project-librarian 0.5.7 → 0.5.9

package/README.ko.md

@@ -104,15 +104,32 @@ Project Librarian은 작은 첫 읽기와 더 깊은 정본으로 가는 경로
 
 여기 수치는 모든 상황에 대한 약속이 아니라 관리자 릴리스 근거입니다. 모든 값은 실제 Codex JSONL 사용량과 로컬 실행 시간이며, Project Librarian이 없는 `organic` 대조군과 격리된 환경에서 비교했습니다. 위키 라우팅 트랙과 코드 그래프 트랙은 분리해 측정하며, 한 트랙의 우위가 다른 트랙의 주장을 뒷받침하지 않습니다.
 
-위키 트랙, 대조군 대비 비용 가중 토큰:
+최신 clean 합성 위키 라우팅 트랙 릴리스 근거는 2026-06-29에 `gpt-5.5`, `perf/small-repo-code-evidence-safeguards` 브랜치의 `ae79390` 커밋에서 측정했습니다. 42개 시나리오, 21개 with/without 쌍을 각 3회 측정하고 1회 예열했으며, 전체 주장 게이트는 통과했고 주장 ledger는 이 보고서를 release-claimable로 분류했습니다. repair 실행은 보존된 raw JSONL에서 claimable 측정 실행 125개를 재사용하고 남은 실패 슬롯 1개를 다시 측정했습니다. 이 주장은 합성 위키 라우팅 트랙과 표기된 작업 유형에 한정되며, 코드 그래프 동작, 실제 저장소, 모든 에이전트 표면, 모든 질문 형태에 대한 주장이 아닙니다.
 
-| 규모 | decision_lookup | aggregation | multi_session |
-| --- | --- | --- | --- |
-| 소형 | 14.4% 적음 | 81.0% 많음 | 22.0% 적음 |
-| 중형 | 52.0% 적음 | 19.0% 적음 | 54.1% 적음 |
-| 대형 | 71.1% 적음 | 29.0% 적음 | 71.8% 적음 |
+위키 트랙 집계, 대조군 대비 변화:
 
-최신 합성 위키 라우팅 트랙 릴리스 후보는 2026-06-19에 `gpt-5.5`로 측정했으며, 42개 시나리오를 각 3회 측정하고 1회 예열했습니다. 전체 주장 게이트는 통과했습니다. 이 주장은 합성 위키 라우팅 트랙과 표기된 작업 유형에 한정되며, 코드 그래프 동작, 실제 저장소, 모든 에이전트 표면, 모든 질문 형태에 대한 주장이 아닙니다.
+| 규모 | 비용 가중 토큰 | 전체 토큰 | 실행 시간 | 명령 수 | 도구 출력 |
+| --- | ---: | ---: | ---: | ---: | ---: |
+| 전체 | 51.39% 적음 | 48.67% 적음 | 19.83% 적음 | 18.40% 적음 | 85.88% 적음 |
+| 소형 | 21.59% 적음 | 9.31% 적음 | 13.91% 적음 | 12.64% 적음 | 43.89% 적음 |
+| 중형 | 45.95% 적음 | 37.42% 적음 | 11.88% 적음 | 4.30% 적음 | 69.12% 적음 |
+| 대형 | 66.97% 적음 | 69.87% 적음 | 31.90% 적음 | 35.19% 적음 | 95.58% 적음 |
+
+위키 트랙 작업 유형별 비용 가중 토큰 변화, 전체 규모 합산:
+
+| 작업 유형 | 변화 |
+| --- | ---: |
+| onboarding | 65.95% 적음 |
+| decision_lookup | 48.46% 적음 |
+| code_impact | 55.61% 적음 |
+| release_policy | 58.22% 적음 |
+| change_location | 29.59% 적음 |
+| multi_session | 52.58% 적음 |
+| aggregation | 42.53% 적음 |
+
+규모별 값은 서로 다릅니다. 소형/중형/대형 작업 유형별 실제 행렬과 아직 역전이 남은 셀은 [벤치마크 근거](docs/ko/benchmarks.md#위키-트랙)에 있습니다.
+
+실행 시간과 명령 수의 한계도 있습니다. 이번 보고서에서 `code_impact`, `change_location`, `aggregation`은 토큰과 출력 바이트가 모두 개선됐지만, 실행 시간이나 명령 수에서는 일부 역전이 남아 있습니다.
 
 코드 그래프 트랙, 대조군 대비 비용 가중 토큰:
 

package/README.md

@@ -104,15 +104,32 @@ See [Usage](docs/usage.md) for install scopes, runner paths, generated files, mi
 
 These numbers are maintainer release evidence, not a blanket promise. Every value is real Codex JSONL usage and local wall-clock time, measured hermetically against an `organic` control with no Project Librarian. The wiki-routing track and the code-graph track are measured separately; a win on one never backs a claim about the other.
 
-Wiki track, cost-weighted tokens vs control:
+Latest clean synthetic wiki-routing track release evidence: 2026-06-29, `gpt-5.5`, branch `perf/small-repo-code-evidence-safeguards` at `ae79390`, 42 scenarios, 21 with/without pairs, 3 measured runs plus 1 warmup each. The overall claim gate **passed**, and the claim ledger classified the report as release-claimable. A repair run reused 125 claimable measured runs from retained raw JSONL and remeasured the remaining failed slot. This is bounded to the synthetic wiki-routing track and listed task families; it is not a claim about code-graph behavior, real repositories, every agent surface, or every question shape.
 
