@orquentra/codebuddy 2.5.0 → 2.6.0

package/dist/{chunk-4BYGTEJT.js → chunk-XBJQKCLU.js}

@@ -1,6 +1,6 @@
 import {
   getModelContextWindow
-} from "./chunk-JMWJ7V6Y.js";
+} from "./chunk-JPHJBCZB.js";
 
 // src/planner/budget.ts
 import { Tiktoken } from "js-tiktoken/lite";
@@ -100,4 +100,4 @@ export {
   computeContentHash,
   computeFactHash
 };
-//# sourceMappingURL=chunk-4BYGTEJT.js.map
+//# sourceMappingURL=chunk-XBJQKCLU.js.map

package/dist/cli/index.js

@@ -3,13 +3,13 @@
 // src/cli/index.ts
 var [command, subcommand] = process.argv.slice(2);
 if (command === "hooks" && subcommand === "lean") {
-  const { runLeanHookCommand } = await import("../lean-command-DA4ZNKG3.js");
+  const { runLeanHookCommand } = await import("../lean-command-C5YPFKHP.js");
   await runLeanHookCommand();
 } else if (command === "hooks" && subcommand === "pretool") {
   const { runPreToolHookCommand } = await import("../pretool-command-MQQ2UNQM.js");
   await runPreToolHookCommand();
 } else {
-  const { createCli } = await import("../main-BDVQAGAS.js");
+  const { createCli } = await import("../main-HFHLGID4.js");
   await createCli().parseAsync();
 }
 //# sourceMappingURL=index.js.map

package/dist/{cloud-GP6G3EAK.js → cloud-7WM3YLI4.js}

@@ -4,8 +4,8 @@ import {
   OpenRouterEmbeddingProvider,
   rerankHuggingFace,
   rerankOpenRouter
-} from "./chunk-KYK6RPAJ.js";
-import "./chunk-JMWJ7V6Y.js";
+} from "./chunk-JFDRBPYF.js";
+import "./chunk-JPHJBCZB.js";
 import "./chunk-PZ5AY32C.js";
 export {
   HF_ROUTER_BASE_URL,
@@ -14,4 +14,4 @@ export {
   rerankHuggingFace,
   rerankOpenRouter
 };
-//# sourceMappingURL=cloud-GP6G3EAK.js.map
+//# sourceMappingURL=cloud-7WM3YLI4.js.map

package/dist/index.js

@@ -19,9 +19,9 @@ import {
   registerCodeBuddyTools,
   resolveRiskPaths,
   startMcpServer
-} from "./chunk-Z5QB367W.js";
+} from "./chunk-VAKBJV26.js";
 import "./chunk-IHQM2ZDV.js";
-import "./chunk-CQPCHMBS.js";
+import "./chunk-SYWJG5NA.js";
 import {
   buildArchitectureMap,
   neighbours,
@@ -45,17 +45,17 @@ import {
   isLangSmithTracingEnabled,
   listFactsPage,
   runDoctor
-} from "./chunk-JDZ3VK26.js";
+} from "./chunk-UN57FKQX.js";
 import "./chunk-GDJH7XJC.js";
 import {
   clampBudgetForCallerModel,
   countTokens
-} from "./chunk-4BYGTEJT.js";
-import "./chunk-KYK6RPAJ.js";
+} from "./chunk-XBJQKCLU.js";
+import "./chunk-JFDRBPYF.js";
 import {
   LocalModelProvider
-} from "./chunk-3XYQGFPB.js";
-import "./chunk-JMWJ7V6Y.js";
+} from "./chunk-U4HAJLIB.js";
+import "./chunk-JPHJBCZB.js";
 import "./chunk-QDBDYCM2.js";
 import {
   PLAN_POLICIES,

