primitive-admin 1.1.0-alpha.8 → 1.1.0-alpha.80

package/dist/src/commands/workflows.js

@@ -1,21 +1,304 @@
-import { readFileSync, writeFileSync, statSync } from "fs";
-import { basename } from "path";
-import * as TOML from "@iarna/toml";
-import { lookup as mimeLookup } from "mime-types";
+import { readFileSync, writeFileSync, readdirSync, } from "fs";
+import * as path from "path";
+import { stringifyConfigToml } from "../lib/config-toml.js";
+import { generateWorkflowTypes, } from "../lib/workflow-codegen/generator.js";
+import { generateWorkflowSwiftTypes } from "../lib/swift-codegen/generator.js";
 import { ApiClient } from "../lib/api-client.js";
 import { resolveAppId } from "../lib/config.js";
+import { resolveWorkflowSelector } from "../lib/block-selector.js";
+import { formatAccessRuleForDisplay } from "../lib/access-rule-display.js";
+import { resolveCodegenSourceDir } from "../lib/codegen-shared/resolveCodegenSourceDir.js";
+import { expandWorkflow } from "../lib/workflow-fragments.js";
+import { confirmPrompt } from "../lib/confirm-prompt.js";
+import { parseStatusFilter } from "../lib/object-status-filter.js";
+import { buildWorkflowRunEnvelope, buildWorkflowStepEnvelope, collectWorkflowFailures, describeEmptyRunPage, describeFailureSweep, formatFailureCodeCell, formatFailureErrorCell, formatFailureStepCell, selectFailedStepRow, toWorkflowRunInspectionRow, DEFAULT_FAILURE_SWEEP_MAX_SCAN, FAILURE_SWEEP_MAX_REQUESTS, SERVER_SCAN_ROWS_PER_REQUEST, } from "../lib/log-inspection.js";
 import chalk from "chalk";
-import { success, error, info, warn, keyValue, formatTable, formatId, formatDate, formatStatus, formatDuration, json, divider, } from "../lib/output.js";
+import { success, error, info, warn, keyValue, result as printResult, formatTable, formatId, formatDate, formatStatus, formatDuration, json, jsonLine, divider, dim, progress, progressEnd, } from "../lib/output.js";
+import { pollUntil, followLoop, watchLoop, parseIntervalSeconds, } from "../lib/watch.js";
+import { APPLY_RUN_STATUSES, RUN_STATUS_LIST, SETTLED_RUN_STATUSES, } from "../lib/run-status.js";
+import { requireTestCaseInputVariables } from "../lib/test-case-variables.js";
+/**
+ * Render the `Step Results` section of `workflows runs status`. Pure
+ * formatting helper — returns lines as strings so it's unit-testable
+ * (the caller console.logs them). For each step, emits:
+ *
+ *   `  <id>: <status-label> <duration>`
+ *
+ * and, when applicable, indented follow-up lines:
+ *
+ * - `error_captured` step with `error` → `    error: <truncated message>`
+ * - forEach output with `errors[]` non-empty →
+ *   `    forEach errors: <N/M> - first: <truncated message>`
+ *
+ * The truncation length matches `runs steps` ERROR column (80 chars,
+ * trailing "..." when truncated). See #688.
+ */
+/**
+ * Stop-polling predicate for the `workflows preview --wait` loop (issues #941,
+ * #2348). True when the run has settled and the wait should stop — as opposed
+ * to a still-in-flight status ("queued"/"running") that keeps the loop going
+ * until the cap.
+ *
+ * Since #2348 the server reconciles every status surface onto the canonical
+ * vocabulary before it goes on the wire, so this reads the one shared set in
+ * `lib/run-status.ts` instead of re-deriving terminality here:
+ *   - "completed" / "failed" / "terminated" — true terminals. "terminated" is
+ *     an aborted preview, a real end state (Codex P2 on #941: it used to be
+ *     omitted, so an abort was misreported as a wait timeout).
+ *   - "apply_pending" / "apply_claimed" — the run finished and is parked
+ *     waiting for a client to apply its result. It never advances on its own,
+ *     so the previous ladder spun here until the 60s cap.
+ *   - "missing" — no live instance to observe; more polling cannot help.
+ *
+ * The raw Cloudflare spellings ("complete"/"errored"/"waiting") are gone: the
+ * server no longer puts them on the wire, and matching them here would have
+ * kept two vocabularies alive.
+ */
+export function isPreviewWaitTerminalStatus(s) {
+    return typeof s === "string" && SETTLED_RUN_STATUSES.has(s);
+}
+/**
+ * Which arm the `workflows preview --wait` render ladder takes (#2636).
+ *
+ * Extracted from the ladder so the branch decision is testable: the point of
+ * the `skipped` arm is that an elided preview — one whose declarative lock was
+ * held under `onContention: "ignore"` — reports that it did not run, instead
+ * of falling through to the timeout branch and exiting 124 as though it were
+ * still going.
