@nexrall/code-core 1.4.0 → 1.4.1

package/dist/agent/loop.d.ts

@@ -53,7 +53,7 @@ export interface ProgressLedger {
 }
 export declare function createLedger(): ProgressLedger;
 /** Record one tool call's effect on the ledger (deterministic, no model call). */
-export declare function ledgerRecord(ledger: ProgressLedger, toolName: string, input: Record<string, unknown> | undefined, ok: boolean): void;
+export declare function ledgerRecord(ledger: ProgressLedger, toolName: string, input: Record<string, unknown> | undefined, ok: boolean, output?: string): void;
 /** Render the ledger as a compact, verbatim block for the compaction preamble. */
 export declare function ledgerSummary(ledger: ProgressLedger): string;
 /**

package/dist/agent/loop.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"loop.d.ts","sourceRoot":"","sources":["../../src/agent/loop.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,OAAO,EAMP,gBAAgB,EACjB,MAAM,UAAU,CAAC;AAqKlB,wBAAgB,oBAAoB,CAClC,WAAW,EAAE,MAAM,GAAG,SAAS,EAC/B,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GACnC,MAAM,CAWR;AA8RD,+EAA+E;AAC/E,wBAAgB,iBAAiB,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,MAAM,CAM7D;AAsBD,iFAAiF;AACjF,eAAO,MAAM,gBAAgB,aAA+G,CAAC;AAC7I,gGAAgG;AAChG,eAAO,MAAM,aAAa,QAA2J,CAAC;AAEtL;;;;;;;;;;;;GAYG;AACH,wBAAgB,gBAAgB,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,CAK5E;AAUD;;;;;;;GAOG;AACH,wBAAgB,YAAY,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,MAAM,CAwBxD;AAoBD,MAAM,WAAW,cAAc;IAC7B,YAAY,EAAE,GAAG,CAAC,MAAM,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC3D,aAAa,EAAE,KAAK,CAAC;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,EAAE,EAAE,OAAO,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAClE,qFAAqF;IACrF,aAAa,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACvD;;;;OAIG;IACH,KAAK,EAAE,MAAM,CAAC;CACf;AAED,wBAAgB,YAAY,IAAI,cAAc,CAE7C;AAED,kFAAkF;AAClF,wBAAgB,YAAY,CAC1B,MAAM,EAAE,cAAc,EACtB,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS,EAC1C,EAAE,EAAE,OAAO,GACV,IAAI,CAkCN;AAED,kFAAkF;AAClF,wBAAgB,aAAa,CAAC,MAAM,EAAE,cAAc,GAAG,MAAM,CA6B5D;AAmBD;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,MAAM,CAmB/D;AA2FD,wBAAsB,YAAY,CAChC,eAAe,EAAE,OAAO,EAAE,EAC1B,OAAO,EAAE,gBAAgB,GACxB,OAAO,CAAC,OAAO,EAAE,CAAC,CA6epB"}
+{"version":3,"file":"loop.d.ts","sourceRoot":"","sources":["../../src/agent/loop.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,OAAO,EAMP,gBAAgB,EACjB,MAAM,UAAU,CAAC;AAqKlB,wBAAgB,oBAAoB,CAClC,WAAW,EAAE,MAAM,GAAG,SAAS,EAC/B,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GACnC,MAAM,CAWR;AA8RD,+EAA+E;AAC/E,wBAAgB,iBAAiB,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,MAAM,CAM7D;AAsBD,iFAAiF;AACjF,eAAO,MAAM,gBAAgB,aAA+G,CAAC;AAC7I,gGAAgG;AAChG,eAAO,MAAM,aAAa,QAA2J,CAAC;AAEtL;;;;;;;;;;;;GAYG;AACH,wBAAgB,gBAAgB,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,CAK5E;AAUD;;;;;;;GAOG;AACH,wBAAgB,YAAY,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,MAAM,CAwBxD;AAoBD,MAAM,WAAW,cAAc;IAC7B,YAAY,EAAE,GAAG,CAAC,MAAM,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC3D,aAAa,EAAE,KAAK,CAAC;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,EAAE,EAAE,OAAO,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAClE,qFAAqF;IACrF,aAAa,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACvD;;;;OAIG;IACH,KAAK,EAAE,MAAM,CAAC;CACf;AAED,wBAAgB,YAAY,IAAI,cAAc,CAE7C;AAED,kFAAkF;AAClF,wBAAgB,YAAY,CAC1B,MAAM,EAAE,cAAc,EACtB,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS,EAC1C,EAAE,EAAE,OAAO,EACX,MAAM,CAAC,EAAE,MAAM,GACd,IAAI,CA4CN;AAED,kFAAkF;AAClF,wBAAgB,aAAa,CAAC,MAAM,EAAE,cAAc,GAAG,MAAM,CA6B5D;AAmBD;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,MAAM,CAmB/D;AA2FD,wBAAsB,YAAY,CAChC,eAAe,EAAE,OAAO,EAAE,EAC1B,OAAO,EAAE,gBAAgB,GACxB,OAAO,CAAC,OAAO,EAAE,CAAC,CAufpB"}