-| Scale | decision_lookup | aggregation | multi_session |
-| --- | --- | --- | --- |
-| Small | 14.4% less | 81.0% more | 22.0% less |
-| Medium | 52.0% less | 19.0% less | 54.1% less |
-| Large | 71.1% less | 29.0% less | 71.8% less |
+Wiki track aggregate deltas vs control:
 
-Latest synthetic wiki-routing track release candidate: 2026-06-19, `gpt-5.5`, 42 scenarios, 3 measured runs plus 1 warmup each. The overall claim gate **passed**. This is bounded to the synthetic wiki-routing track and listed task families; it is not a claim about code-graph behavior, real repositories, every agent surface, or every question shape.
+| Scale | Cost-weighted tokens | Total tokens | Wall time | Commands | Tool output |
+| --- | ---: | ---: | ---: | ---: | ---: |
+| All | 51.39% less | 48.67% less | 19.83% less | 18.40% less | 85.88% less |
+| Small | 21.59% less | 9.31% less | 13.91% less | 12.64% less | 43.89% less |
+| Medium | 45.95% less | 37.42% less | 11.88% less | 4.30% less | 69.12% less |
+| Large | 66.97% less | 69.87% less | 31.90% less | 35.19% less | 95.58% less |
+
+Wiki track task-family cost-weighted token deltas vs control, all scales combined:
+
+| Task family | Delta |
+| --- | ---: |
+| onboarding | 65.95% less |
+| decision_lookup | 48.46% less |
+| code_impact | 55.61% less |
+| release_policy | 58.22% less |
+| change_location | 29.59% less |
+| multi_session | 52.58% less |
+| aggregation | 42.53% less |
+
+Scale-specific values vary; see [Benchmark Evidence](docs/benchmarks.md#wiki-track) for the small/medium/large task-family matrix and the cells that still regress.
+
+Timing and command-count caveat: `code_impact`, `change_location`, and `aggregation` still had wall-time or command-count regressions in this report, even though token and output-byte metrics improved for every task family.
 
 Code-graph track, cost-weighted tokens vs control:
 