package/dist/{lean-command-DA4ZNKG3.js → lean-command-C5YPFKHP.js}

@@ -1,15 +1,15 @@
 import {
   runLeanHookCommand
-} from "./chunk-ROQTT5ZO.js";
+} from "./chunk-N34XHWBA.js";
 import "./chunk-PYLH4MH5.js";
-import "./chunk-CQPCHMBS.js";
+import "./chunk-SYWJG5NA.js";
 import "./chunk-L5AYZC3F.js";
 import "./chunk-PQK3GKZH.js";
-import "./chunk-4BYGTEJT.js";
-import "./chunk-JMWJ7V6Y.js";
+import "./chunk-XBJQKCLU.js";
+import "./chunk-JPHJBCZB.js";
 import "./chunk-776EQXDV.js";
 import "./chunk-PZ5AY32C.js";
 export {
   runLeanHookCommand
 };
-//# sourceMappingURL=lean-command-DA4ZNKG3.js.map
+//# sourceMappingURL=lean-command-C5YPFKHP.js.map

package/dist/{local-3NGNESNB.js → local-OTOY56B4.js}

@@ -2,12 +2,12 @@ import {
   LocalModelProvider,
   modelCacheDir,
   rerankLocally
-} from "./chunk-3XYQGFPB.js";
-import "./chunk-JMWJ7V6Y.js";
+} from "./chunk-U4HAJLIB.js";
+import "./chunk-JPHJBCZB.js";
 import "./chunk-PZ5AY32C.js";
 export {
   LocalModelProvider,
   modelCacheDir,
   rerankLocally
 };
-//# sourceMappingURL=local-3NGNESNB.js.map
+//# sourceMappingURL=local-OTOY56B4.js.map

package/dist/{main-BDVQAGAS.js → main-HFHLGID4.js}

@@ -10,7 +10,7 @@ import {
 import {
   renderLeanBlock,
   runLeanHookCommand
-} from "./chunk-ROQTT5ZO.js";
+} from "./chunk-N34XHWBA.js";
 import {
   runPreToolHookCommand
 } from "./chunk-WHZSVY2L.js";
@@ -45,12 +45,12 @@ import {
   startMcpServer,
   tracePath,
   watchRepository
-} from "./chunk-Z5QB367W.js";
+} from "./chunk-VAKBJV26.js";
 import "./chunk-IHQM2ZDV.js";
 import {
   SessionCapsuleLedger,
   estimateTokens
-} from "./chunk-CQPCHMBS.js";
+} from "./chunk-SYWJG5NA.js";
 import {
   buildArchitectureMap,
   generateGraphifyGraph,
@@ -77,24 +77,24 @@ import {
   pingDatabase,
   runDoctor,
   runMigrations
-} from "./chunk-JDZ3VK26.js";
+} from "./chunk-UN57FKQX.js";
 import {
   redactPersisted
 } from "./chunk-GDJH7XJC.js";
 import {
   computeFactHash,
   generateEntityId
-} from "./chunk-4BYGTEJT.js";
-import "./chunk-KYK6RPAJ.js";
+} from "./chunk-XBJQKCLU.js";
+import "./chunk-JFDRBPYF.js";
 import {
   modelCacheDir
-} from "./chunk-3XYQGFPB.js";
+} from "./chunk-U4HAJLIB.js";
 import {
   DEFAULT_CLOUD_EMBEDDING_MODELS,
   DEFAULT_RERANK_MODELS,
   LOCAL_EMBEDDING_MODEL,
   getModelDefinition
-} from "./chunk-JMWJ7V6Y.js";
+} from "./chunk-JPHJBCZB.js";
 import {
   DEFAULT_ZVEC_MODEL,
   buildZvecIndex,
@@ -106,6 +106,7 @@ import {
 import {
   PLAN_POLICIES,
   configPath,
+  defaultNamespaceForCwd,
   initConfigFile,
   initProjectScaffold,
   loadPlanPolicy,
@@ -633,9 +634,9 @@ function resolveCliEntrypoint() {
 }
 function buildEntry(options) {
   const env = {
-    CODEBUDDY_NAMESPACE: options.namespace
+    CODEBUDDY_NAMESPACE: options.namespace,
+    CODEBUDDY_PROJECT_ROOT: resolve2(options.projectRoot ?? process.cwd())
   };
-  if (options.projectRoot) env.CODEBUDDY_PROJECT_ROOT = options.projectRoot;
   if (options.hfToken) env.HF_TOKEN = options.hfToken;
   if (options.groqApiKey) env.GROQ_API_KEY = options.groqApiKey;
   if (options.databaseUrl) env.DATABASE_URL = options.databaseUrl;
@@ -3540,6 +3541,7 @@ async function pingOnce(databaseUrl) {
   }
 }
 async function runInitWizard() {
+  const projectRoot = process.cwd();
   const sigintHandler = () => bail();
   process.on("SIGINT", sigintHandler);
   try {
@@ -3562,7 +3564,7 @@ async function runInitWizard() {
     const namespace = unwrap(
       await text({
         message: "Default namespace",
-        initialValue: process.env.CODEBUDDY_NAMESPACE ?? "default",
+        initialValue: process.env.CODEBUDDY_NAMESPACE ?? defaultNamespaceForCwd(projectRoot),
         validate: (value) => value && /^[a-zA-Z0-9_.:-]+$/.test(value) ? void 0 : "Letters, digits, _ . : - only."
       })
     );
@@ -3638,6 +3640,7 @@ async function runInitWizard() {
       try {
         const install = await installClaudeEntry({
           namespace,
+          projectRoot,
           hfToken,
           databaseUrl
         });
@@ -3883,8 +3886,8 @@ import pc13 from "picocolors";
 