package/dist/agent/loop.js

@@ -551,7 +551,7 @@ function createLedger() {
     return { filesTouched: new Map(), verifications: [], testIntegrity: [], epoch: 0 };
 }
 /** Record one tool call's effect on the ledger (deterministic, no model call). */
-function ledgerRecord(ledger, toolName, input, ok) {
+function ledgerRecord(ledger, toolName, input, ok, output) {
     if (exports.WRITE_TOOL_NAMES.has(toolName)) {
         if (!ok)
             return; // a FAILED write changed nothing — not a durable fact
@@ -565,10 +565,22 @@ function ledgerRecord(ledger, toolName, input, ok) {
         }
         // Reward-hacking guard: if this write WEAKENED a test file, record it so the
         // signal survives compaction and can be surfaced before the agent finishes.
-        const ti = (0, testIntegrity_1.analyzeWriteToolForTestIntegrity)(toolName, input);
-        if (ti?.suspicious && p) {
-            for (const f of ti.findings) {
-                ledger.testIntegrity.push({ path: p, reason: f.reason });
+        const reasons = [];
+        // write_file overwrites carry a marker computed by the executor (which had the
+        // prior on-disk content) — it detects REMOVED assertions/cases, not just
+        // additive skips/tautologies. Prefer it when present.
+        const markerReasons = toolName === 'write_file' ? (0, testIntegrity_1.decodeTestIntegrityMarker)(output) : [];
+        if (markerReasons.length) {
+            reasons.push(...markerReasons);
+        }
+        else {
+            const ti = (0, testIntegrity_1.analyzeWriteToolForTestIntegrity)(toolName, input);
+            if (ti?.suspicious)
+                reasons.push(...ti.findings.map((f) => f.reason));
+        }
+        if (reasons.length && p) {
+            for (const reason of reasons) {
+                ledger.testIntegrity.push({ path: p, reason });
             }
             if (ledger.testIntegrity.length > LEDGER_MAX_NOTES * 2) {
                 ledger.testIntegrity.splice(0, ledger.testIntegrity.length - LEDGER_MAX_NOTES);
@@ -1110,16 +1122,25 @@ async function runAgentLoop(initialMessages, options) {
                         }
                     }
                 }
+                // Capture the raw output (may carry the invisible test-integrity
+                // marker from a write_file overwrite) for the ledger, then strip the
+                // marker so neither the UI (onToolResult) nor the model ever see it.
+                let rawOutput;
+                if (name === 'write_file' && result.output) {
+                    rawOutput = result.output;
+                    result.output = (0, testIntegrity_1.stripTestIntegrityMarker)(result.output);
+                }
                 // Notify caller about result
                 options.onToolResult(name, result);
-                return { block: { ...block, id }, result };
+                return { block: { ...block, id }, result, rawOutput };
             }));
             // Track whether files were mutated / verified this run, for the one-shot
             // end-of-task nudge below (GAP D — see declaration above).
-            for (const { block, result } of toolResults) {
+            for (const { block, result, rawOutput } of toolResults) {
                 const ok = result.error === undefined;
                 // GAP E — feed every successful effect into the deterministic ledger.
-                ledgerRecord(ledger, block.name, block.input, ok);
+                // rawOutput carries the (already-stripped-from-view) test-integrity marker.
+                ledgerRecord(ledger, block.name, block.input, ok, rawOutput ?? result.output);
                 if (!ok)
                     continue; // failed calls don't count either way
                 if (exports.WRITE_TOOL_NAMES.has(block.name))

