tickmarkr 2.2.1 → 2.4.0

package/dist/compile/ownership.js

@@ -22,14 +22,31 @@ function testSources(repoRoot) {
         return [];
     }
 }
-function namedSources(test, tasks) {
+function sourceFiles(repoRoot) {
+    const root = join(repoRoot, "src");
+    try {
+        return readdirSync(root, { recursive: true, encoding: "utf8" })
+            .filter((path) => /\.(?:[cm]?[jt]sx?)$/.test(path))
+            .map((path) => `src/${normalize(path)}`)
+            .sort();
+    }
+    catch {
+        return [];
+    }
+}
+// OBS-898: `allSources` is indexed once per ownership pass by the caller, never per test source.
+function namedSources(test, tasks, allSources) {
     const stem = basename(test).replace(/\.test\.ts$/, "");
     const matches = new Map();
     for (const task of tasks) {
-        for (const entry of task.files.map(normalize).filter((path) => path.startsWith("src/") && !/[*?{[]/.test(path))) {
-            const source = basename(entry, extname(entry));
-            if (stem === source || stem.startsWith(`${source}-`)) {
-                matches.set(`${task.id}:${entry}`, { taskId: task.id, source: entry });
+        for (const entry of task.files.map(normalize).filter((path) => path.startsWith("src/"))) {
+            const fromGlob = /[*?{[]/.test(entry);
+            const entries = fromGlob ? allSources.filter(filesGlob(entry)) : [entry];
+            for (const sourcePath of entries) {
+                const source = basename(sourcePath, extname(sourcePath));
+                if (stem === source || stem.startsWith(`${source}-`)) {
+                    matches.set(`${task.id}:${sourcePath}`, { taskId: task.id, source: sourcePath, fromGlob });
+                }
             }
         }
     }
@@ -148,19 +165,27 @@ export function ownershipFindings(tasks, repoRoot) {
             predictedBy.set(hit, ids);
         }
     }
+    const globOwnedTests = new Map();
+    const allSources = sourceFiles(repoRoot);
     for (const source of sources) {
         const ids = predictedBy.get(source.path) ?? new Set();
-        for (const { taskId } of namedSources(source.path, tasks))
-            ids.add(taskId);
+        for (const named of namedSources(source.path, tasks, allSources)) {
+            ids.add(named.taskId);
+            if (named.fromGlob) {
+                const owners = globOwnedTests.get(source.path) ?? new Set();
+                owners.add(named.taskId);
+                globOwnedTests.set(source.path, owners);
+            }
+        }
         if (ids.size > 0)
             predictedBy.set(source.path, ids);
     }
     const findings = [];
     for (const [test, taskIds] of predictedBy) {
-        if (owners(test).length === 0) {
+        if (owners(test).length === 0 && !(globOwnedTests.get(test)?.size)) {
             const ids = [...taskIds].sort();
             const source = sourceByPath.get(test);
-            const evidence = corroboration(source, namedSources(test, tasks));
+            const evidence = corroboration(source, namedSources(test, tasks, allSources));
             findings.push({
                 code: "unowned-test",
                 test,

package/dist/config/config.d.ts

@@ -266,6 +266,7 @@ export declare const TickmarkrConfigSchema: z.ZodObject<{
     }, z.core.$strip>;
     review: z.ZodObject<{
         complexityThreshold: z.ZodNumber;
+        timeoutMs: z.ZodNumber;
         required: z.ZodBoolean;
         prefer: z.ZodOptional<z.ZodArray<z.ZodString>>;
         policy: z.ZodOptional<z.ZodEnum<{