package/dist/code-index/modes.js

@@ -131,6 +131,36 @@ function emitCodeIndexPhaseTimings(timings) {
         return;
     console.error(`code_index_phase_timings ${JSON.stringify(timings)}`);
 }
+function finishCodeEvidencePhaseTimings(started, timings) {
+    timings.total_ms = Number(elapsedMs(started).toFixed(3));
+    emitCodeIndexPhaseTimings(timings);
+}
+function runWithCodeEvidenceDatabase(runtime, operation, options = {}) {
+    const totalStarted = process.hrtime.bigint();
+    const phaseTimings = {};
+    let database;
+    try {
+        measurePhase(phaseTimings, "require_existing_index_ms", () => runtime.requireExistingIndex());
+        options.beforeOpen?.();
+        database = measurePhase(phaseTimings, "open_database_ms", () => runtime.openDatabase(runtime.codeEvidenceDatabasePath().absolutePath));
+        operation(database, phaseTimings);
+    }
+    finally {
+        if (database) {
+            const openedDatabase = database;
+            measurePhase(phaseTimings, "close_database_ms", () => openedDatabase.close());
+        }
+        finishCodeEvidencePhaseTimings(totalStarted, phaseTimings);
+    }
+}
+function requireCompatibleDatabaseTimed(database, runtime, timings) {
+    measurePhase(timings, "compatibility_ms", () => requireCompatibleDatabase(database, runtime));
+}
+function checkedCodeIndexStaleness(database, runtime, timings) {
+    const staleness = measurePhase(timings, "staleness_ms", () => runtime.codeIndexStaleness(database));
+    runtime.warnIfCodeIndexStale(database, staleness);
+    return staleness;
+}
 function configureBulkWriteConnection(database) {
     database.exec(`
     PRAGMA synchronous = OFF;
@@ -199,13 +229,11 @@ function runCodeIndexMode(runtime) {
         if (!args_1.codeIndexIncrementalMode) {
             try {
                 measurePhase(phaseTimings, "native_helper_ms", () => runtime.runNativeCodeIndexMode({ databasePath, discoveredFiles, parserMode, requestedEngine, scopes }));
-                phaseTimings.total_ms = Number(elapsedMs(totalStarted).toFixed(3));
-                emitCodeIndexPhaseTimings(phaseTimings);
+                finishCodeEvidencePhaseTimings(totalStarted, phaseTimings);
                 return;
             }
             catch (error) {
-                phaseTimings.total_ms = Number(elapsedMs(totalStarted).toFixed(3));
-                emitCodeIndexPhaseTimings(phaseTimings);
+                finishCodeEvidencePhaseTimings(totalStarted, phaseTimings);
                 throw error;
             }
         }
@@ -278,8 +306,7 @@ function runCodeIndexMode(runtime) {
                     scopes,
                     unchangedFiles,
                 }));
-                phaseTimings.total_ms = Number(elapsedMs(totalStarted).toFixed(3));
-                emitCodeIndexPhaseTimings(phaseTimings);
+                finishCodeEvidencePhaseTimings(totalStarted, phaseTimings);
                 return;
             }
         }
@@ -303,7 +330,6 @@ function runCodeIndexMode(runtime) {
                 (0, schema_1.createSecondaryIndexes)(activeDatabase);
             activeDatabase.exec("COMMIT");
         });
-        phaseTimings.total_ms = Number(elapsedMs(totalStarted).toFixed(3));
         console.log("Project wiki code evidence index complete.");
         console.log(`database: ${databasePath.relativePath}`);
         console.log(`mode: ${incremental ? "incremental" : "full"}`);
@@ -316,7 +342,7 @@ function runCodeIndexMode(runtime) {
         console.log(`reindexed_files: ${reindexedFiles.length}`);
         console.log(`deleted_files: ${deletedPaths.length}`);
         console.log(`unchanged_files: ${unchangedFiles}`);
-        emitCodeIndexPhaseTimings(phaseTimings);
+        finishCodeEvidencePhaseTimings(totalStarted, phaseTimings);
     }
     catch (error) {
         try {
@@ -336,123 +362,93 @@ function runCodeQueryMode(runtime) {
         console.error("missing SQL: use --code-query \"select ...\"");
         process.exit(1);
     }
-    runtime.requireExistingIndex();
-    if (!(0, code_index_sql_1.isReadOnlySql)(args_1.codeQuerySql)) {
-        console.error("code queries must be read-only SQL starting with SELECT or WITH");
-        process.exit(1);
-    }
-    const database = runtime.openDatabase(runtime.codeEvidenceDatabasePath().absolutePath);
-    try {
+    runWithCodeEvidenceDatabase(runtime, (database, phaseTimings) => {
         database.exec("PRAGMA query_only = ON");
-        requireCompatibleDatabase(database, runtime);
-        runtime.warnIfCodeIndexStale(database);
-        printRows(database.prepare(args_1.codeQuerySql).all());
-    }
-    finally {
-        database.close();
-    }
+        requireCompatibleDatabaseTimed(database, runtime, phaseTimings);
+        checkedCodeIndexStaleness(database, runtime, phaseTimings);
+        const rows = measurePhase(phaseTimings, "query_ms", () => database.prepare(args_1.codeQuerySql).all());
+        measurePhase(phaseTimings, "render_ms", () => printRows(rows));
+    }, {
+        beforeOpen() {
+            if (!(0, code_index_sql_1.isReadOnlySql)(args_1.codeQuerySql)) {
+                console.error("code queries must be read-only SQL starting with SELECT or WITH");
+                process.exit(1);
+            }
+        },
+    });
 }
 function runCodeReportMode(runtime) {
-    runtime.requireExistingIndex();
-    const database = runtime.openDatabase(runtime.codeEvidenceDatabasePath().absolutePath);
-    try {
-        requireCompatibleDatabase(database, runtime);
-        const staleness = runtime.codeIndexStaleness(database);
-        runtime.warnIfCodeIndexStale(database, staleness);
-        const report = runtime.codeReportForRequestedSection(database, args_1.codeReportSection, { staleness });
+    runWithCodeEvidenceDatabase(runtime, (database, phaseTimings) => {
+        requireCompatibleDatabaseTimed(database, runtime, phaseTimings);
+        const staleness = checkedCodeIndexStaleness(database, runtime, phaseTimings);
+        const report = measurePhase(phaseTimings, "query_ms", () => runtime.codeReportForRequestedSection(database, args_1.codeReportSection, { staleness }));
         if (!report)
             runtime.fail((0, reports_1.invalidCodeReportSectionMessage)(args_1.codeReportSection));
-        printJson(report);
-    }
-    finally {
-        database.close();
-    }
+        measurePhase(phaseTimings, "render_ms", () => printJson(report));
+    });
 }
 function runCodeStatusMode(runtime) {
-    runtime.requireExistingIndex();
-    const database = runtime.openDatabase(runtime.codeEvidenceDatabasePath().absolutePath);
-    try {
-        requireCompatibleDatabase(database, runtime);
-        const rows = database.prepare(`
+    runWithCodeEvidenceDatabase(runtime, (database, phaseTimings) => {
+        requireCompatibleDatabaseTimed(database, runtime, phaseTimings);
+        const rows = measurePhase(phaseTimings, "query_ms", () => database.prepare(`
       SELECT 'files' AS metric, count(*) AS value FROM files
       UNION ALL SELECT 'symbols', count(*) FROM symbols
       UNION ALL SELECT 'imports', count(*) FROM imports
       UNION ALL SELECT 'routes', count(*) FROM routes
       UNION ALL SELECT 'edges', count(*) FROM edges
       UNION ALL SELECT 'configs', count(*) FROM configs
-    `).all();
-        const staleness = runtime.codeIndexStaleness(database);
+    `).all());
+        const staleness = measurePhase(phaseTimings, "staleness_ms", () => runtime.codeIndexStaleness(database));
         rows.push({ metric: "stale_files", value: staleness.added + staleness.changed + staleness.deleted }, { metric: "stale_changed_files", value: staleness.changed }, { metric: "stale_added_files", value: staleness.added }, { metric: "stale_deleted_files", value: staleness.deleted });
-        printRows(rows);
-    }
-    finally {
-        database.close();
-    }
+        measurePhase(phaseTimings, "render_ms", () => printRows(rows));
+    });
 }
 function runCodeIndexHealthMode(runtime) {
     printJson(runtime.codeIndexHealth());
 }
 function runCodeFilesMode(runtime) {
-    runtime.requireExistingIndex();
-    const database = runtime.openDatabase(runtime.codeEvidenceDatabasePath().absolutePath);
-    try {
-        requireCompatibleDatabase(database, runtime);
-        runtime.warnIfCodeIndexStale(database);
-        printRows(database.prepare("SELECT path, language, profile, kind, lines, bytes FROM files ORDER BY path").all());
-    }
-    finally {
-        database.close();
-    }
+    runWithCodeEvidenceDatabase(runtime, (database, phaseTimings) => {
+        requireCompatibleDatabaseTimed(database, runtime, phaseTimings);
+        checkedCodeIndexStaleness(database, runtime, phaseTimings);
+        const rows = measurePhase(phaseTimings, "query_ms", () => database.prepare("SELECT path, language, profile, kind, lines, bytes FROM files ORDER BY path").all());
+        measurePhase(phaseTimings, "render_ms", () => printRows(rows));
+    });
 }
 function runCodeImpactMode(runtime) {
     if (!args_1.codeImpactTarget.trim()) {
         console.error("missing impact target: use --code-impact \"path-or-symbol-or-module\"");
         process.exit(1);
     }
-    runtime.requireExistingIndex();
-    const database = runtime.openDatabase(runtime.codeEvidenceDatabasePath().absolutePath);
-    try {
-        requireCompatibleDatabase(database, runtime);
-        const staleness = runtime.codeIndexStaleness(database);
-        runtime.warnIfCodeIndexStale(database, staleness);
-        printJson(runtime.codeImpact(database, args_1.codeImpactTarget.trim(), { staleness }));
-    }
-    finally {
-        database.close();
-    }
+    runWithCodeEvidenceDatabase(runtime, (database, phaseTimings) => {
+        requireCompatibleDatabaseTimed(database, runtime, phaseTimings);
+        const staleness = checkedCodeIndexStaleness(database, runtime, phaseTimings);
+        const impact = measurePhase(phaseTimings, "query_ms", () => runtime.codeImpact(database, args_1.codeImpactTarget.trim(), { staleness }));
+        measurePhase(phaseTimings, "render_ms", () => printJson(impact));
+    });
 }
 function runCodeContextPackMode(runtime) {
     if (!args_1.codeContextPackTarget.trim()) {
         console.error("missing context pack query: use --code-context-pack \"path-or-symbol-or-route\"");
         process.exit(1);
     }
-    runtime.requireExistingIndex();
-    const database = runtime.openDatabase(runtime.codeEvidenceDatabasePath().absolutePath);
-    try {
-        requireCompatibleDatabase(database, runtime);
-        const staleness = runtime.codeIndexStaleness(database);
-        runtime.warnIfCodeIndexStale(database, staleness);
-        console.log(runtime.codeContextPack(database, args_1.codeContextPackTarget.trim(), { staleness }));
-    }
-    finally {
-        database.close();
-    }
+    runWithCodeEvidenceDatabase(runtime, (database, phaseTimings) => {
+        requireCompatibleDatabaseTimed(database, runtime, phaseTimings);
+        const staleness = checkedCodeIndexStaleness(database, runtime, phaseTimings);
+        const pack = measurePhase(phaseTimings, "query_ms", () => runtime.codeContextPack(database, args_1.codeContextPackTarget.trim(), { staleness }));
+        measurePhase(phaseTimings, "render_ms", () => console.log(pack));
+    });
 }
 function runCodeSearchSymbolMode(runtime) {
     if (!args_1.codeSearchSymbol.trim()) {
         console.error("missing symbol search term: use --code-search-symbol \"term\"");
         process.exit(1);
     }
-    runtime.requireExistingIndex();
-    const database = runtime.openDatabase(runtime.codeEvidenceDatabasePath().absolutePath);
-    try {
-        requireCompatibleDatabase(database, runtime);
-        runtime.warnIfCodeIndexStale(database);
-        printRows((0, search_1.searchSymbols)(database, args_1.codeSearchSymbol.trim()));
-    }
-    finally {
-        database.close();
-    }
+    runWithCodeEvidenceDatabase(runtime, (database, phaseTimings) => {
+        requireCompatibleDatabaseTimed(database, runtime, phaseTimings);
+        checkedCodeIndexStaleness(database, runtime, phaseTimings);
+        const rows = measurePhase(phaseTimings, "query_ms", () => (0, search_1.searchSymbols)(database, args_1.codeSearchSymbol.trim()));
+        measurePhase(phaseTimings, "render_ms", () => printRows(rows));
+    });
 }
 function isCodeEvidenceModeFor(flags) {
     return Boolean(flags.codeContextPackTarget)

