agentlas 1.0.28 → 1.0.29

package/CHANGELOG.md

@@ -1,5 +1,38 @@
 # Changelog
 
+## 1.0.29 — 2026-08-05
+
+Terminal-wide audit release. Every command was executed for real; what follows fixes what that audit found.
+
+- Native federated staffing: `hep-network` / `hep-local` / `hep-cloud` / `hep-hub` now run this
+  terminal's own workforce loop with the local Agentlas-OS core federating the declared source scope.
+  (They previously passed through to an external CLI stub that always answered exit 3.)
+  Honest stop with install guidance when the local core is missing — no silent fallback to the public Hub.
+- `workforce` now declares its sourceScope ("hub") explicitly instead of relying on the server default.
+- `doctor` distinguishes installed from signed-in: local sign-in evidence per runtime, a warning
+  (not "all clear") when the active runtime has none.
+- New `roles` command: view and set orchestrator/worker model roles from the terminal
+  (`roles set <role> <runtime> [--model id] [--effort lv]`, `roles set worker --inherit`).
+  REPL `/model`·`/runtime` now say they are session-scoped and point to `agentlas roles`.
+- Setup wizard prints install/sign-in guidance when the chosen runtime is missing or unauthenticated.
+- `creds list` (names and stores only — values never printed), `mcp list`, Korean `help` body.
+- The graph command group ships in the npm package for the first time (it landed after 1.0.28 was published).
+- New gates: user-scenarios (93 real CLI invocations), local-core transport wire contract,
+  doctor auth-evidence, roles round-trip, onboarding guidance, EN/KO help sync.
+- `agentlas <command> --help` reaches the command's own help. It printed a two-line stub
+  scraped from the help table, so `graph --help` never showed the eight lines `graph help` has —
+  and that is the spelling people try first. (The gate had pinned the stub as correct; its
+  contract now asks whether real help was shown.)
+- `graph install --name "<new name>"` works. The documented flag had never worked: its value was
+  appended to the file path, so the install died with "no such file".
+- `graph show` stops indenting a straight chain deeper at every step — fourteen steps meant
+  twenty-eight columns. Depth now marks real branches (a fork, a failure exit).
+- Building a graph follows your language setting. One Korean character anywhere in the request
+  forced the interview to Korean while `graph show`, the list, and errors stayed English.
+- Graph steps written as code declare the pip packages they import, and a verification step is
+  now required whenever a step that changes something outside sends out a value an earlier step
+  computed — an unattended run must not ship an empty or invented result.
+
 ## 1.0.26 — 2026-08-03
 
 - **`agentlas list` stops letting the terminal cut its own output.** Slugs were

package/README.md

@@ -94,10 +94,13 @@ agentlas variant resolve --base-release <id> # Preview local variant compatibili
 ```sh
 agentlas storm "<goal>"          # Goal + UltraCode harness: plan -> assign -> execute -> verify [--research]
 agentlas swarm "<goal>"          # Emergent multi-agent swarm [--parallel N]
-agentlas workforce "<prompt>"    # Route through Agent Workforce Ontology
+agentlas workforce "<prompt>"    # Route through Agent Workforce Ontology (public Hub menu)
 agentlas network "<prompt>"      # Alias for workforce
 agentlas taskforce "<prompt>"    # Alias for workforce
-agentlas legacy-network "<prompt>" # Legacy Hephaestus solver (explicit invocation only)
+agentlas hep-network "<prompt>"  # Staff across Local + owner Cloud + public Hub (local Core federation)
+agentlas hep-local "<prompt>"    # Same staffing, registered Local agents only
+agentlas hep-cloud "<prompt>"    # Same staffing, owner Agent Cloud only
+agentlas hep-hub "<prompt>"      # Same staffing, public Hub only
 agentlas call "a,b" "<context>"  # Call explicitly named Hub or Cloud agents
