 = 262144) but lose coherence far below it. Keying
+       * compaction/pruning budgets off the raw window means Omnius never trims until
+       * the model is already saturated and doom-looping (observed: a 35B run pinned
+       * at 120K tokens with a 196K threshold that never fired). This caps the budget
+       * denominator to a realistic, tier-aware working window so compaction actually
+       * runs and the active context stays focused.
+       *
+       * Tier defaults (tokens): small 28K, medium 48K, large 72K. deepContext relaxes
+       * ×1.4. Override with OMNIUS_MAX_CONTEXT_TOKENS. Cloud/large-context users who
+       * want to use their full window set that env high (or per-tier via the option).
+       */
+      absoluteContextCeiling() {
+        const env2 = Number(process.env["OMNIUS_MAX_CONTEXT_TOKENS"]);
+        if (Number.isFinite(env2) && env2 > 0)
+          return Math.floor(env2);
+        const tier = this.options.modelTier ?? "large";
+        const deep = this.options.deepContext ?? false;
+        const base3 = tier === "small" ? 28e3 : tier === "medium" ? 48e3 : 72e3;
+        return deep ? Math.floor(base3 * 1.4) : base3;
+      }
+      /**
+       * Context window (tokens) used for compaction/pruning BUDGET math — the
+       * smaller of the model's declared window and the effective working ceiling.
+       * The raw declared window is still used for the backend num_ctx and the
+       * generation-token budget; this only governs WHEN we compact/prune.
+       */
+      effectiveContextWindow() {
+        const raw = this.options.contextWindowSize ?? 0;
+        const ceiling = this.absoluteContextCeiling();
+        return raw > 0 ? Math.min(raw, ceiling) : ceiling;
+      }
       contextLimits() {
         const ctx3 = this.options.contextWindowSize;
         const tier = this.options.modelTier ?? "large";
@@ -586158,6 +586191,7 @@ ${body}`;
         } else {
           compactionThreshold = deep ? 8e4 : this.options.compactionThreshold;
         }
+        compactionThreshold = Math.min(compactionThreshold, this.absoluteContextCeiling());
         const keepRecentMax = deep ? 24 : 12;
         let keepRecent;
         if (ctx3 > 0) {
@@ -587989,6 +588023,86 @@ ${this._lastPprMemoryLines.slice(0, 5).join("\n")}` : null;
         });
         return frame.content;
       }
+      /**
+       * Feed the ContextTree one tool call, detect phase transitions, and — on a
+       * transition — capture the outgoing phase's message slice, contract inactive
+       * phases (LLM-summarized via makePhaseSummarizer, anchors retained), and
+       * archive the full message bodies to `.omnius/phases/<phase>-<id>-turnN.jsonl`,
+       * leaving only a pointer + high-signal summary in the working context. This is
+       * the core "offload large block → disk pointer + searchable signature"
+       * mechanism; `contextTree.expand(phase)` re-hydrates on demand.
+       *
+       * Wired into BOTH the primary loop and the brute-force loop. The brute-force
+       * path is the ACTIVE main-agent path on live 35B runs; omitting the feed here
+       * is why the context tree "never kicked off" — it rendered an empty phase
+       * status every turn while the raw history grew unbounded. Returns the
+       * transition (so the primary loop can reset per-phase tool budgets) or null.