package/dist/config/config.js

@@ -355,6 +355,7 @@ export const TickmarkrConfigSchema = z.object({
         // because routing hints and the setup cockpit still read it as a complexity landmark; NOTHING in
         // src/gates/review.ts reads it any more, and a value here can no longer disable a review.
         complexityThreshold: z.number(),
+        timeoutMs: z.number().int().positive(),
         required: z.boolean(),
         prefer: z.array(z.string()).optional(),
         // R3: the operator's participation floor. "full" forces a cross-vendor review on every task;
@@ -484,8 +485,15 @@ export const DEFAULT_CONFIG = {
                 // 2026-07-16 (cursor-agent 2026.07.09); gives cursor a cheap-tier channel so low-complexity shapes
                 // stop burning its mid. Operator-approved 2026-07-16.
                 "composer-2.5-fast": "cheap",
+                // OBS-871: LiveBench 2026_06_25, fetched 2026-09-03; released 2026-09-01.
+                "claude-fable-5-1": "frontier",
+                // OBS-871: LiveBench 2026_06_25, fetched 2026-09-03; released 2026-09-02.
+                "gemini-3.8-flash": "mid",
+            },
+            windows: {
+                "composer-2.5": 200_000, "composer-2.5-fast": 200_000,
+                "claude-fable-5-1": 1_000_000, "gemini-3.8-flash": 1_000_000,
             },
-            windows: { "composer-2.5": 200_000, "composer-2.5-fast": 200_000 },
         },
         // GLM-5.2 → mid per benchmark policy (2026-07): SWE-bench Pro 62.1 (> GPT-5.5 58.6), FrontierSWE 74.4 ≈ Opus 4.8;
         // no independent Terminal-Bench score → conservative mid, overlays may raise.
@@ -507,8 +515,46 @@ export const DEFAULT_CONFIG = {
         // cross-harness) is pre-existing, FLEET-05 owns it.
         pi: {
             vendor: "zhipu", channel: "sub",
-            models: { "zai/glm-5.2": "mid" },
-            windows: { "zai/glm-5.2": 1_000_000 },
+            models: {
+                "zai/glm-5.2": "mid",
+                // OBS-871: LiveBench 2026_06_25, fetched 2026-09-03; GLM-5.3 released 2026-08-18.
+                "zai/glm-5.3": "frontier",
+                // OBS-871: same LiveBench source and 2026-09-03 fetch; speed variant remains mid.
+                "zai/glm-5.3-flash": "mid",
+            },
+            windows: { "zai/glm-5.2": 1_000_000, "zai/glm-5.3": 1_000_000, "zai/glm-5.3-flash": 200_000 },
+        },
+        // OBS-871: gateway ids and bands from LiveBench 2026_06_25, fetched 2026-09-03.
+        omp: {
+            vendor: "mixed", channel: "sub",
+            models: {
+                "google/gemini-3.8-flash": "mid",
+                "zai/glm-5.3": "frontier",
+                "alibaba/qwen3.8-max": "frontier",
+            },
+            modelOverrides: {
+                "google/gemini-3.8-flash": { vendor: "google" },
+                "zai/glm-5.3": { vendor: "zhipu" },
+                "alibaba/qwen3.8-max": { vendor: "alibaba" },
+            },
+            windows: {
+                "google/gemini-3.8-flash": 1_000_000,
+                "zai/glm-5.3": 1_000_000,
+                "alibaba/qwen3.8-max": 1_000_000,
+            },
+        },
+        // OBS-871: Qwen 3.8 Max scored 64.6 in LiveBench 2026_06_25, fetched 2026-09-03.
+        qwen: {
+            vendor: "alibaba", channel: "sub",
+            models: { "qwen3.8-max": "frontier" },
+            windows: { "qwen3.8-max": 1_000_000 },
+        },
+        // OBS-871: prime-agent listed this prime-inference route on 2026-09-03; GLM-5.2 remains mid.
+        "prime-agent": {
+            vendor: "mixed", channel: "sub",
+            models: { "prime-inference/z-ai/glm-5.2": "mid" },
+            modelOverrides: { "prime-inference/z-ai/glm-5.2": { vendor: "zhipu" } },
+            windows: { "prime-inference/z-ai/glm-5.2": 1_000_000 },
         },
         // Native grok CLI (Phase 40). grok-4.5 → mid: same benchmark provenance as the retired cursor-agent
         // grok-4.5-xhigh seed above (AA 54, TB2.1 83.3%, SWE-b Pro 64.7% — researched 2026-07-10).
@@ -543,7 +589,7 @@ export const DEFAULT_CONFIG = {
     judge: { adapter: "claude-code", model: "fable" },
     // R3: no `policy` floor — the neutral floor leaves the compiler's per-task assignment standing, so
     // the path-keyed rule is reachable out of the box rather than raised to full by construction.
-    review: { complexityThreshold: 7, required: true, criticalPaths: [...DEFAULT_REVIEW_CRITICAL_PATHS] },
+    review: { complexityThreshold: 7, timeoutMs: 900_000, required: true, criticalPaths: [...DEFAULT_REVIEW_CRITICAL_PATHS] },
     consult: { adapter: "claude-code", model: "fable", stallMinutes: 15 },
     // v1.4: gate LLM calls (judge/review/consult) run headless by default; pane opts back into visible agents.
     // v1.2: workers are the real agent TUI in the pane; "print" restores the -p-rendered-in-pane path.
@@ -713,7 +759,7 @@ export function overlayBytesLoadError(repoRoot, bytes, opts = {}) {
 export function configTemplate(overlay) {
     const base = `# tickmarkr config overlay — merges over built-in defaults (repo beats global beats defaults)
 # concurrency: 3
-# driver: auto            # auto | herdr | subprocess | orca
+# driver: auto            # auto: herdr when HERDR_ENV=1, then orca when TERM_PROGRAM=Orca + ORCA_TERMINAL_HANDLE, else subprocess; name orca explicitly outside an Orca terminal
 # taskTimeoutMinutes: 30
 # contextWarnTokens: 170000   # v1.23: journal+notify once per attempt when live worker context crosses this (status shows the sample)
 # setup: npm ci --prefer-offline   # run in each fresh task worktree before dispatch
@@ -765,7 +811,7 @@ export function configTemplate(overlay) {
 #   test: npm test
 #   byShape:
 #     docs: { acceptance: false, review: false }  # baseline, evidence, and scope are mandatory
-# review: { complexityThreshold: 7, required: true, prefer: [codex:gpt-5.6-sol, kimi] }
+# review: { complexityThreshold: 7, timeoutMs: 900000, required: true, prefer: [codex:gpt-5.6-sol, kimi] }
 #                         # prefer: ordered reviewer seat preference (adapter | adapter:model); ranks
 #                         # diversity-eligible channels only — never admits a same-vendor/same-model reviewer
 # consult: { adapter: claude-code, model: fable, stallMinutes: 15, prefer: [codex:gpt-5.6-sol, kimi:kimi-code/k3] }