package/dist/agent/testIntegrity.d.ts

@@ -53,4 +53,10 @@ export declare function analyzeTestEdit(path: string, oldText: string, newText:
  * - write_file: ('', content) — additive-only detection (no prior content here)
  */
 export declare function analyzeWriteToolForTestIntegrity(toolName: string, input: Record<string, unknown> | undefined): TestIntegrityResult | null;
+/** Encode findings as an invisible marker to append to a tool's output string. */
+export declare function encodeTestIntegrityMarker(findings: TestEditFinding[]): string;
+/** Extract weakening reasons from a tool output that may carry the marker. */
+export declare function decodeTestIntegrityMarker(output: string | undefined): string[];
+/** Strip the (invisible) marker from output before it is shown to the model. */
+export declare function stripTestIntegrityMarker(output: string | undefined): string;
 //# sourceMappingURL=testIntegrity.d.ts.map

package/dist/agent/testIntegrity.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"testIntegrity.d.ts","sourceRoot":"","sources":["../../src/agent/testIntegrity.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AAEH,MAAM,WAAW,eAAe;IAC9B,gEAAgE;IAChE,MAAM,EAAE,MAAM,CAAC;IACf,0CAA0C;IAC1C,IAAI,EAAE,YAAY,GAAG,mBAAmB,GAAG,iBAAiB,GAAG,kBAAkB,GAAG,gBAAgB,CAAC;CACtG;AAED,MAAM,WAAW,mBAAmB;IAClC,UAAU,EAAE,OAAO,CAAC;IACpB,UAAU,EAAE,OAAO,CAAC;IACpB,QAAQ,EAAE,eAAe,EAAE,CAAC;CAC7B;AAQD,wBAAgB,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAGhD;AA4ED;;;GAGG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,mBAAmB,CAmDnG;AAED;;;;;;;GAOG;AACH,wBAAgB,gCAAgC,CAC9C,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS,GACzC,mBAAmB,GAAG,IAAI,CA2B5B"}
+{"version":3,"file":"testIntegrity.d.ts","sourceRoot":"","sources":["../../src/agent/testIntegrity.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AAEH,MAAM,WAAW,eAAe;IAC9B,gEAAgE;IAChE,MAAM,EAAE,MAAM,CAAC;IACf,0CAA0C;IAC1C,IAAI,EAAE,YAAY,GAAG,mBAAmB,GAAG,iBAAiB,GAAG,kBAAkB,GAAG,gBAAgB,CAAC;CACtG;AAED,MAAM,WAAW,mBAAmB;IAClC,UAAU,EAAE,OAAO,CAAC;IACpB,UAAU,EAAE,OAAO,CAAC;IACpB,QAAQ,EAAE,eAAe,EAAE,CAAC;CAC7B;AAQD,wBAAgB,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAGhD;AA4ED;;;GAGG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,mBAAmB,CAmDnG;AAED;;;;;;;GAOG;AACH,wBAAgB,gCAAgC,CAC9C,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS,GACzC,mBAAmB,GAAG,IAAI,CA8B5B;AAcD,kFAAkF;AAClF,wBAAgB,yBAAyB,CAAC,QAAQ,EAAE,eAAe,EAAE,GAAG,MAAM,CAI7E;AAED,8EAA8E;AAC9E,wBAAgB,yBAAyB,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,EAAE,CAQ9E;AAED,gFAAgF;AAChF,wBAAgB,wBAAwB,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,CAQ3E"}