+       */
+      _advanceContextTreeOnToolCall(toolName, argsKey, turn, messages2) {
+        if (!this._contextTree)
+          return null;
+        this._contextTree.observeToolCall(toolName, argsKey, turn);
+        const transition = this._contextTree.maybeTransition(turn);
+        if (!transition)
+          return null;
+        const startIdx = Math.min(this._phaseMessageStartIdx, messages2.length);
+        const phaseSlice = messages2.slice(startIdx);
+        if (phaseSlice.length > 0) {
+          const fromNode = this._contextTree.getSnapshot().phases[transition.from];
+          if (fromNode)
+            fromNode.messages = phaseSlice;
+        }
+        this.emit({
+          type: "status",
+          content: `Phase: ${transition.from} → ${transition.to}`,
+          timestamp: (/* @__PURE__ */ new Date()).toISOString()
+        });
+        const summarizer = this.makePhaseSummarizer();
+        const contracted = this._contextTree.contractInactive(turn, summarizer ? (msgs) => summarizer(transition.from, msgs) : void 0);
+        if (contracted.length > 0) {
+          this.emit({
+            type: "status",
+            content: `Phase contraction: ${contracted.join(", ")} → contracted (${summarizer ? "LLM-summarized" : "stub-summarized"}, anchors retained)`,
+            timestamp: (/* @__PURE__ */ new Date()).toISOString()
+          });
+          try {
+            const archDir = _pathJoin(this.omniusStateDir(), "phases");
+            _fsMkdirSync(archDir, { recursive: true });
+            for (const phaseName of contracted) {
+              const node = this._contextTree.getSnapshot().phases[phaseName];
+              if (!node)
+                continue;
+              const stamp = `${phaseName}-${Date.now().toString(36)}-turn${turn}`;
+              const archPath = _pathJoin(archDir, `${stamp}.jsonl`);
+              const archived = {
+                phase: phaseName,
+                contractedAtTurn: turn,
+                contractedAtIso: (/* @__PURE__ */ new Date()).toISOString(),
+                anchors: node.anchors,
+                summary: node.summary ?? null,
+                messages: node.messages ?? []
+              };
+              _fsWriteFileSync(archPath, JSON.stringify(archived) + "\n", "utf-8");
+              this._contextTree.archive(phaseName, archPath);
+            }
+            this.emit({
+              type: "status",
+              content: `Phase archive: ${contracted.length} phase(s) written to .omnius/phases/`,
+              timestamp: (/* @__PURE__ */ new Date()).toISOString()
+            });
+          } catch (archErr) {
+            this.emit({
+              type: "status",
+              content: `Phase archive failed (non-fatal): ${archErr instanceof Error ? archErr.message : String(archErr)}`,
+              timestamp: (/* @__PURE__ */ new Date()).toISOString()
+            });
+          }
+        }
+        this._phaseMessageStartIdx = messages2.length;
+        this._taskState.phase = transition.to;
+        this._taskState.phaseSince = turn;
+        return transition;
+      }
       makePhaseSummarizer() {
         if (process.env["OMNIUS_DISABLE_PHASE_SUMMARIZER"] === "1")
           return null;
@@ -592316,77 +592430,18 @@ Corrective action: try a different approach first: read relevant files, adjust a
                 return { tc, output: blockMsg, success: false };
               }
               this._toolLastUsedTurn.set(tc.name, turn);
-              if (this._contextTree) {
-                this._contextTree.observeToolCall(tc.name, argsKey, turn);
-                const transition = this._contextTree.maybeTransition(turn);
-                if (transition) {
-                  const phaseSlice = messages2.slice(this._phaseMessageStartIdx);
-                  if (phaseSlice.length > 0) {
-                    const fromNode = this._contextTree.getSnapshot().phases[transition.from];
-                    if (fromNode) {
-                      fromNode.messages = phaseSlice;
-                    }
-                  }
-                  this.emit({
-                    type: "status",
-                    content: `Phase: ${transition.from} → ${transition.to}`,
-                    timestamp: (/* @__PURE__ */ new Date()).toISOString()
-                  });
-                  const summarizer = this.makePhaseSummarizer();