 agentlas browser [...]           # Real browser hardpoint launcher
 agentlas route "<prompt>" [--json] # Preview routing decisions without execution

package/engine/agentlas-i18n.cjs

@@ -74,7 +74,6 @@ const STRINGS = {
     "effortUsage": "usage: /effort low|medium|high|max|off|auto",
     "permSet": "permission → %s",
     "permUsage": "usage: /permission read|write|full",
-    "legacyNetworkUsage": "usage: /legacy-network <request>",
     "stormUsage": "usage: /storm <goal> [--research]",
     "buildUsage": "usage: /build <agent-or-team request>",
     "routeUsage": "usage: /route <request> [--json]",
@@ -190,7 +189,6 @@ const STRINGS = {
     "help.search": "discover agents in the Hub",
     "help.install": "install an agent from the Hub by slug",
     "help.network": "let the host LLM staff and execute an ontology-grounded exact-release task force",
-    "help.legacyNetwork": "run the explicit compatibility-only Hephaestus network route",
     "help.browser": "real browser execution hardpoint",
     "help.connect": "wire Telegram / platforms to an agent team",
     "help.swarm": "fan out an emergent agent swarm on a goal",
@@ -251,6 +249,8 @@ const STRINGS = {
     "wiz.runtimeAuto": "Use the app's active runtime",
     "wiz.runtimeInstalled": "installed",
     "wiz.runtimeMissing": "not installed",
+    "wiz.runtimeInstallHint": "This runtime is not installed yet. Install it first: %s",
+    "wiz.runtimeLoginHint": "No local sign-in evidence for this runtime. Sign in before your first run (run: %s, then follow its login flow).",
     "wiz.permQ": "How much should agents be allowed to do by default?",
     "wiz.permRead": "read — inspect only; workspace writes blocked",
     "wiz.permWrite": "write — workspace + runtime temp writes; external MCP off (recommended)",
@@ -323,7 +323,6 @@ const STRINGS = {
     "effortUsage": "사용법: /effort low|medium|high|max|off|auto",
     "permSet": "권한 → %s",
     "permUsage": "사용법: /permission read|write|full",
-    "legacyNetworkUsage": "사용법: /legacy-network <요청>",
     "stormUsage": "사용법: /storm <목표> [--research]",
     "buildUsage": "사용법: /build <만들 에이전트·팀 설명>",
     "routeUsage": "사용법: /route <요청> [--json]",
@@ -438,7 +437,6 @@ const STRINGS = {
     "help.search": "Hub에서 에이전트 발견",
     "help.install": "식별자로 Hub 에이전트 설치",
     "help.network": "상위 LLM이 온톨로지에서 정확한 릴리스 TF를 선발·실행",
-    "help.legacyNetwork": "호환성 전용 Hephaestus 네트워크 경로를 명시적으로 실행",
     "help.browser": "실제 브라우저 실행 하드포인트",
     "help.connect": "Telegram/플랫폼을 에이전트 팀에 연결",
     "help.swarm": "목표에 창발형 에이전트 스웜 전개",
@@ -497,6 +495,8 @@ const STRINGS = {
     "wiz.runtimeAuto": "앱의 활성 런타임 사용",
     "wiz.runtimeInstalled": "설치됨",
     "wiz.runtimeMissing": "미설치",
+    "wiz.runtimeInstallHint": "이 런타임은 아직 설치되어 있지 않습니다. 먼저 설치하세요: %s",
+    "wiz.runtimeLoginHint": "이 런타임의 로그인 흔적이 없습니다. 첫 실행 전에 로그인하세요 (%s 실행 후 로그인 절차 진행).",
     "wiz.permQ": "에이전트가 기본적으로 어디까지 할 수 있게 할까요?",
     "wiz.permRead": "read — 조회만; 작업 공간 쓰기 차단",
     "wiz.permWrite": "write — 작업 공간·런타임 임시 경로 쓰기; 외부 MCP 꺼짐 (권장)",

package/engine/agentlas-input.cjs

@@ -300,7 +300,6 @@ const SLASH_COMMAND_META = [
   { command: "/search", description: "Discover agents in the Hub", category: "Hub", usage: "/search <what you need>", detail: "Search the Agentlas Hub + local for an agent that fits the task (hep-search)." },
   { command: "/install", description: "Install an agent from the Hub by slug", category: "Hub", usage: "/install <slug>", detail: "Install a marketplace agent into this terminal." },
   { command: "/network", description: "Staff and execute an ontology-grounded task force", category: "Engine", usage: "/network <request> [--benchmark]", detail: "The active top host LLM creates the work order, searches the Hub workforce ontology, selects exact releases, and executes a receipt-backed task force.", aliases: ["/taskforce", "/workforce"] },