+ */
+export function previewWaitOutcome(status) {
+    if (status === "completed")
+        return "completed";
+    if (status === "failed")
+        return "failed";
+    if (status === "terminated")
+        return "terminated";
+    if (status === "skipped")
+        return "skipped";
+    if (typeof status === "string" && APPLY_RUN_STATUSES.has(status))
+        return "apply";
+    if (status === "missing")
+        return "missing";
+    return "timeout";
+}
+export function renderRunStatusStepResults(stepResults) {
+    const lines = [];
+    for (const step of stepResults || []) {
+        const statusLabel = step.skipped
+            ? chalk.gray("skipped")
+            : step.status === "error_captured"
+                ? chalk.yellow("error_captured")
+                : step.status === "failed"
+                    ? chalk.red("failed")
+                    : step.status === "completed"
+                        ? chalk.green("completed")
+                        : step.status || "?";
+        const durationLabel = formatDuration(step.durationMs);
+        lines.push(`  ${step.id}: ${statusLabel} ${durationLabel}`);
+        // Single-step error_captured: show the captured error message.
+        if (step.status === "error_captured" && step.error) {
+            const errMsg = typeof step.error === "string" ? step.error : String(step.error);
+            const truncated = errMsg.length > 80 ? errMsg.slice(0, 77) + "..." : errMsg;
+            lines.push(`    ${chalk.dim("error:")} ${chalk.yellow(truncated)}`);
+        }
+        else if (step.status === "error_captured" && step.errorDetails) {
+            // No top-level error but errorDetails present — hint to drill in.
+            lines.push(`    ${chalk.dim("error: <see runs step-detail for full errorDetails>")}`);
+        }
+        // forEach aggregated errors: a one-line summary surfaces the count
+        // of `error_captured` iterations and a sample message (#688).
+        const fo = step.output;
+        if (fo && typeof fo === "object" && Array.isArray(fo.errors) && fo.errors.length > 0) {
+            const total = (Array.isArray(fo.items) ? fo.items.length : 0)
+                || (fo.totalSucceeded ?? 0) + (fo.totalFailed ?? 0);
+            const firstErr = (Array.isArray(fo.errorMessages) && fo.errorMessages.length > 0
+                ? fo.errorMessages[0]
+                : (fo.errors[0]?.error ?? "?"));
+            const firstStr = typeof firstErr === "string" ? firstErr : String(firstErr);
+            const truncated = firstStr.length > 80 ? firstStr.slice(0, 77) + "..." : firstStr;
+            lines.push(`    ${chalk.dim("forEach errors:")} ${chalk.yellow(`${fo.totalFailed ?? fo.errors.length}/${total}`)} ${chalk.dim("- first:")} ${chalk.yellow(truncated)}`);
+        }
+    }
+    return lines;
+}
+/**
+ * #2771 — how long a step row has been going.
+ *
+ * A step that is still executing is persisted with status `running` and a
+ * `durationMs` of 0 (the column is required; the real value is written when the
+ * step ends). Printing that 0 hides the one number that matters while a run is
+ * in flight — how long the current step has been stuck — so a running row is
+ * measured from its `startedAt` instead. Finished rows keep their stored
+ * duration. Exported for testability.
+ */
+export function stepRunDisplayDurationMs(step, now = Date.now()) {
+    const stored = typeof step?.durationMs === "number" ? step.durationMs : null;
+    if (step?.status !== "running")
+        return stored;
+    const started = step?.startedAt ? new Date(step.startedAt).getTime() : NaN;
+    if (Number.isNaN(started))
+        return stored;
+    return Math.max(0, now - started);
+}
+/**
+ * #2771 — the step lines `workflows runs status` should render.
+ *
+ * Cloudflare's instance status carries `stepResults` only once the run reaches
+ * a terminal state, so a live run used to print `Status: running` and nothing
+ * else. The persisted step rows now exist DURING the run, so they are the
+ * fallback: mapped onto the shape `renderRunStatusStepResults` reads, with a
+ * running step's elapsed time in place of its 0 placeholder. The instance's own
+ * results always win when it has them. Exported for testability.
+ */
+export function stepResultsForRunStatus(instanceStepResults, persistedStepRuns, now = Date.now()) {
+    if (instanceStepResults && instanceStepResults.length > 0)
+        return instanceStepResults;
+    return (persistedStepRuns || []).map((row) => ({
+        id: row?.stepId,
+        kind: row?.stepKind,
+        status: row?.status,
+        durationMs: stepRunDisplayDurationMs(row, now),
+        error: row?.error,
+        errorDetails: row?.errorDetails,
+        output: row?.output,
+    }));
+}
+/**
+ * #1367 — annotate step-run rows with the inter-step gap: the wall-clock time
+ * between the previous step's `endedAt` and this step's `startedAt`. Sorts by
+ * `startedAt` when every row has a parseable one (tie-break `stepIndex`),
+ * otherwise by `stepIndex` for the whole list — the ordering mode must be
+ * chosen once for all rows, not per pair: a per-pair timestamp-vs-index
+ * fallback is not transitive when some rows lack a timestamp (A<B by index,
+ * B<C by index, C<A by time), and a non-transitive comparator leaves
+ * Array.prototype.sort free to return an arbitrary order, computing gaps
+ * against the wrong predecessor. The first step has no predecessor, so its
+ * `gapMs` is `null`. A gap is only meaningful when both adjacent timestamps
+ * parse and the gap is positive. When the immediate predecessor lacks a valid
+ * `endedAt`, the gap is left `null` rather than measured from an older,
+ * non-adjacent step — otherwise a later step could raise a false stall
+ * warning against a step it does not actually follow. Exported for
+ * testability.