-                  const contracted = this._contextTree.contractInactive(turn, summarizer ? (msgs) => summarizer(transition.from, msgs) : void 0);
-                  if (contracted.length > 0) {
-                    this.emit({
-                      type: "status",
-                      content: `Phase contraction: ${contracted.join(", ")} → contracted (${summarizer ? "LLM-summarized" : "stub-summarized"}, anchors retained)`,
-                      timestamp: (/* @__PURE__ */ new Date()).toISOString()
-                    });
-                    try {
-                      const archDir = _pathJoin(this.omniusStateDir(), "phases");
-                      _fsMkdirSync(archDir, { recursive: true });
-                      for (const phaseName of contracted) {
-                        const node = this._contextTree.getSnapshot().phases[phaseName];
-                        if (!node)
-                          continue;
-                        const stamp = `${phaseName}-${Date.now().toString(36)}-turn${turn}`;
-                        const archPath = _pathJoin(archDir, `${stamp}.jsonl`);
-                        const archived = {
-                          phase: phaseName,
-                          contractedAtTurn: turn,
-                          contractedAtIso: (/* @__PURE__ */ new Date()).toISOString(),
-                          anchors: node.anchors,
-                          summary: node.summary ?? null,
-                          messages: node.messages ?? []
-                        };
-                        _fsWriteFileSync(archPath, JSON.stringify(archived) + "\n", "utf-8");
-                        this._contextTree.archive(phaseName, archPath);
-                      }
-                      this.emit({
-                        type: "status",
-                        content: `Phase archive: ${contracted.length} phase(s) written to .omnius/phases/`,
-                        timestamp: (/* @__PURE__ */ new Date()).toISOString()
-                      });
-                    } catch (archErr) {
-                      this.emit({
-                        type: "status",
-                        content: `Phase archive failed (non-fatal): ${archErr instanceof Error ? archErr.message : String(archErr)}`,
-                        timestamp: (/* @__PURE__ */ new Date()).toISOString()
-                      });
-                    }
-                  }
-                  this._phaseMessageStartIdx = messages2.length;
-                  this._taskState.phase = transition.to;
-                  this._taskState.phaseSince = turn;
-                  for (const [tool2, budget] of Object.entries(toolBudgets)) {
-                    toolCallBudget.set(tool2, budget);
-                  }
-                  this._writesSinceLastTodoWrite = 0;
-                  this._progressGateActive = false;
-                  this.emit({
-                    type: "status",
-                    content: `Tool budgets reset for new phase (${Object.keys(toolBudgets).length} tools)`,
-                    timestamp: (/* @__PURE__ */ new Date()).toISOString()
-                  });
+              const _treeTransition = this._advanceContextTreeOnToolCall(tc.name, argsKey, turn, messages2);
+              if (_treeTransition) {
+                for (const [tool2, budget] of Object.entries(toolBudgets)) {
+                  toolCallBudget.set(tool2, budget);
                 }
+                this._writesSinceLastTodoWrite = 0;
+                this._progressGateActive = false;
+                this.emit({
+                  type: "status",
+                  content: `Tool budgets reset for new phase (${Object.keys(toolBudgets).length} tools)`,
+                  timestamp: (/* @__PURE__ */ new Date()).toISOString()
+                });
               }
               const REG61_EDIT_TOOLS = /* @__PURE__ */ new Set([
                 "file_write",
@@ -595757,6 +595812,7 @@ ${this.options.maxTurns && this.options.maxTurns > 0 ? `You have ${this.options.
             }
             this.proactivePrune(compactedMsgs, this._taskState.toolCallCount);
             this.microcompact(compactedMsgs);
+            this._retireStaleEvidenceInPlace(compactedMsgs, this._taskState.toolCallCount);
             let bfEnvironmentBlock = null;
             if (this.options.environmentProvider) {
               try {
@@ -595922,6 +595978,8 @@ ${this.options.maxTurns && this.options.maxTurns > 0 ? `You have ${this.options.