-  { command: "/legacy-network", description: "Run the compatibility Hephaestus network route", category: "Engine", usage: "/legacy-network <request>", detail: "Explicit compatibility escape hatch; never used as a fallback by /network." },
   { command: "/browser", description: "Real browser execution hardpoint", category: "Engine", usage: "/browser [sub]", detail: "Runs the Agentlas browser hardpoint (hep-browser)." },
   { command: "/connect", description: "Wire Telegram / platforms to an agent team", category: "Hub", usage: "/connect", detail: "Runs Hephaestus hep-connect for platform integration." },
   { command: "/doctor", description: "Check runtimes and local data", category: "Health", usage: "/doctor", detail: "Run local checks for runtimes, data, credentials, and setup." },
@@ -351,7 +350,6 @@ const HELP_KEY_BY_COMMAND = {
   "/research": "help.research",
   "/search": "help.search",
   "/network": "help.network",
-  "/legacy-network": "help.legacyNetwork",
   "/browser": "help.browser",
   "/connect": "help.connect",
   "/doctor": "help.doctor",

package/engine/agentlas-onboard.cjs

@@ -142,6 +142,26 @@ async function runOnboard({ ui, rl, helpers, persist }) {
   rtOpts.forEach((o, i) => optionLine(i + 1, o.label, i === 0));
   const ri = await pickNum(rtOpts.length);
   const runtime = rtOpts[ri - 1].value;
+  // missing인 런타임을 골라도 저장은 한다(미리 설정) — 하지만 조용히 저장하면
+  // 첫 실행 실패가 배신이 된다(2026-08-05 감사 결함 A). 설치 명령을 그 자리에서
+  // 알려주고, 로그인 흔적이 없으면 로그인 절차까지 말한다.
+  if (runtime !== "auto") {
+    const installHint = {
+      "claude-code": "npm i -g @anthropic-ai/claude-code",
+      codex: "npm i -g @openai/codex",
+      gemini: "npm i -g @google/gemini-cli",
+    }[runtime];
+    if (!H.which(H.RUNTIME_BIN[runtime])) {
+      if (installHint) printIndented(ui.t("wiz.runtimeInstallHint", installHint), c.dim);
+    } else {
+      try {
+        const { runtimeAuthEvidence } = require("./runtimes/auth-evidence.cjs");
+        if (runtimeAuthEvidence(runtime).status === "none") {
+          printIndented(ui.t("wiz.runtimeLoginHint", H.RUNTIME_BIN[runtime]), c.dim);
+        }
+      } catch { /* evidence unavailable — say nothing rather than guess */ }
+    }
+  }
 
   // Step 3 — default permission
   ui.line("");

package/engine/agentlas-workforce.cjs

@@ -1097,10 +1097,15 @@ function validateWorkOrder(value) {
 function validateCandidateSet(value, workOrder, now = new Date(), options = {}) {
   const set = assertObject(value, "candidateSet");
   assertNoForbiddenFitSignals(set);
-  assertExactKeys(set, [
+  // projection은 로컬 Core(연합) 응답에만 있는 메뉴 투영 메타데이터다(실측
+  // 2026-08-05, reference-first: fullDossier=false). 원격 서버는 보내지 않는다.
+  // 없애고 되돌려 보내면 Core 쪽 다이제스트 대조가 위험하므로 선택 키로 허용한다.
+  const exactKeys = [
     "schemaVersion", "selectionSessionId", "workOrderId", "ontologyVersion",
     "candidateSetDigest", "decisionOwner", "historyInfluence", "slots", "issuedAt", "expiresAt",
-  ], "candidateSet", "candidate_set_invalid");
+  ];
+  if (Object.prototype.hasOwnProperty.call(set, "projection")) exactKeys.push("projection");
+  assertExactKeys(set, exactKeys, "candidateSet", "candidate_set_invalid");
   if (set.schemaVersion !== "agentlas.workforce-candidate-set.v1") fail("candidate_set_invalid", "unsupported candidate set schema");
   assertId(set.selectionSessionId, "candidateSet.selectionSessionId");
   if (set.workOrderId !== workOrder.workOrderId) fail("candidate_set_invalid", "candidate set workOrderId mismatch");
@@ -1126,11 +1131,15 @@ function validateCandidateSet(value, workOrder, now = new Date(), options = {})
     const releases = new Set();
     for (const candidate of assertArray(slotResult.candidates, `candidateSet.${slotId}.candidates`, 100)) {
       assertObject(candidate, "candidate");
-      assertExactKeys(candidate, [
+      // missingMandatory는 로컬 Core(연합) 응답에만 있는 미충족 필수 표식이다
+      // (실측 2026-08-05, fullDossier=true에도 동봉). 원격 서버는 보내지 않는다.