 // evals/fact-relevance.json
 var fact_relevance_default = {
-  version: 1,
-  note: "Task prompts vs durable project facts. Queries are paraphrased so most share few or no words with their fact; every unlisted (query, fact) pair is irrelevant.",
+  version: 2,
+  note: "Task prompts vs durable project facts. Queries are paraphrased so most share few or no words with their fact; every unlisted (query, fact) pair is irrelevant. Hard negatives share a subsystem with a relevant fact but do not answer its task. Queries alternate dev/test by position (even index = dev): floors are chosen on dev and reported on test.",
   facts: {
     cookie: "OAuth callback breaks unless the session cookie is SameSite=Lax; Strict drops it on the cross-site redirect.",
     pool: "Postgres connections go through one bounded pool of 10; never open ad-hoc clients in request handlers.",
@@ -3905,43 +3908,174 @@ var fact_relevance_default = {
     avatar: "Profile images are resized to 256px and stored as WebP in the media bucket.",
     flags: "Feature flags are read once at boot; changing one requires a restart of the worker.",
     pagination: "List endpoints use cursor pagination with an opaque base64url cursor of createdAt and id.",
-    i18n: "UI strings live in locales/en.json; never hard-code user-facing text in components."
+    i18n: "UI strings live in locales/en.json; never hard-code user-facing text in components.",
+    cookieDomain: "Session cookies are scoped to the app subdomain; never set them on the apex domain.",
+    poolMetrics: "Database pool metrics are exported to Prometheus under the codebuddy_pool_ prefix.",
+    releaseNotes: "Release notes are drafted from merged pull request titles by the changelog script.",
+    hookInstall: "Hooks are written to .claude/settings.json and merged idempotently with the user's own hooks.",
+    migrationNaming: "Migration files are numbered by drizzle-kit; never rename or edit a migration that has been applied.",
+    testsParallel: "Vitest runs test files in parallel, so every test must use its own temporary directory.",
+    logsStderr: "Server logs go to stderr through pino; never log full request bodies.",
+    retry5xx: "The HTTP client retries 5xx responses twice with jittered backoff.",
+    dateFormat: "Dates in the UI are formatted with the user's locale through Intl.DateTimeFormat.",
+    flagNames: "Feature flag names use kebab-case and are declared in flags.yaml."
   },
   queries: [
     {
       q: "users get bounced back to sign-in forever after the identity provider returns",
       relevant: ["cookie"]
     },
-    { q: "handlers are exhausting connections to the database under load", relevant: ["pool"] },
-    { q: "cut a new version and push the package to the registry", relevant: ["release"] },
-    { q: "the injected block sometimes never shows up in the chat", relevant: ["hook"] },
-    { q: "fresh database setup fails on the embedding column", relevant: ["migration"] },
-    { q: "an API key ended up in a saved note", relevant: ["redaction"] },
-    { q: "context is sent again after I switch tasks and come back", relevant: ["ledger"] },
-    { q: "semantic code search crashes and the agent gets no files", relevant: ["zvec"] },
-    { q: "why do the db tests not run on my machine", relevant: ["tests"] },
-    { q: "module not found when importing a sibling ts file at runtime", relevant: ["esm"] },
-    { q: "Claude Desktop says Unexpected token in the server response", relevant: ["stdout"] },
-    { q: "the summarizer quietly stopped calling Groq in the afternoon", relevant: ["budget"] },
-    { q: "my uncommitted changes vanished after another session ran", relevant: ["worktree"] },
+    {
+      q: "handlers are exhausting connections to the database under load",
+      relevant: ["pool"]
+    },
+    {
+      q: "cut a new version and push the package to the registry",
+      relevant: ["release"]
+    },
+    {
+      q: "the injected block sometimes never shows up in the chat",
+      relevant: ["hook"]
+    },
+    {
+      q: "fresh database setup fails on the embedding column",