package/dist/code-index/ownership.js

@@ -65,15 +65,126 @@ function readJsonObject(relativePath) {
 function workspacePatternsFromRootPackage() {
     const rootPackage = readJsonObject("package.json");
     const workspaces = rootPackage?.workspaces;
-    if (Array.isArray(workspaces))
-        return workspaces.filter((value) => typeof value === "string");
+    if (Array.isArray(workspaces)) {
+        return workspaces
+            .filter((value) => typeof value === "string")
+            .map((pattern) => ({ pattern, source: "package.json workspaces" }));
+    }
     if (workspaces && typeof workspaces === "object" && !Array.isArray(workspaces)) {
         const packages = workspaces.packages;
-        if (Array.isArray(packages))
-            return packages.filter((value) => typeof value === "string");
+        if (Array.isArray(packages)) {
+            return packages
+                .filter((value) => typeof value === "string")
+                .map((pattern) => ({ pattern, source: "package.json workspaces" }));
+        }
     }
     return [];
 }
+function trimYamlComment(value) {
+    let quote = "";
+    for (let index = 0; index < value.length; index += 1) {
+        const character = value[index] ?? "";
+        if (quote) {
+            if (character === quote)
+                quote = "";
+            continue;
+        }
+        if (character === "\"" || character === "'") {
+            quote = character;
+            continue;
+        }
+        if (character === "#" && (index === 0 || /\s/.test(value[index - 1] ?? "")))
+            return value.slice(0, index).trimEnd();
+    }
+    return value.trimEnd();
+}
+function parseYamlStringScalar(value) {
+    const trimmed = trimYamlComment(value).trim();
+    if (!trimmed)
+        return null;
+    if (trimmed.startsWith("\"")) {
+        const match = trimmed.match(/^"((?:\\.|[^"\\])*)"/);
+        if (!match)
+            return null;
+        return match[1]?.replace(/\\"/g, "\"") ?? "";
+    }
+    if (trimmed.startsWith("'")) {
+        const match = trimmed.match(/^'((?:''|[^'])*)'/);
+        if (!match)
+            return null;
+        return match[1]?.replace(/''/g, "'") ?? "";
+    }
+    return trimmed;
+}
+function splitYamlFlowArray(value) {
+    const trimmed = trimYamlComment(value).trim();
+    if (!trimmed.startsWith("[") || !trimmed.endsWith("]"))
+        return null;
+    const entries = [];
+    let quote = "";
+    let current = "";
+    for (const character of trimmed.slice(1, -1)) {
+        if (quote) {
+            current += character;
+            if (character === quote)
+                quote = "";
+            continue;
+        }
+        if (character === "\"" || character === "'") {
+            quote = character;
+            current += character;
+            continue;
+        }
+        if (character === ",") {
+            entries.push(current);
+            current = "";
+            continue;
+        }
+        current += character;
+    }
+    entries.push(current);
+    return entries;
+}
+function workspacePatternsFromPnpmWorkspace() {
+    const filePath = "pnpm-workspace.yaml";
+    if (!(0, workspace_1.containedProjectFileStat)(filePath))
+        return [];
+    const lines = (0, workspace_1.read)(filePath).split(/\r?\n/);
+    const patterns = [];
+    let packagesIndent = null;
+    for (const line of lines) {
+        if (!line.trim() || line.trimStart().startsWith("#"))
+            continue;
+        const indent = line.match(/^\s*/)?.[0].length ?? 0;
+        const packagesMatch = line.match(/^(\s*)packages\s*:\s*(.*)$/);
+        if (packagesMatch) {
+            packagesIndent = packagesMatch[1]?.length ?? 0;
+            const inlineArray = splitYamlFlowArray(packagesMatch[2] ?? "");
+            if (inlineArray) {
+                for (const entry of inlineArray) {
+                    const pattern = parseYamlStringScalar(entry);
+                    if (pattern && !pattern.startsWith("!"))
+                        patterns.push({ pattern, source: "pnpm-workspace.yaml packages" });
+                }
+            }
+            continue;
+        }
+        if (packagesIndent === null)
+            continue;
+        if (indent <= packagesIndent)
+            break;
+        const item = line.slice(indent).match(/^-\s+(.+)$/);
+        if (!item)
+            continue;
+        const pattern = parseYamlStringScalar(item[1] ?? "");
+        if (pattern && !pattern.startsWith("!"))
+            patterns.push({ pattern, source: "pnpm-workspace.yaml packages" });
+    }
+    return patterns;
+}
+function workspacePatterns() {
+    return [...workspacePatternsFromRootPackage(), ...workspacePatternsFromPnpmWorkspace()];
+}
 function workspacePatternCandidates(pattern) {
     const normalized = (0, workspace_1.normalizePath)(pattern).replace(/^\/+/, "").replace(/\/+$/, "");
     if (!normalized || normalized.includes(".."))
@@ -94,8 +205,8 @@ function workspacePatternCandidates(pattern) {
 }
 function workspacePackages() {
     const packages = new Map();
-    for (const pattern of workspacePatternsFromRootPackage()) {
-        for (const candidate of workspacePatternCandidates(pattern)) {
+    for (const workspacePattern of workspacePatterns()) {
+        for (const candidate of workspacePatternCandidates(workspacePattern.pattern)) {
             const packageJsonPath = (0, workspace_1.normalizePath)(path.join(candidate, "package.json"));
             if (!(0, workspace_1.containedProjectFileStat)(packageJsonPath))
                 continue;
@@ -104,8 +215,8 @@ function workspacePackages() {
             packages.set(candidate, {
                 name: packageName,
                 root: candidate,
-                source: "package.json workspaces",
-                workspace_pattern: pattern,
+                source: workspacePattern.source,
+                workspace_pattern: workspacePattern.pattern,
             });
         }
     }