+      const candidateKeys = [
         "agentDefinitionId", "agentReleaseId", "releaseVersion", "packageHash", "contentDigest",
         "entityKind", "name", "communities", "fitEvidence", "qualificationEvidence", "optionalGaps",
         "semanticSnapshot", "operational",
-      ], "candidate", "candidate_set_invalid");
+      ];
+      if (Object.prototype.hasOwnProperty.call(candidate, "missingMandatory")) candidateKeys.push("missingMandatory");
+      assertExactKeys(candidate, candidateKeys, "candidate", "candidate_set_invalid");
       assertId(candidate.agentDefinitionId, "candidate.agentDefinitionId");
       const releaseId = assertId(candidate.agentReleaseId, "candidate.agentReleaseId");
       if (releases.has(releaseId)) fail("candidate_set_invalid", `duplicate release ${releaseId} in ${slotId}`);
@@ -1151,10 +1160,19 @@ function validateCandidateSet(value, workOrder, now = new Date(), options = {})
       if (typeof operational.callable !== "boolean" || typeof operational.installable !== "boolean") fail("candidate_set_invalid", "candidate operational flags are invalid");
       assertIds(operational.unavailableReasons || [], "candidate.operational.unavailableReasons");
       const semantic = assertObject(candidate.semanticSnapshot, "candidate.semanticSnapshot");
-      assertExactKeys(semantic, [
+      // knowledge·modalities는 로컬 Core 스냅샷에만 있는 확장 어휘다(실측 2026-08-05).
+      // 원격 서버는 보내지 않는다 — missingMandatory·projection과 같은 규칙으로
+      // "있으면 검증하고 허용", 원격 계약의 exact-keys는 그대로 둔다.
+      const semanticKeys = [
         "summaries", "roles", "skills", "toolCapabilities", "consumes", "produces",
         "authorities", "runtimes", "languages",
-      ], "candidate.semanticSnapshot", "candidate_set_invalid");
+      ];
+      for (const optional of ["knowledge", "modalities"]) {
+        if (Object.prototype.hasOwnProperty.call(semantic, optional)) semanticKeys.push(optional);
+      }
+      assertExactKeys(semantic, semanticKeys, "candidate.semanticSnapshot", "candidate_set_invalid");
+      if (semantic.knowledge !== undefined) assertIds(semantic.knowledge, "candidate.semanticSnapshot.knowledge");
+      if (semantic.modalities !== undefined) assertStrings(semantic.modalities, "candidate.semanticSnapshot.modalities");
       assertStrings(semantic.summaries, "candidate.semanticSnapshot.summaries");
       assertIds(semantic.roles, "candidate.semanticSnapshot.roles");
       assertLeveledConcepts(semantic.skills, "candidate.semanticSnapshot.skills");
@@ -2393,6 +2411,16 @@ function create(deps = {}) {
 
   async function workforceRun(db, rawTask, ctx = {}) {
     const task = assertString(rawTask, "task", 20_000);
+    // 이 표면이 보는 후보 메뉴의 소스 스코프. 원격 MCP 경로의 정직한 기본은 "hub"
+    // (터미널 workforce = 공개 Hub 메뉴 스태핑). 다른 값은 로컬 Core 전송을 가진
+    // 호출자만 넘길 수 있다 — 여기서 조용히 넓히지 않는다.
+    const sourceScope = (() => {
+      const value = ctx.sourceScope === undefined ? "hub" : ctx.sourceScope;
+      if (!["network", "local", "cloud", "hub"].includes(value)) {
+        fail("source_scope_invalid", `sourceScope must be network|local|cloud|hub, got: ${String(value)}`);
+      }
+      return value;
+    })();
     const ui = ctx.ui || newUi();
     const runtime = ctx.runtime || D.resolveRuntime(db, ctx.runtimeOverride);
     const cwd = ctx.cwd || (typeof D.projectCwd === "function" ? D.projectCwd() : process.cwd());
@@ -2744,7 +2772,9 @@ function create(deps = {}) {
     };
 
     const supersedeCandidateSearch = (workOrder, refinementNumber, triggerKind) => {
-      const requestDigest = sha256({ workOrder });
+      // hubStage가 저장한 requestDigest와 같은 인자 모양이어야 행을 찾는다 —
+      // search 인자에 sourceScope가 실리므로(2026-08-05) 여기서도 함께 계산한다.