package/dist/drivers/herdr.d.ts

@@ -56,6 +56,7 @@ export declare class HerdrDriver implements ExecutorDriver {
     private journal?;
     id: string;
     interactive: boolean;
+    readSource: string;
     private groups;
     private groupSerial;
     private deliverySerial;

package/dist/drivers/herdr.js

@@ -142,6 +142,7 @@ export class HerdrDriver {
     journal;
     id = "herdr";
     interactive = true;
+    readSource = "recent-unwrapped";
     groups = new Map();
     // grouped slot()/close() mutate shared group state across awaits — serialize them so two
     // concurrent first members can never both create the group tab (mergeSerial idiom, daemon.ts)
@@ -1339,7 +1340,16 @@ export class HerdrDriver {
                 if (alive.has(tab))
                     continue;
                 const closed = await this.herdr(`tab close ${shq(tab)}`);
-                if (closed.code !== 0) {
+                if (closed.code !== 0 && /tab_not_found/i.test(`${closed.stderr}\n${closed.stdout}`)) {
+                    this.journalReconcile(journalHandle, "tab-reconcile-close-skipped", undefined, {
+                        tabId: tab,
+                        runId,
+                        sweeperRunId: runId,
+                        exitCode: 0,
+                        reason: "tab_not_found",
+                    });
+                }
+                else if (closed.code !== 0) {
                     this.journalReconcile(journalHandle, "tab-reconcile-close-failed", undefined, {
                         tabId: tab,
                         runId,

package/dist/drivers/index.d.ts

@@ -2,6 +2,12 @@ import type { TickmarkrConfig } from "../config/config.js";
 import type { ExecutorDriver } from "./types.js";
 export declare const DRIVER_CHOICES: readonly ["auto", "herdr", "subprocess", "orca"];
 export type DriverChoice = (typeof DRIVER_CHOICES)[number];
+/**
+ * Orca authors both markers on every terminal it creates. Requiring the pair avoids treating an
+ * unrelated TERM_PROGRAM value or a copied terminal handle as host identity. This is deliberately
+ * environment-only: selection must not execute a binary or contact the Orca runtime.
+ */
+export declare function orcaHostDetected(env?: NodeJS.ProcessEnv): boolean;
 /** Validate argv at the CLI boundary rather than casting an arbitrary string into a driver choice. */
 export declare function parseDriverOverride(override?: string): DriverChoice | undefined;
 export declare function driverEvidence(cfg: TickmarkrConfig, driver: ExecutorDriver, override?: string): string;