package/dist/code-index/schema.js

@@ -47,7 +47,7 @@ exports.incrementalCompatibility = incrementalCompatibility;
 exports.codeIndexSnapshot = codeIndexSnapshot;
 const crypto = __importStar(require("node:crypto"));
 const workspace_1 = require("../workspace");
-exports.codeIndexSchemaVersion = "5";
+exports.codeIndexSchemaVersion = "6";
 function stableFtsRowid(parts) {
     const hash = crypto.createHash("sha256");
     for (const part of parts) {
@@ -158,7 +158,7 @@ function createIndexStatements(database) {
         insertMeta: database.prepare("INSERT OR REPLACE INTO meta (key, value) VALUES (?, ?)"),
         insertRoute: database.prepare("INSERT INTO routes (method, route, file_path, line, handler) VALUES (?, ?, ?, ?, ?)"),
         insertSymbol: database.prepare("INSERT INTO symbols (name, kind, file_path, line, signature) VALUES (?, ?, ?, ?, ?)"),
-        insertSymbolFts: database.prepare("INSERT INTO symbols_fts (name, kind, file_path, signature) VALUES (?, ?, ?, ?)"),
+        insertSymbolFts: database.prepare("INSERT INTO symbols_fts (rowid, name, kind, file_path, signature) VALUES (last_insert_rowid(), ?, ?, ?, ?)"),
     };
 }
 function removeIndexedFile(filePath, statements) {

package/dist/code-index/search.js

@@ -130,11 +130,7 @@ function searchSymbols(database, term, limit = 50) {
         const ftsRows = database.prepare(`
       SELECT symbols.name, symbols.kind, symbols.file_path, symbols.line, symbols.signature
       FROM symbols_fts
-      JOIN symbols
-        ON symbols.name = symbols_fts.name
-       AND symbols.kind = symbols_fts.kind
-       AND symbols.file_path = symbols_fts.file_path
-       AND symbols.signature = symbols_fts.signature
+      JOIN symbols ON symbols.id = symbols_fts.rowid
       WHERE symbols_fts MATCH ?
       ORDER BY bm25(symbols_fts, 8.0, 1.0, 4.0, 2.0), symbols.file_path, symbols.line
       LIMIT ?