+      const requestDigest = sha256({ workOrder, sourceScope });
       for (const row of receipt.hubTools) {
         if (row.tool !== "workforce.search_candidates" || row.requestDigest !== requestDigest || row.authoritativeChain !== true) continue;
         row.authoritativeChain = false;
@@ -3065,7 +3095,10 @@ function create(deps = {}) {
 
       let refinementsUsed = 0;
       const searchCurrentWorkOrder = async () => {
-        const candidateRaw = await hubStage("workforce.search_candidates", { workOrder });
+        // sourceScope는 MCP 스키마상 required다. 예전에는 싣지 않아 서버 기본값
+        // ("hub")에 의존했다 — 기본값이 바뀌면 이 표면의 실제 스코프가 조용히
+        // 넓어지거나 좁아진다. 이 표면이 보는 메뉴를 스스로 선언한다.
+        const candidateRaw = await hubStage("workforce.search_candidates", { workOrder, sourceScope });
         candidateSet = validateCandidateSet(
           candidateRaw,
           workOrder,

package/engine/agentlas.cjs

@@ -107,7 +107,13 @@ function main() {
   if (helpRequested && helpCommand) {
     const ctx = buildCtx();
     const command = commands.resolveCommandName(helpCommand);
-    const code = require("./commands/help.cjs").runForCommand(ctx, command);
+    // ★명령이 자기 도움말을 갖고 있으면 그것을 보여준다.
+    //   예전에는 무조건 표 한 줄을 긁어(`runForCommand`) "Usage: agentlas graph [options]"
+    //   두 줄만 나왔다 — `graph help`에는 8줄짜리 제대로 된 안내가 있는데도 `--help`로는
+    //   영원히 닿지 못했다(사용자가 가장 먼저 치는 것이 `--help`다).
+    const code = commands.SELF_HELP_COMMANDS.has(command)
+      ? commands.dispatch(ctx, [command, "help"])
+      : require("./commands/help.cjs").runForCommand(ctx, command);
     process.exit(typeof code === "number" ? code : 0);
   }
   // 옵션 정규화: -h/--help/-V/--version 은 하위 명령으로 변환

package/engine/commands/billing.cjs

@@ -25,6 +25,9 @@ function usage(ko) {
     ko
       ? "  구독 계좌(A)와 렌트수익 계좌(B) 잔액을 표시합니다."
       : "  Shows the subscription account (A) and rental-earnings account (B) balances.",
+    ko
+      ? "  크레딧은 Hub 에이전트 호출(공개 에이전트 3·팀 10, 활성 리스는 0)에 쓰입니다."
+      : "  Credits pay for Hub agent calls (public agent 3 · team 10; active leases cost 0).",
     ko
       ? "  참고: 렌트수익(B) → 구독(A) 전송은 Agentlas Desktop 에서만 가능합니다 (터미널 전송 명령 없음)."
       : "  Note: earnings (B) → subscription (A) transfer is Desktop-only (no transfer command in the terminal).",

package/engine/commands/creds.cjs

@@ -134,12 +134,60 @@ async function cmdCredsFile(ctx, args) {
   return 0;
 }
 
+/** KEY=value 파일에서 KEY 이름만 뽑는다 — 값은 절대 읽어 돌려주지 않는다. */
+function envKeyNames(filePath) {
+  try {
+    return fs.readFileSync(filePath, "utf8")
+      .split(/\r?\n/)
+      .map((line) => line.match(/^\s*(?:export\s+)?([A-Za-z_][A-Za-z0-9_]*)\s*=/))
+      .filter(Boolean)
+      .map((m) => m[1]);
+  } catch {
+    return null; // 파일 없음/읽기 불가 — 목록에서 침묵이 아니라 부재로 표시
+  }
+}
+
+/*
+ * creds list — 어떤 자격증명 "이름"이 어디에 저장돼 있는지.
+ * 배경(2026-08-05 감사, 결함 H): save/file만 있어 저장한 것을 확인할 방법이
+ * 없었다. 계약은 상단과 동일 — 값은 어떤 경로로도 출력하지 않는다. 키 이름과
+ * 저장 위치까지만. (키체인 vault는 standalone에서 keytar를 부르면 무한 대기하는
+ * v1 실사고가 있어 조회조차 하지 않는다 — 부재가 아니라 "미조회"로 표시.)