package/dist/drivers/index.js

@@ -3,6 +3,14 @@ import { OrcaDriver } from "./orca.js";
 import { SubprocessDriver } from "./subprocess.js";
 export const DRIVER_CHOICES = ["auto", "herdr", "subprocess", "orca"];
 const overrideByDriver = new WeakMap();
+/**
+ * Orca authors both markers on every terminal it creates. Requiring the pair avoids treating an
+ * unrelated TERM_PROGRAM value or a copied terminal handle as host identity. This is deliberately
+ * environment-only: selection must not execute a binary or contact the Orca runtime.
+ */
+export function orcaHostDetected(env = process.env) {
+    return env.TERM_PROGRAM === "Orca" && env.ORCA_TERMINAL_HANDLE !== undefined;
+}
 /** Validate argv at the CLI boundary rather than casting an arbitrary string into a driver choice. */
 export function parseDriverOverride(override) {
     if (override === undefined)
@@ -20,8 +28,13 @@ export function driverEvidence(cfg, driver, override) {
     if (want !== "auto")
         return `${driver.id} (config)`;
     const herdrAvailable = process.env.HERDR_ENV === "1";
-    if (driver.id === (herdrAvailable ? "herdr" : "subprocess")) {
-        return `auto → ${driver.id} (${herdrAvailable ? "HERDR_ENV=1" : "HERDR_ENV unset"})`;
+    if (herdrAvailable && driver.id === "herdr")
+        return "auto → herdr (HERDR_ENV=1)";
+    if (!herdrAvailable && orcaHostDetected() && driver.id === "orca") {
+        return "auto → orca (TERM_PROGRAM+ORCA_TERMINAL_HANDLE)";
+    }
+    if (!herdrAvailable && !orcaHostDetected() && driver.id === "subprocess") {
+        return "auto → subprocess (HERDR_ENV unset)";
     }
     return `auto → ${driver.id} (runtime)`;
 }
@@ -34,9 +47,11 @@ export function pickDriver(cfg, override) {
     const driver = want === "herdr" ? new HerdrDriver("herdr", cfg.visibility.workersPerTab)
         : want === "subprocess" ? new SubprocessDriver()
             // Orca is an operator-selected execution surface. Its runtime failure stays on Orca; selection
-            // must never substitute a hidden subprocess worker after this explicit choice.
+            // must never substitute a hidden subprocess worker after an explicit or detected choice.
             : want === "orca" ? new OrcaDriver()
-                : HerdrDriver.available() ? new HerdrDriver("herdr", cfg.visibility.workersPerTab) : new SubprocessDriver();
+                : HerdrDriver.available() ? new HerdrDriver("herdr", cfg.visibility.workersPerTab)
+                    : orcaHostDetected() ? new OrcaDriver()
+                        : new SubprocessDriver();
     if (selectedOverride !== undefined)
         overrideByDriver.set(driver, selectedOverride);
     return driver;