package/dist/code-index.js

@@ -648,16 +648,22 @@ function codeContextScaleLine(fileCount) {
         ? `scale small (${fileCount} indexed files < ${code_index_file_policy_1.SMALL_REPO_FILE_THRESHOLD}); direct reads are usually cheaper for simple lookups`
         : `scale large (${fileCount} indexed files >= ${code_index_file_policy_1.SMALL_REPO_FILE_THRESHOLD}); indexed traversal is useful for impact-style context`;
 }
-function structuralSignature(value) {
-    const signature = (0, shared_1.oneLine)(String(value ?? ""));
-    const bodyStart = signature.indexOf("{");
-    return bodyStart >= 0 ? signature.slice(0, bodyStart).trimEnd() : signature;
-}
-function codeContextPack(database, query, options = {}) {
-    const normalized = query.trim();
-    if (!normalized)
-        return 'Code context pack: missing query; use --code-context-pack "path-or-symbol-or-route".';
-    const evidence = (0, evidence_1.collectCodeEvidence)(database, normalized, {
+function codeContextCollectorOptions(fileCount) {
+    if (fileCount < code_index_file_policy_1.SMALL_REPO_FILE_THRESHOLD) {
+        return {
+            edgeLimit: 8,
+            fileLimit: 8,
+            includeEdgeEvidenceMatches: false,
+            includeOwnerCodeowners: false,
+            includeRouteEdges: false,
+            importLimit: 12,
+            ownerSampleLimit: 3,
+            routeEdgeLimit: 0,
+            routeLimit: 8,
+            symbolLimit: 12,
+        };
+    }
+    return {
         edgeLimit: 30,
         fileLimit: 12,
         includeEdgeEvidenceMatches: true,
@@ -668,14 +674,26 @@ function codeContextPack(database, query, options = {}) {
         routeEdgeLimit: 0,
         routeLimit: 20,
         symbolLimit: 20,
-    });
-    const staleness = options.staleness ?? codeIndexStaleness(database);
+    };
+}
+function structuralSignature(value) {
+    const signature = (0, shared_1.oneLine)(String(value ?? ""));
+    const bodyStart = signature.indexOf("{");
+    return bodyStart >= 0 ? signature.slice(0, bodyStart).trimEnd() : signature;
+}
+function codeContextPack(database, query, options = {}) {
+    const normalized = query.trim();
+    if (!normalized)
+        return 'Code context pack: missing query; use --code-context-pack "path-or-symbol-or-route".';
     const coverage = (0, reports_1.evidenceCoverage)(database);
+    const indexedFileCount = Number(coverage.files ?? 0);
+    const evidence = (0, evidence_1.collectCodeEvidence)(database, normalized, codeContextCollectorOptions(indexedFileCount));
+    const staleness = options.staleness ?? codeIndexStaleness(database);
     const staleLabel = staleness.stale
         ? `STALE ${staleness.changed} changed, ${staleness.added} added, ${staleness.deleted} deleted`
         : "fresh";
     const lines = [
-        `Code context pack "${normalized}": ${evidence.files.length} file matches, ${evidence.symbols.length} symbols, ${evidence.routes.length} routes, ${evidence.imports.length} imports, ${evidence.incomingEdges.length} incoming / ${evidence.outgoingEdges.length} outgoing edges; index ${staleLabel}; ${codeContextScaleLine(Number(coverage.files ?? 0))}.`,
+        `Code context pack "${normalized}": ${evidence.files.length} file matches, ${evidence.symbols.length} symbols, ${evidence.routes.length} routes, ${evidence.imports.length} imports, ${evidence.incomingEdges.length} incoming / ${evidence.outgoingEdges.length} outgoing edges; index ${staleLabel}; ${codeContextScaleLine(indexedFileCount)}.`,
         "Evidence is structural only: paths, lines, signatures, routes, imports, edges, and owners; no source snippets are included.",
     ];
     pushBudgetedSection(lines, "Files:", evidence.files, 8, (row) => `  file-match ${String(row.path)} (${String(row.language)}, ${String(row.profile)}, ${Number(row.lines ?? 0)} lines)`);

package/dist/native/project-librarian-indexer-manifest.json

@@ -5,7 +5,7 @@
       "architecture": "arm64",
       "format": "mach-o",
       "path": "dist/native/darwin-arm64/project-librarian-indexer",
-      "sha256": "f5c69130f6c0441044141c761f1ccdc7e6ee4d06e327ce0f50d43a7aed76cc95",
+      "sha256": "848c29daf166e55672add5febf22fcd2ba261619a37969d9bc87cff06815787c",
       "size": 2697520,
       "triple": "darwin-arm64"
     },
@@ -13,7 +13,7 @@
       "architecture": "x64",
       "format": "mach-o",
       "path": "dist/native/darwin-x64/project-librarian-indexer",
-      "sha256": "25041398ba2de3ed749c14103d358ed4a7934a05459eb63c6cf5f6ae58c7fb03",
+      "sha256": "b419d82b7560defe1d43fc4d99d96b10222f1ead5bf6d7a6acabbae425ee7545",
       "size": 2770592,
       "triple": "darwin-x64"
     },
@@ -21,7 +21,7 @@
       "architecture": "arm64",
       "format": "elf",
       "path": "dist/native/linux-arm64/project-librarian-indexer",
-      "sha256": "01bf4437fd62446958438ac58e9ca8bef811bdc6d70a63fc1a0ca1b627521e67",
+      "sha256": "f3cc884acca4f5863fb6a6545a4a9f9a37a67872bfe780f4b14f0c4a8a667f16",
       "size": 3065520,
       "triple": "linux-arm64"
     },
@@ -29,31 +29,31 @@
       "architecture": "arm64",
       "format": "elf",
       "path": "dist/native/linux-arm64-musl/project-librarian-indexer",
-      "sha256": "b7287995bea678a6f55bd03761e5b0985d483e875d7594395b81bdb1e90a8dd9",
-      "size": 3100840,
+      "sha256": "80f7440c4d115d05805b63b8118a172ddeb5a374b7154593e5b1d0b804a938ae",
+      "size": 3100816,
       "triple": "linux-arm64-musl"
     },
     {
       "architecture": "x64",
       "format": "elf",
       "path": "dist/native/linux-x64/project-librarian-indexer",
-      "sha256": "5d93bb7576afbdde0acbde7983a874f2666f0d59227524c1eb42cc75b457c5ed",
-      "size": 3110696,
+      "sha256": "3d93408db344567d7de6f477b58fdaf6aba19c943336fb7ab9720e3e9fa68c31",
+      "size": 3110832,
       "triple": "linux-x64"
     },
     {
       "architecture": "x64",
       "format": "elf",
       "path": "dist/native/linux-x64-musl/project-librarian-indexer",
-      "sha256": "518aea0fc53ed7731751db4bb436a0be98f8583659d51aeff769bedb7a7cb8ab",
-      "size": 3247384,
+      "sha256": "0055ec9fefa385d6d637fa1e457308c38f892e0fbdb389db3fcbdf546f0daca8",
+      "size": 3247400,
       "triple": "linux-x64-musl"
     },
     {
       "architecture": "arm64",
       "format": "pe",
       "path": "dist/native/win32-arm64/project-librarian-indexer.exe",
-      "sha256": "e8e5200660a8852d1b7554daf0d16361e29e5599baa5cd406eb3f2266abd98f6",
+      "sha256": "f8ef16c98ff97c4f1dd38e2e08a034a40bb219f114f8270d165feaddd8166d8f",
       "size": 2098688,
       "triple": "win32-arm64"
     },
@@ -61,7 +61,7 @@
       "architecture": "x64",
       "format": "pe",
       "path": "dist/native/win32-x64/project-librarian-indexer.exe",
-      "sha256": "dc076bd61271f23ab87b428c7cbfa3af45911c00f7bf7986a2ed9d517e1cfb6f",
+      "sha256": "5a81b330741f11d10ab5d296e8cbc8c60f1596c30a45c5f31e41fab49e5b2603",
       "size": 2375680,
       "triple": "win32-x64"
     }