+ */
+function cmdCredsList(ctx) {
+  const ko = ctx.lang !== "en";
+  const project = activeProjectPath(ctx.db());
+  const stores = [];
+  if (project) stores.push({ label: (ko ? "프로젝트 .env — " : "project .env — ") + path.join(project, ".env"), keys: envKeyNames(path.join(project, ".env")) });
+  stores.push({ label: (ko ? "전역 credentials.env — " : "global credentials.env — ") + path.join(userDataDir(), "credentials.env"), keys: envKeyNames(path.join(userDataDir(), "credentials.env")) });
+  stores.push({ label: "~/.agentlas/credentials.env", keys: envKeyNames(path.join(os.homedir(), ".agentlas", "credentials.env")) });
+
+  let any = false;
+  for (const store of stores) {
+    if (store.keys === null) continue; // 파일 자체가 없으면 줄을 만들지 않는다
+    any = true;
+    ctx.out(ctx.ui.bold(store.label));
+    if (!store.keys.length) ctx.out(ctx.ui.dim(ko ? "  (비어 있음)" : "  (empty)"));
+    for (const key of store.keys) ctx.out(`  ${key}${ctx.ui.dim("=•••")}`);
+  }
+  if (isElectronRuntime()) {
+    ctx.out(ctx.ui.dim(ko ? "키체인 vault: 데스크탑에서 확인" : "keychain vault: inspect in Desktop"));
+  } else {
+    ctx.out(ctx.ui.dim(ko
+      ? "키체인 vault는 standalone 터미널에서 조회하지 않습니다(서명 없는 keytar 호출은 키체인에 막혀 멈춥니다)."
+      : "The keychain vault is not queried from the standalone terminal (unsigned keytar calls hang on the OS keychain)."));
+  }
+  if (!any) ctx.out(ko ? "저장된 자격증명 파일이 없습니다. agentlas creds save 로 추가하세요." : "No credential files yet. Add one with: agentlas creds save");
+  return 0;
+}
+
 async function run(ctx, args) {
   const fail = (msg) => { ctx.err(msg); return 1; };
   const sub = args[0];
   if (sub === "file") return cmdCredsFile(ctx, args.slice(1));
+  if (sub === "list" || sub === "ls") return cmdCredsList(ctx);
   if (sub !== "save") {
-    return fail('usage: agentlas creds save --provider <name> --key <ENV_NAME> --value <value> [--project <path>] OR agentlas creds file --source <path> [--env <ENV_NAME>]');
+    return fail('usage: agentlas creds save --provider <name> --key <ENV_NAME> --value <value> [--project <path>] OR agentlas creds file --source <path> [--env <ENV_NAME>] OR agentlas creds list');
   }
   const f = parseCredFlags(args.slice(1));
   const key = typeof f.key === "string" ? f.key.trim() : "";

package/engine/commands/doctor.cjs

@@ -9,6 +9,7 @@ const fs = require("node:fs");
 const path = require("node:path");
 const { dbPath, userDataDir } = require("../core/paths.cjs");
 const { listAvailableCliRuntimes, activeRuntimeRow } = require("../runtimes/detect.cjs");
+const { runtimeAuthEvidence } = require("../runtimes/auth-evidence.cjs");
 const { resolvedModelRole } = require("../runtimes/roles.cjs");
 
 function roleDetail(selection, role, en) {
@@ -26,8 +27,10 @@ function roleDetail(selection, role, en) {
 function run(ctx) {
   const en = ctx.lang === "en";
   let failures = 0;
+  let warnings = 0;
   const ok = (label, detail) => ctx.out(`  ${ctx.ui.green("✓")} ${label}${detail ? ctx.ui.dim(" — " + detail) : ""}`);
   const bad = (label, detail) => { failures += 1; ctx.out(`  ${ctx.ui.red("✗")} ${label}${detail ? ctx.ui.dim(" — " + detail) : ""}`); };
+  const warn = (label, detail) => { warnings += 1; ctx.out(`  ${ctx.ui.yellow ? ctx.ui.yellow("!") : "!"} ${label}${detail ? ctx.ui.dim(" — " + detail) : ""}`); };
 
   // 1) 데이터
   const p = dbPath();
@@ -43,10 +46,28 @@ function run(ctx) {
     bad(en ? "database" : "데이터베이스", (en ? "missing: " : "없음: ") + p);
   }
 
-  // 2) 런타임
+  // 2) 런타임 — 설치 여부와 인증 흔적은 다른 축이다. which()만 보면 로그아웃
+  // 상태에서도 all clear가 나간다(2026-08-05 실측). 각 CLI가 로그인 시 남기는
+  // 로컬 산출물을 증거로 관측하고, 증거≠증명이므로 문구도 단정하지 않는다.