+      relevant: ["migration"]
+    },
+    {
+      q: "an API key ended up in a saved note",
+      relevant: ["redaction"]
+    },
+    {
+      q: "context is sent again after I switch tasks and come back",
+      relevant: ["ledger"]
+    },
+    {
+      q: "semantic code search crashes and the agent gets no files",
+      relevant: ["zvec"]
+    },
+    {
+      q: "why do the db tests not run on my machine",
+      relevant: ["tests"]
+    },
+    {
+      q: "module not found when importing a sibling ts file at runtime",
+      relevant: ["esm"]
+    },
+    {
+      q: "Claude Desktop says Unexpected token in the server response",
+      relevant: ["stdout"]
+    },
+    {
+      q: "the summarizer quietly stopped calling Groq in the afternoon",
+      relevant: ["budget"]
+    },
+    {
+      q: "my uncommitted changes vanished after another session ran",
+      relevant: ["worktree"]
+    },
     {
       q: "clients see too many requests errors during the import job",
       relevant: ["rateLimit"]
     },
-    { q: "meeting times show up one hour off for users in Berlin", relevant: ["timezone"] },
-    { q: "uploaded profile pictures look blurry", relevant: ["avatar"] },
-    { q: "I toggled the new checkout switch but nothing changed", relevant: ["flags"] },
-    { q: "infinite scroll repeats items on the second page", relevant: ["pagination"] },
-    { q: "translate the onboarding screen to German", relevant: ["i18n"] },
-    { q: "format the code before I commit", relevant: ["biome"] },
-    { q: "make the sidebar collapsible", relevant: [] },
-    { q: "add dark mode to the dashboard", relevant: [] },
-    { q: "write a haiku about the project", relevant: [] },
-    { q: "rename the button label on the settings page", relevant: [] }
+    {
+      q: "meeting times show up one hour off for users in Berlin",
+      relevant: ["timezone"]
+    },
+    {
+      q: "uploaded profile pictures look blurry",
+      relevant: ["avatar"]
+    },
+    {
+      q: "I toggled the new checkout switch but nothing changed",
+      relevant: ["flags"]
+    },
+    {
+      q: "infinite scroll repeats items on the second page",
+      relevant: ["pagination"]
+    },
+    {
+      q: "translate the onboarding screen to German",
+      relevant: ["i18n"]
+    },
+    {
+      q: "format the code before I commit",
+      relevant: ["biome"]
+    },
+    {
+      q: "make the sidebar collapsible",
+      relevant: []
+    },
+    {
+      q: "add dark mode to the dashboard",
+      relevant: []
+    },
+    {
+      q: "write a haiku about the project",
+      relevant: []
+    },
+    {
+      q: "rename the button label on the settings page",
+      relevant: []
+    },
+    {
+      q: "where do I declare a new feature flag",
+      relevant: ["flagNames"]
+    },
+    {
+      q: "show dates in the user's own language and order",
+      relevant: ["dateFormat"]
+    },
+    {
+      q: "can I tidy up the name of the 0004 migration file",
+      relevant: ["migrationNaming"]
+    },
+    {
+      q: "tests interfere with each other when the whole suite runs",
+      relevant: ["testsParallel"]
+    },
+    {
+      q: "where can I see how busy the connection pool is",
+      relevant: ["poolMetrics"]
+    },
+    {
+      q: "how is the text for the changelog produced",
+      relevant: ["releaseNotes"]
+    },
+    {
+      q: "the login cookie shows up on example.com instead of app.example.com",
+      relevant: ["cookieDomain"]
+    },
+    {
+      q: "is it fine to print the incoming payload for debugging",
+      relevant: ["logsStderr"]
+    },
+    {
+      q: "our hook setup appears twice in the Claude settings file",
+      relevant: ["hookInstall"]
+    },
+    {
+      q: "flaky upstream 502s should not fail the request on the first try",
+      relevant: ["retry5xx"]
+    },
+    {
+      q: "rewrite the marketing copy on the landing page",
+      relevant: []
+    },
+    {
+      q: "upgrade React to the next major version",
+      relevant: []
+    },
+    {
+      q: "add a CSV export button to the reports table",
+      relevant: []
+    }
   ]
 };
 