package/dist/templates.js

@@ -199,7 +199,7 @@ exports.startup = `${(0, exports.metadata)("startup-router", "short", "wiki/meta
 - This project is in an initial planning state unless the canonical wiki says otherwise.
 - Project truth lives in \`wiki/canonical/\`, future work in \`wiki/roadmaps/\` and \`wiki/plans/\`, project decisions in \`wiki/decisions/\`, and sources in \`wiki/sources/\`.
 - Wiki operating rules and wiki operating decisions live in \`wiki/meta/\`.
-- At session start, read only this file and \`wiki/index.md\` first; read detailed files on demand.
+- At session start, read only this file and \`wiki/index.md\` first; use the index as a route table, open matching detail files directly, and avoid broad repo/wiki search unless no route matches.
 - Project canonical content language is not fixed by this bootstrap. The LLM should choose the language that best matches the user and project context.
 - Completed roadmaps/plans are removed after truth/rationale/evidence capture.
 - Update the wiki in the same turn when project-planning content changes.
@@ -232,7 +232,7 @@ exports.startup = `${(0, exports.metadata)("startup-router", "short", "wiki/meta
 ## Token Discipline
 
 - Codex, Claude Code, Cursor, and Gemini CLI session-start hooks inject only this file and \`wiki/index.md\`.
-- Detailed files are selected by the "read when" rules in \`wiki/index.md\`.
+- Detailed files are selected by \`wiki/index.md\`; use broad wiki search only when no route matches or evidence conflicts.
 - Long decision history is not injected wholesale; read only relevant Decision Packs or ADRs.
 `;
 exports.index = `${(0, exports.metadata)("wiki-router", "short", "wiki/meta/wiki-ops-v1-decisions.md", "wiki page added, moved, removed, or routing changes")}
@@ -240,7 +240,7 @@ exports.index = `${(0, exports.metadata)("wiki-router", "short", "wiki/meta/wiki
 
 ## How To Use This Index
 
-This file is a router, not a file to expand into every answer. Read only the files that are relevant to the current question.
+This is a route table, not a page inventory. Open the matching route first; use broad wiki search or file listing only when no route matches or evidence conflicts.
 
 ## Language Policy
 

package/docs/benchmarks.md

@@ -17,17 +17,33 @@ Generated benchmark reports under `benchmarks/reports/llm/` are ignored by defau
 
 ## Wiki Track
 
-Cost-weighted tokens, Project Librarian vs control:
-
-| Scale | decision_lookup | aggregation | multi_session (2nd session) |
-| --- | --- | --- | --- |
-| Small | 14.4% less | 81.0% more | 22.0% less |
-| Medium | 52.0% less | 19.0% less | 54.1% less |
-| Large | 71.1% less | 29.0% less | 71.8% less |
-
-Latest synthetic wiki-track release candidate: 2026-06-19, `gpt-5.5`, 42 scenarios, 3 measured runs plus 1 warmup each. The overall claim gate **passed**: 42/42 scenarios passed correctness, all 42 scenarios were claimable, and every corpus gate met the 3-run minimum.
-
-The release claim is bounded to the synthetic wiki-routing track and the listed task families. It is not a claim about code-graph behavior, real repositories, every agent surface, or every question shape. Published boundaries remain visible: small `aggregation` still costs 81.0% more with the wiki, small `release_policy` costs 9.4% more in the full report, and `aggregation` stays slower at every scale even when token cost drops.
+Latest clean synthetic wiki-track release evidence: 2026-06-29, `gpt-5.5`, branch `perf/small-repo-code-evidence-safeguards` at `ae79390`, 42 scenarios, 21 with/without pairs, 3 measured runs plus 1 warmup each. The overall claim gate **passed**, and the claim ledger classified the report as release-claimable. A repair run reused 125 claimable measured runs from retained raw JSONL and remeasured the remaining failed slot.
+
+Cost-weighted tokens, Project Librarian vs control, all scales combined:
+
+| Task family | Delta |
+| --- | ---: |
+| onboarding | 65.95% less |
+| decision_lookup | 48.46% less |
+| code_impact | 55.61% less |
+| release_policy | 58.22% less |
+| change_location | 29.59% less |
+| multi_session | 52.58% less |
+| aggregation | 42.53% less |
+
+Scale-specific cost-weighted token deltas:
+
+| Task family | Small | Medium | Large |
+| --- | ---: | ---: | ---: |
+| onboarding | 60.88% less | 65.53% less | 69.17% less |
+| decision_lookup | 19.18% less | 51.29% less | 60.12% less |
+| code_impact | 26.61% less | 54.81% less | 68.43% less |
+| release_policy | 30.23% less | 62.16% less | 67.26% less |
+| change_location | 7.79% less | 2.86% more | 57.27% less |
+| multi_session | 54.29% more | 35.32% less | 82.77% less |
+| aggregation | 7.66% less | 51.32% less | 49.17% less |
+
+The release claim is bounded to the synthetic wiki-routing track and the listed task families. It is not a claim about code-graph behavior, real repositories, every agent surface, or every question shape. Published boundaries remain visible: small `multi_session` and medium `change_location` cost more with the wiki in the cost-weighted token metric. `code_impact`, `change_location`, and `aggregation` also still have wall-time or command-count regressions in this report, even though token and output-byte metrics improved at the task-family aggregate level.
 