   const clis = listAvailableCliRuntimes();
   if (clis.length) {
-    ok(en ? "runtimes" : "런타임", clis.map((c) => `${c.kind}`).join(", "));
+    const evidences = clis.map((c) => ({ kind: c.kind, evidence: runtimeAuthEvidence(c.kind) }));
+    const annotated = evidences.map(({ kind, evidence }) => {
+      if (evidence.status === "none") return `${kind}(${en ? "no sign-in evidence" : "로그인 흔적 없음"})`;
+      return kind; // evidence 또는 unknown — 검사법이 없는 런타임을 미로그인으로 표시하지 않는다
+    });
+    ok(en ? "runtimes" : "런타임", annotated.join(", "));
+    // 신규 설치 상태: 검사 가능한 런타임 전부가 흔적 없음이면 첫 실행이 거의
+    // 확실히 실패한다 — 활성 런타임 행이 아직 없어도 여기서 경고한다.
+    const checkable = evidences.filter(({ evidence }) => evidence.status !== "unknown");
+    if (checkable.length && checkable.every(({ evidence }) => evidence.status === "none")) {
+      warn(
+        en ? "runtime sign-in" : "런타임 로그인",
+        en
+          ? "no detected runtime has local sign-in evidence — sign in to one before running (e.g. claude / codex login / gemini)"
+          : "감지된 어떤 런타임에도 로그인 흔적이 없습니다 — 실행 전에 하나는 로그인하세요 (예: claude / codex login / gemini)",
+      );
+    }
   } else {
     bad(en ? "runtimes" : "런타임", en
       ? "no agent CLI on PATH (claude / codex / gemini / kimi / grok / cursor-agent)"
@@ -57,7 +78,23 @@ function run(ctx) {
   try {
     const db = ctx.db();
     const active = activeRuntimeRow(db);
-    if (active) ok(en ? "active runtime" : "활성 런타임", `${active.kind}${active.model ? ` (${active.model})` : ""}`);
+    if (active) {
+      const detail = `${active.kind}${active.model ? ` (${active.model})` : ""}`;
+      // 활성 런타임은 모든 실행이 지나는 문이다 — 로그인 흔적이 없으면 all clear
+      // 가 아니라 경고다. 흔적 없음 = 미로그인 "가능성"이므로 단정하지 않는다.
+      const evidence = runtimeAuthEvidence(active.kind);
+      if (evidence.status === "none") {
+        const bin = { "claude-code": "claude", codex: "codex", gemini: "gemini" }[active.kind] || active.kind;
+        warn(
+          en ? "active runtime" : "활성 런타임",
+          en
+            ? `${detail} — no local sign-in evidence; runs may fail until you sign in (try: ${bin} login)`
+            : `${detail} — 로그인 흔적이 없습니다. 로그인 전에는 실행이 실패할 수 있습니다 (시도: ${bin} login)`,
+        );
+      } else {
+        ok(en ? "active runtime" : "활성 런타임", detail);
+      }
+    }
     const orchestrator = resolvedModelRole(db, "orchestrator");
     const worker = resolvedModelRole(db, "worker");
     if (orchestrator && worker) {
@@ -83,6 +120,12 @@ function run(ctx) {
     ctx.out(en ? `doctor: ${failures} problem(s) found` : `doctor: 문제 ${failures}건`);
     return 1;
   }
+  if (warnings) {
+    // 경고가 있으면 "이상 없음"이라고 말하지 않는다 — 그 문구가 첫 실행 실패를
+    // 배신으로 만든다. 경고는 실행을 막지 않으므로 exit 0.
+    ctx.out(en ? `doctor: ${warnings} warning(s) — runnable, but check the lines above` : `doctor: 경고 ${warnings}건 — 실행은 가능하나 위 항목을 확인하세요`);
+    return 0;
+  }
   ctx.out(en ? "doctor: all clear" : "doctor: 이상 없음");
   return 0;
 }