                 if (!r2)
                   continue;
                 messages2.push(this.buildModelFacingToolMessage(r2.output, r2.tc.id, r2.tc.name, r2.tc.arguments, r2.success));
+                this._toolLastUsedTurn.set(r2.tc.name, turn);
+                this._advanceContextTreeOnToolCall(r2.tc.name, this._buildExactArgsKey(r2.tc.arguments ?? {}), turn, messages2);
                 if (r2.systemGuidance) {
                   messages2.push({
                     role: "system",
@@ -598896,7 +598954,7 @@ Describe what you see and integrate this into your current approach.` : "[User s
         const head = messages2.slice(0, headEndIdx);
         if (messages2.length <= headEndIdx + keepRecent)
           return messages2;
-        const tailTokenBudget = Math.max(4e3, Math.floor((this.options.contextWindowSize || 32e3) * 0.15));
+        const tailTokenBudget = Math.max(4e3, Math.floor((this.effectiveContextWindow() || 32e3) * 0.15));
         let recentStart = messages2.length - keepRecent;
         {
           let accumulated = 0;

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.494",
+  "version": "1.0.496",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.494",
+      "version": "1.0.496",
       "bundleDependencies": [
         "image-to-ascii"
       ],
@@ -1077,14 +1077,14 @@
       }
     },
     "node_modules/@libp2p/peer-record": {
-      "version": "9.0.12",
-      "resolved": "https://registry.npmjs.org/@libp2p/peer-record/-/peer-record-9.0.12.tgz",
-      "integrity": "sha512-tYGHo9gSO1ZqQripsEtpzxMrWDnmCMgFe8c1TKcml8dZUBCAsM9k5js25RKb0HA1aqZ2VX54XPFazpbjVs3tow==",
+      "version": "9.0.13",
+      "resolved": "https://registry.npmjs.org/@libp2p/peer-record/-/peer-record-9.0.13.tgz",
+      "integrity": "sha512-lkdZUNC6h86YECpRE3dB4GLYSiLLuXFBleiM01oEmWtAdTCfRd+vRqNkguSoC6R3tGk/YWRm7cfz39ATPuq7zQ==",
       "license": "Apache-2.0 OR MIT",
       "dependencies": {
-        "@libp2p/crypto": "^5.1.20",
-        "@libp2p/interface": "^3.2.4",
-        "@libp2p/peer-id": "^6.0.11",
+        "@libp2p/crypto": "^5.1.21",
+        "@libp2p/interface": "^3.2.5",
+        "@libp2p/peer-id": "^6.0.12",
         "@multiformats/multiaddr": "^13.0.3",
         "multiformats": "^14.0.0",
         "protons-runtime": "^7.0.0",
@@ -1193,9 +1193,9 @@
       }
     },
     "node_modules/@libp2p/record": {
-      "version": "4.0.14",
-      "resolved": "https://registry.npmjs.org/@libp2p/record/-/record-4.0.14.tgz",
-      "integrity": "sha512-YObHdE1dkAiVDPR+YHIFcCD5YMppP0lj2shnMj0L5PgLka8lnZgaY5AjiQo+DMvLFhRvNIMHyxlE9WQUbrM1wA==",
+      "version": "4.0.15",
+      "resolved": "https://registry.npmjs.org/@libp2p/record/-/record-4.0.15.tgz",
+      "integrity": "sha512-EaHFtQAlZuM9vCpd+0KnZhTVgV26Uzje7DkyUseGNKTy+Wit9Q94C/+wDThpZuRHnK1FH/CvPxJ4kLj7AoQ1WA==",
       "license": "Apache-2.0 OR MIT",
       "dependencies": {
         "protons-runtime": "^7.0.0",
@@ -1358,19 +1358,19 @@
       }
     },
     "node_modules/@libp2p/webrtc": {
-      "version": "6.0.25",