package/dist/agent/testIntegrity.js

@@ -32,6 +32,9 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.isTestFile = isTestFile;
 exports.analyzeTestEdit = analyzeTestEdit;
 exports.analyzeWriteToolForTestIntegrity = analyzeWriteToolForTestIntegrity;
+exports.encodeTestIntegrityMarker = encodeTestIntegrityMarker;
+exports.decodeTestIntegrityMarker = decodeTestIntegrityMarker;
+exports.stripTestIntegrityMarker = stripTestIntegrityMarker;
 // ── Is this a test file? ──────────────────────────────────────────────────────
 // Covers the common conventions across JS/TS, Python, Go, Java/Kotlin, Ruby,
 // Rust, PHP, C#. Path-based (fast, language-agnostic).
@@ -192,8 +195,55 @@ function analyzeWriteToolForTestIntegrity(toolName, input) {
     if (toolName === 'write_file') {
         const content = typeof input.content === 'string' ? input.content : '';
         // No prior content available at this layer → additive-only ('' → content).
+        // The executor closes this gap: when write_file OVERWRITES an existing test
+        // file it has the prior on-disk content, runs the full old→new analysis, and
+        // embeds a TEST_INTEGRITY_MARKER in its output which ledgerRecord parses.
         return analyzeTestEdit(path, '', content);
     }
     return null;
 }
+// ── write_file overwrite bridge ────────────────────────────────────────────────
+// A write_file that overwrites an existing test file can silently REMOVE
+// assertions / test cases — invisible to analyzeWriteToolForTestIntegrity above
+// because the prior content isn't available at the loop layer. The executor DOES
+// have the prior content (it reads it to preserve mode / run the cross-file
+// check), so it runs analyzeTestEdit(path, priorContent, newContent) and, when
+// suspicious, appends a machine-readable marker to its tool output. ledgerRecord
+// parses that marker so a write_file overwrite gets the exact same ledger entry
+// and pre-finish nudge as an edit_file weakening would.
+const TEST_INTEGRITY_MARKER_OPEN = '\u0000NEXRALL_TEST_INTEGRITY\u0000';
+const TEST_INTEGRITY_MARKER_CLOSE = '\u0000/NEXRALL_TEST_INTEGRITY\u0000';
+/** Encode findings as an invisible marker to append to a tool's output string. */
+function encodeTestIntegrityMarker(findings) {
+    if (!findings.length)
+        return '';
+    const payload = findings.map((f) => f.reason).join('\u0001');
+    return `${TEST_INTEGRITY_MARKER_OPEN}${payload}${TEST_INTEGRITY_MARKER_CLOSE}`;
+}
+/** Extract weakening reasons from a tool output that may carry the marker. */
+function decodeTestIntegrityMarker(output) {
+    if (!output)
+        return [];
+    const start = output.indexOf(TEST_INTEGRITY_MARKER_OPEN);
+    if (start === -1)
+        return [];
+    const from = start + TEST_INTEGRITY_MARKER_OPEN.length;
+    const end = output.indexOf(TEST_INTEGRITY_MARKER_CLOSE, from);
+    if (end === -1)
+        return [];
+    return output.slice(from, end).split('\u0001').filter(Boolean);
+}
+/** Strip the (invisible) marker from output before it is shown to the model. */
+function stripTestIntegrityMarker(output) {
+    if (!output)
+        return output ?? '';
+    const start = output.indexOf(TEST_INTEGRITY_MARKER_OPEN);
+    if (start === -1)
+        return output;
+    const end = output.indexOf(TEST_INTEGRITY_MARKER_CLOSE, start);
+    if (end === -1)
+        return output;
+    const cleaned = output.slice(0, start) + output.slice(end + TEST_INTEGRITY_MARKER_CLOSE.length);
+    return cleaned.replace(/\n{3,}/g, '\n\n').trimEnd();
+}
 //# sourceMappingURL=testIntegrity.js.map