 // src/providers/calibration.ts
+function splitOf(index) {
+  return index % 2 === 0 ? "dev" : "test";
+}
 async function calibrateEmbedder(embed) {
   const data = fact_relevance_default;
   const factIds = Object.keys(data.facts);
@@ -3950,47 +4084,59 @@ async function calibrateEmbedder(embed) {
   const dot = (a, b) => a.reduce((sum, value, index) => sum + value * (b[index] ?? 0), 0);
   const scores = queryVectors.map((query) => factVectors.map((fact) => dot(query, fact)));
   const relevant = (qi, fi) => data.queries[qi]?.relevant.includes(factIds[fi] ?? "") ?? false;
+  const answerable = (qi) => (data.queries[qi]?.relevant.length ?? 0) > 0;
+  const rows = (split) => scores.flatMap((row2, qi) => splitOf(qi) === split ? [{ row: row2, qi }] : []);
   const positives = [];
   const negatives = [];
-  for (const [qi, row2] of scores.entries()) {
-    for (const [fi, score] of row2.entries()) (relevant(qi, fi) ? positives : negatives).push(score);
-  }
-  const auc = positives.reduce((wins, p) => wins + negatives.filter((n) => p > n).length, 0) / (positives.length * negatives.length);
   let hits = 0;
   let reciprocal = 0;
-  let answerable = 0;
-  for (const [qi, row2] of scores.entries()) {
-    if (!data.queries[qi]?.relevant.length) continue;
-    answerable += 1;
+  let answered = 0;
+  for (const { row: row2, qi } of rows("test")) {
+    for (const [fi, score] of row2.entries()) (relevant(qi, fi) ? positives : negatives).push(score);
+    if (!answerable(qi)) continue;
+    answered += 1;
     const order = row2.map((score, fi) => ({ score, fi })).sort((a, b) => b.score - a.score);
     const rank = order.findIndex(({ fi }) => relevant(qi, fi)) + 1;
     if (rank === 1) hits += 1;
     reciprocal += 1 / rank;
   }
-  const at = (floor) => {
+  const auc = positives.reduce((wins, p) => wins + negatives.filter((n) => p > n).length, 0) / (positives.length * negatives.length);
+  const at = (floor, split = "test") => {
     let tp = 0;
     let fp = 0;
+    let relevantPairs = 0;
     let abstained = 0;
     let unrelated = 0;
-    for (const [qi, row2] of scores.entries()) {
+    for (const { row: row2, qi } of rows(split)) {
+      relevantPairs += data.queries[qi]?.relevant.length ?? 0;
       const admitted = row2.flatMap((score, fi) => score >= floor ? [fi] : []);
       for (const fi of admitted) {
         if (relevant(qi, fi)) tp += 1;
         else fp += 1;
       }
-      if (!data.queries[qi]?.relevant.length) {
+      if (!answerable(qi)) {
         unrelated += 1;
         if (admitted.length === 0) abstained += 1;
       }
     }
     const precision = tp + fp === 0 ? 1 : tp / (tp + fp);
-    const recall = tp / positives.length;
+    const recall = relevantPairs === 0 ? 0 : tp / relevantPairs;
     const f05 = precision + recall === 0 ? 0 : 1.25 * precision * recall / (0.25 * precision + recall);
-    return { floor, precision, recall, f05, abstain: abstained / unrelated };
+    return { floor, precision, recall, f05, abstain: unrelated === 0 ? 1 : abstained / unrelated };
   };
-  const sweep = Array.from({ length: 96 }, (_, i) => at(Math.round((0.05 + i * 0.01) * 100) / 100));
+  const sweep = Array.from(
+    { length: 96 },
+    (_, i) => at(Math.round((0.05 + i * 0.01) * 100) / 100, "dev")
+  );
   const best = sweep.reduce((left, right) => right.f05 > left.f05 ? right : left);
-  return { auc, recallAt1: hits / answerable, mrr: reciprocal / answerable, best, at };
+  return {
+    auc,
+    recallAt1: answered === 0 ? 0 : hits / answered,
+    mrr: answered === 0 ? 0 : reciprocal / answered,
+    best,
+    test: at(best.floor, "test"),
+    at
+  };
 }
 