 ## Code-Graph Track
 
@@ -52,7 +68,11 @@ Repositories under test:
 
 Question types:
 
+- **onboarding** - summarize what the project is, current risks, and where to read next.
 - **decision_lookup** - find the latest project decision and its date from the wiki.
+- **code_impact** - identify likely impacted files or areas for a benchmark report schema change.
+- **release_policy** - find the checks required before publishing or making benchmark claims.
+- **change_location** - find where an agent should edit to implement the Codex LLM benchmark runner.
 - **aggregation** - answer a question whose facts are scattered across several pages and must be synthesized.
 - **multi_session** - a second session on the same project, measuring whether the durable wiki helps the next session, not just the first.
 - **impact_trace** - trace the full set of direct and indirect importers for a changing module.

package/docs/ko/benchmarks.md

@@ -17,17 +17,33 @@ npm run benchmark:release -- --allow-codex-run
 
 ## 위키 트랙
 
-비용 가중 토큰, Project Librarian 대 대조군:
-
-| 규모 | decision_lookup | aggregation | multi_session (2번째 세션) |
-| --- | --- | --- | --- |
-| 소형 | 14.4% 적음 | 81.0% 많음 | 22.0% 적음 |
-| 중형 | 52.0% 적음 | 19.0% 적음 | 54.1% 적음 |
-| 대형 | 71.1% 적음 | 29.0% 적음 | 71.8% 적음 |
-
-최신 합성 위키 트랙 릴리스 후보는 2026-06-19에 `gpt-5.5`로 측정했으며, 42개 시나리오를 각 3회 측정하고 1회 예열했습니다. 전체 주장 게이트는 **통과**했습니다. 42/42개 시나리오가 정확성 검사를 통과했고, 42개 시나리오가 모두 claimable이었으며, 모든 corpus gate가 3-run 최소 조건을 충족했습니다.
-
-이 릴리스 주장은 합성 위키 라우팅 트랙과 표기된 작업 유형에 한정됩니다. 코드 그래프 동작, 실제 저장소, 모든 에이전트 표면, 모든 질문 형태에 대한 주장이 아닙니다. 숨기지 않는 한계도 남습니다. 소형 `aggregation`은 위키를 켰을 때 81.0% 더 비싸고, 전체 보고서의 소형 `release_policy`도 9.4% 더 비싸며, `aggregation`은 토큰 비용이 줄어드는 규모에서도 실행 시간은 매번 더 느립니다.
+최신 clean 합성 위키 트랙 릴리스 근거는 2026-06-29에 `gpt-5.5`, `perf/small-repo-code-evidence-safeguards` 브랜치의 `ae79390` 커밋에서 측정했습니다. 42개 시나리오, 21개 with/without 쌍을 각 3회 측정하고 1회 예열했으며, 전체 주장 게이트는 **통과**했고 주장 ledger는 이 보고서를 release-claimable로 분류했습니다. repair 실행은 보존된 raw JSONL에서 claimable 측정 실행 125개를 재사용하고 남은 실패 슬롯 1개를 다시 측정했습니다.
+
+비용 가중 토큰, Project Librarian 대 대조군, 전체 규모 합산:
+
+| 작업 유형 | 변화 |
+| --- | ---: |
+| onboarding | 65.95% 적음 |
+| decision_lookup | 48.46% 적음 |
+| code_impact | 55.61% 적음 |
+| release_policy | 58.22% 적음 |
+| change_location | 29.59% 적음 |
+| multi_session | 52.58% 적음 |
+| aggregation | 42.53% 적음 |
+
+규모별 비용 가중 토큰 변화:
+
+| 작업 유형 | 소형 | 중형 | 대형 |
+| --- | ---: | ---: | ---: |
+| onboarding | 60.88% 적음 | 65.53% 적음 | 69.17% 적음 |
+| decision_lookup | 19.18% 적음 | 51.29% 적음 | 60.12% 적음 |
+| code_impact | 26.61% 적음 | 54.81% 적음 | 68.43% 적음 |
+| release_policy | 30.23% 적음 | 62.16% 적음 | 67.26% 적음 |
+| change_location | 7.79% 적음 | 2.86% 많음 | 57.27% 적음 |
+| multi_session | 54.29% 많음 | 35.32% 적음 | 82.77% 적음 |
+| aggregation | 7.66% 적음 | 51.32% 적음 | 49.17% 적음 |
+
+이 릴리스 주장은 합성 위키 라우팅 트랙과 표기된 작업 유형에 한정됩니다. 코드 그래프 동작, 실제 저장소, 모든 에이전트 표면, 모든 질문 형태에 대한 주장이 아닙니다. 숨기지 않는 한계도 남습니다. 비용 가중 토큰 기준으로 소형 `multi_session`과 중형 `change_location`은 위키를 켰을 때 더 비쌉니다. 또한 `code_impact`, `change_location`, `aggregation`은 작업 유형 집계에서 토큰과 출력 바이트가 개선됐더라도 이번 보고서의 실행 시간이나 명령 수에는 일부 역전이 남아 있습니다.
 
 ## 코드 그래프 트랙
 
@@ -52,7 +68,11 @@ SHA로 고정한 오픈소스 저장소 2곳에서, 손으로 작성한 정답
 
 질문 유형:
 
+- **onboarding** - 프로젝트가 무엇인지, 현재 리스크, 다음에 읽을 문서를 요약합니다.
 - **decision_lookup** - 위키에서 가장 최근 프로젝트 결정과 그 날짜를 찾습니다.
+- **code_impact** - 벤치마크 보고서 스키마 변경 시 영향받을 파일이나 영역을 찾습니다.
+- **release_policy** - publish 또는 benchmark claim 전에 필요한 체크를 찾습니다.
+- **change_location** - Codex LLM benchmark runner 구현 시 어디를 수정해야 하는지 찾습니다.
 - **aggregation** - 여러 페이지에 흩어진 사실을 종합해야 답할 수 있는 질문입니다.
 - **multi_session** - 같은 프로젝트의 두 번째 세션으로, 지속되는 위키가 다음 세션에도 도움이 되는지 측정합니다.
 - **impact_trace** - 바뀌는 모듈의 직접/간접 importer 전체를 추적합니다.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "project-librarian",
-  "version": "0.5.7",
+  "version": "0.5.9",
   "description": "Create and maintain compact project context for humans and LLM coding agents.",
   "license": "MIT",
   "type": "commonjs",