package/dist/drivers/orca.d.ts

@@ -1,7 +1,8 @@
 import { type ShResult } from "../run/git.js";
+import { type JournalEvent } from "../run/journal.js";
 import { type ExecutorDriver, type NotifyOpts, type Slot, type SlotOpts } from "./types.js";
 /** The response families the ONE shared envelope parser serves. There is no second JSON seam. */
-export declare const ORCA_RESPONSE_FAMILIES: readonly ["status", "create", "list", "read", "send", "wait", "show", "close"];
+export declare const ORCA_RESPONSE_FAMILIES: readonly ["status", "create", "list", "read", "send", "wait", "show", "close", "worktree-current", "worktree-set", "hooks-status"];
 export type OrcaFamily = (typeof ORCA_RESPONSE_FAMILIES)[number];
 export declare const ORCA_FIXTURE_VERSION = "1.4.195";
 export declare const ORCA_CLI_COMMAND_ENV = "ORCA_CLI_COMMAND";
@@ -13,6 +14,7 @@ export declare const NOT_WRITABLE_CODE = "terminal_not_writable";
 /** The ONLY terminal status that licenses reading a terminal's bytes or its agent state. */
 export declare const RUNNING_STATUS = "running";
 export declare const STATUS_GOVERNED_METHODS: readonly ["read", "waitOutput", "status", "waitAgentStatus"];
+export declare const WORKTREE_ADOPTION_TIMEOUT_MS = 60000;
 export interface OrcaExec {
     (args: string[], cwd: string, timeoutMs?: number): Promise<ShResult>;
 }
@@ -113,6 +115,11 @@ export declare class OrcaDriver implements ExecutorDriver {
     private pageLines;
     private pollMs;
     private probeStalenessMs;
+    private journalRoots;
+    private narrate?;
+    private hookCoverage?;
+    private taskWorktrees;
+    private pendingProjects;
     constructor(opts?: OrcaDriverOpts);
     private call;
     /** The live runtime's identity, or an explicit failure. A missing or unreachable runtime is a
@@ -126,8 +133,23 @@ export declare class OrcaDriver implements ExecutorDriver {
     private state;
     private latched;
     private assertAvailable;
+    describe(slot: Slot): {
+        surface?: string;
+        hostPlatform?: string;
+    };
     run(slot: Slot, cmd: string): Promise<void>;
+    private sendText;
+    private sendReceipt;
     private create;
+    /**
+     * A freshly-created git checkout does not become a valid Orca selector atomically. Ask
+     * `worktree current` FROM that checkout until Orca itself resolves the exact filesystem identity;
+     * an enclosing checkout is still not adoption. Only selector_not_found is a retryable refusal —
+     * malformed envelopes and every other refusal remain explicit driver failures.
+     */
+    private awaitWorktreeAdoption;
+    /** Same repo/run/narration path Herdr uses for its driver-owned dispatch-retry row. */
+    private appendAdoptionWait;
     /**
      * Every terminal-addressed call — read AND write — goes through here, and the runtime identity is
      * established BEFORE the runtime-scoped handle goes on the wire. Discarding a lookalike's answer
@@ -149,6 +171,8 @@ export declare class OrcaDriver implements ExecutorDriver {
     private liveShowTerm;
     private tailText;
     private readPage;
+    /** A rendered-frame liveness read. `--screen` and `--cursor` are mutually exclusive in Orca. */
+    private readScreen;
     /** A single UNPAGED tail read — exactly what the caller asked for and nothing more. Markers split
      *  across cursor pages are not reassembled here; that is waitOutput's job. */
     read(slot: Slot, lines: number): Promise<string>;
@@ -163,12 +187,22 @@ export declare class OrcaDriver implements ExecutorDriver {
         regex?: boolean;
     }): Promise<boolean>;
     status(slot: Slot): Promise<string>;
+    private hookAgent;
+    /** true = hooked, false = definitively unhooked, undefined = agent absent from Orca's table. */
+    private agentHookAvailable;
+    private loadHookCoverage;
     /** One `terminal wait` through the full identity machinery. The recorded 1.4.186 elapsed answer
      *  is rc 1 + ok:true + {handle, condition, satisfied:false, status:"running"}; it is "not yet"
      *  only after this method validates all four fields. Any malformed/refused wait remains explicit. */
     private waitCondition;
     waitAgentStatus(slot: Slot, status: string, timeoutMs: number): Promise<boolean>;
+    sendKey(slot: Slot, key: string): Promise<void>;
+    nudge(slot: Slot, message: string): Promise<boolean>;
+    narrator(cwd: string, command: string, runId?: string): Promise<Slot>;
+    project(taskId: string, state: "in-progress" | "in-review" | "completed"): Promise<void>;
+    private setWorkspaceStatus;
     notify(msg: string, opts?: NotifyOpts): Promise<void>;
+    narrateWith(narrate: (event: JournalEvent) => void): void;
     close(slot: Slot): Promise<void>;
     /**
      * The one destructive call in this driver, for a slot's own terminal AND for a reconcile candidate