 // src/cli/commands/models.ts
@@ -4225,7 +4371,7 @@ before keeping one on. A reranker never admits a fact, it only orders them.`
   });
   models.command("pull").description("Download the local models now instead of on first use.").action(async () => {
     await runSafely2(async () => {
-      const { LocalModelProvider, rerankLocally } = await import("./local-3NGNESNB.js");
+      const { LocalModelProvider, rerankLocally } = await import("./local-OTOY56B4.js");
       const file = await readConfigFile();
       let started = Date.now();
       await withSpinner(
@@ -4261,13 +4407,13 @@ before keeping one on. A reranker never admits a fact, it only orders them.`
       console.log(
         row(
           "ranking",
-          `AUC ${result.auc.toFixed(3)} \xB7 right fact first ${pct(result.recallAt1)}`
+          `AUC ${result.auc.toFixed(3)} \xB7 right fact first ${pct(result.recallAt1)} ${pc13.dim("(held-out tasks)")}`
         )
       );
       console.log(
         row(
           "floor",
-          `${result.best.floor}: precision ${pct(result.best.precision)}, recall ${pct(result.best.recall)}, silent on ${pct(result.best.abstain)} of unrelated tasks`
+          `${result.best.floor}: precision ${pct(result.test.precision)}, recall ${pct(result.test.recall)}, silent on ${pct(result.test.abstain)} of unrelated tasks ${pc13.dim("(chosen on dev, measured on test)")}`
         )
       );
       if (opts.dryRun) return;
@@ -5591,4 +5737,4 @@ function printDbDoctor(report) {
 export {
   createCli
 };
-//# sourceMappingURL=main-BDVQAGAS.js.map
+//# sourceMappingURL=main-HFHLGID4.js.map