-      "resolved": "https://registry.npmjs.org/@libp2p/webrtc/-/webrtc-6.0.25.tgz",
-      "integrity": "sha512-c7QwXOjqBR+xgCIE8bv/A1NGQQMX3Sfrek+CD5n+XJur2TOoEmNjYyRX98Yt2h5YzbD083/bZK7ITZiu9xSdmQ==",
+      "version": "6.0.26",
+      "resolved": "https://registry.npmjs.org/@libp2p/webrtc/-/webrtc-6.0.26.tgz",
+      "integrity": "sha512-6IXDWXyE+X/J7IQFBq8AkxWwTVIVjC/2tD17kny4u8p2VOkXmC1aOhQYtKQRv4TSolraiF+AVVGTyf9Rn+w4VQ==",
       "license": "Apache-2.0 OR MIT",
       "dependencies": {
         "@chainsafe/is-ip": "^2.1.0",
         "@chainsafe/libp2p-noise": "^17.0.0",
-        "@libp2p/crypto": "^5.1.20",
-        "@libp2p/interface": "^3.2.4",
-        "@libp2p/interface-internal": "^3.1.7",
-        "@libp2p/keychain": "^6.1.3",
-        "@libp2p/peer-id": "^6.0.11",
-        "@libp2p/utils": "^7.2.3",
+        "@libp2p/crypto": "^5.1.21",
+        "@libp2p/interface": "^3.2.5",
+        "@libp2p/interface-internal": "^3.1.8",
+        "@libp2p/keychain": "^6.1.4",
+        "@libp2p/peer-id": "^6.0.12",
+        "@libp2p/utils": "^7.2.4",
         "@multiformats/multiaddr": "^13.0.3",
         "@multiformats/multiaddr-matcher": "^3.0.2",
         "@peculiar/webcrypto": "^1.5.0",
@@ -2554,9 +2554,9 @@
       }
     },
     "node_modules/bare-fs": {
-      "version": "4.7.3",
-      "resolved": "https://registry.npmjs.org/bare-fs/-/bare-fs-4.7.3.tgz",
-      "integrity": "sha512-xRgplks8SvcKkdlv2M6Z2LZmRsmqd+x0nXXGXeMEjwdibj1HSDrlnqBRLeYdMvsgCox7Bq0e+DHwfczOfsn6IA==",
+      "version": "4.7.4",
+      "resolved": "https://registry.npmjs.org/bare-fs/-/bare-fs-4.7.4.tgz",
+      "integrity": "sha512-y1kC+ffIx/tPLdTE693uNjHfzTfr+ravR5tvWlMXe25nELbkqV400S71qHDwbkAQ1FVEZobB1NFRzFbCCcyBCQ==",
       "license": "Apache-2.0",
       "optional": true,
       "dependencies": {
@@ -2578,25 +2578,12 @@
         }
       }
     },
-    "node_modules/bare-os": {
-      "version": "3.9.3",
-      "resolved": "https://registry.npmjs.org/bare-os/-/bare-os-3.9.3.tgz",
-      "integrity": "sha512-fF4Q7QsyKVF5Rj0qvI8BgUNjqzC2JvQlpTaPLjVJVxYVUX5Zr9un+y3w1HmA4nNKdFmRBT8z/WmrjvXzXVerKQ==",
-      "license": "Apache-2.0",
-      "optional": true,
-      "engines": {
-        "bare": ">=1.14.0"
-      }
-    },
     "node_modules/bare-path": {
-      "version": "3.0.1",
-      "resolved": "https://registry.npmjs.org/bare-path/-/bare-path-3.0.1.tgz",
-      "integrity": "sha512-ghj2DSK/2e99a1anTVPCV4m4YIYtrbXhfM7V3D7XZLOTsybnYyaJloymGqssQc8l/or0UoDyRtNQkmkEF/ysgQ==",
+      "version": "3.1.0",
+      "resolved": "https://registry.npmjs.org/bare-path/-/bare-path-3.1.0.tgz",
+      "integrity": "sha512-322oSBHOhMhOm2CVwn7b3+gWQqwzSgIY4gNpKKb+Ha5jx2dEl9ClOgz/SK57DwuDql8YQg8JRpb9U0o8K3pW9g==",
       "license": "Apache-2.0",
-      "optional": true,
-      "dependencies": {
-        "bare-os": "^3.0.1"
-      }
+      "optional": true
     },
     "node_modules/bare-stream": {
       "version": "2.13.3",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.494",
+  "version": "1.0.496",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",