package/dist/tools/executor.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"executor.d.ts","sourceRoot":"","sources":["../../src/tools/executor.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,UAAU,CAAC;AAC3C,OAAO,EAAyB,KAAK,aAAa,EAAE,MAAM,WAAW,CAAC;AAy8DtE,wBAAsB,WAAW,CAC/B,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAC9B,WAAW,CAAC,EAAE;IAAE,OAAO,EAAE,OAAO,CAAA;CAAE,EAClC,OAAO,CAAC,EAAE,aAAa,EACvB,OAAO,CAAC,EAAE,MAAM,EAChB,UAAU,CAAC,EAAE,MAAM,GAClB,OAAO,CAAC,UAAU,CAAC,CAiBrB"}
+{"version":3,"file":"executor.d.ts","sourceRoot":"","sources":["../../src/tools/executor.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,UAAU,CAAC;AAC3C,OAAO,EAAyB,KAAK,aAAa,EAAE,MAAM,WAAW,CAAC;AAk9DtE,wBAAsB,WAAW,CAC/B,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAC9B,WAAW,CAAC,EAAE;IAAE,OAAO,EAAE,OAAO,CAAA;CAAE,EAClC,OAAO,CAAC,EAAE,aAAa,EACvB,OAAO,CAAC,EAAE,MAAM,EAChB,UAAU,CAAC,EAAE,MAAM,GAClB,OAAO,CAAC,UAAU,CAAC,CAiBrB"}

package/dist/tools/executor.js

@@ -44,6 +44,7 @@ const child_process_1 = require("child_process");
 const sandbox_1 = require("./sandbox");
 const auth_1 = require("../auth");
 const editCompleteness_1 = require("../agent/editCompleteness");
+const testIntegrity_1 = require("../agent/testIntegrity");
 const crossFile_1 = require("../agent/crossFile");
 const client_1 = require("../api/client");
 const symbols_1 = require("./symbols");
@@ -370,7 +371,16 @@ async function writeFile(input, workDir) {
             return { output: `Created ${resolved} (${lines} lines, ${bytes} bytes)` };
         }
         const xfile = crossFileBreakageWarning(resolved, normalizeLF(priorContent), normalizeLF(content), workDir ?? process.cwd());
-        return { output: `Overwrote ${resolved} (${lines} lines, ${bytes} bytes)${xfile}` };
+        // Reward-hacking guard: a write_file that OVERWRITES an existing test file can
+        // silently drop assertions / test cases. The loop layer can't see the prior
+        // content — but we can (we just read it). Run the full old→new analysis and
+        // embed an invisible marker that ledgerRecord parses, so this path gets the
+        // same ledger entry + pre-finish nudge as an edit_file weakening would.
+        let tiMarker = '';
+        const ti = (0, testIntegrity_1.analyzeTestEdit)(resolved, normalizeLF(priorContent), normalizeLF(content));
+        if (ti.suspicious)
+            tiMarker = (0, testIntegrity_1.encodeTestIntegrityMarker)(ti.findings);
+        return { output: `Overwrote ${resolved} (${lines} lines, ${bytes} bytes)${xfile}${tiMarker}` };
     }
     catch (err) {
         return { error: err.message };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nexrall/code-core",
-  "version": "1.4.0",
+  "version": "1.4.1",
   "description": "Core agent loop, tools, and extension primitives for Nexrall Code — embed an AI coding agent in any Node.js application.",
   "license": "MIT",
   "author": "Nexrall <support@nexrall.com> (https://nexrall.com)",