+ */
+export function annotateStepGaps(steps) {
+    const rows = [...(steps || [])];
+    const allTimed = rows.every((r) => r?.startedAt && !Number.isNaN(new Date(r.startedAt).getTime()));
+    rows.sort((a, b) => {
+        if (allTimed) {
+            const sa = new Date(a.startedAt).getTime();
+            const sb = new Date(b.startedAt).getTime();
+            if (sa !== sb)
+                return sa - sb;
+        }
+        return (a?.stepIndex ?? 0) - (b?.stepIndex ?? 0);
+    });
+    let prevEnded = null;
+    for (const row of rows) {
+        const started = row?.startedAt ? new Date(row.startedAt).getTime() : NaN;
+        if (prevEnded != null && !Number.isNaN(started)) {
+            const gap = started - prevEnded;
+            row.gapMs = gap > 0 ? gap : 0;
+        }
+        else {
+            row.gapMs = null;
+        }
+        // Only genuinely adjacent steps yield a gap: if this step lacks a valid
+        // `endedAt`, clear the predecessor so the NEXT step's gap is left null
+        // instead of being measured from a non-adjacent earlier step.
+        const ended = row?.endedAt ? new Date(row.endedAt).getTime() : NaN;
+        prevEnded = Number.isNaN(ended) ? null : ended;
+    }
+    return rows;
+}
+/**
+ * #1367 — compute min / p50 / p95 / max over a set of latency samples (ms).
+ * Ignores non-finite values. Returns `null` when there are no valid samples.
+ * Percentiles use nearest-rank on the sorted samples. Exported for testability.
+ */
+export function summarizeLatencyDistribution(values) {
+    const nums = (values || []).filter((v) => typeof v === "number" && Number.isFinite(v));
+    if (nums.length === 0)
+        return null;
+    nums.sort((a, b) => a - b);
+    const pct = (p) => {
+        const rank = Math.ceil((p / 100) * nums.length);
+        return nums[Math.min(nums.length - 1, Math.max(0, rank - 1))];
+    };
+    return {
+        count: nums.length,
+        min: nums[0],
+        p50: pct(50),
+        p95: pct(95),
+        max: nums[nums.length - 1],
+    };
+}
+/**
+ * Issue #687 (review feedback): render an unambiguous label for a workflow
+ * config slot ("active config", "draft config", etc.) when the slot may be
+ * unnamed. Falls back to a short ID prefix so users staging multiple unnamed
+ * configs can still tell them apart.
+ *
+ *   formatConfigSlotLabel("active config", "v2", "01KRT...")        → 'active config "v2"'
+ *   formatConfigSlotLabel("active config", null, "01KRTABCDXYZ...") → 'active config (01KRTABCDXYZ…)'
+ *   formatConfigSlotLabel("active config", null, null)              → 'active config'
+ *
+ * Exported for testability.
+ */
+export function formatConfigSlotLabel(slot, configName, configId, prefixLength = 12) {
+    if (configName) {
+        return `${slot} "${configName}"`;
+    }
+    if (configId) {
+        return `${slot} (${String(configId).slice(0, prefixLength)}…)`;
+    }
+    return slot;
+}
+/**
+ * Resolve a workflow reference — its server id, or the author-chosen key its
+ * TOML file is named for — for the `tests` verbs (#3249).
+ *
+ * The key is what an operator has: `workflows/<key>.toml` is the object and
+ * `workflows/<key>.tests/` its cases. The resolution lives in
+ * `block-selector.ts`, shared with `prompts`.
+ */
+async function resolveWorkflowId(client, appId, ref) {
+    return (await resolveWorkflowSelector(client, appId, ref)).blockId;
+}
 export function registerWorkflowsCommands(program) {
     const workflows = program
         .command("workflows")
-        .description("Build multi-step workflows, publish revisions, and monitor runs")
+        .description("Inspect workflows, run and monitor them, and take one out of service")
         .addHelpText("after", `
 Examples:
   $ primitive workflows list
-  $ primitive workflows create --from-file process-doc.toml
-  $ primitive workflows publish 01HXY...
+  $ primitive config create workflow process-doc       # scaffold workflows/process-doc.toml
+  $ primitive config push --only workflow/process-doc  # apply it
   $ primitive workflows runs list 01HXY...
+  $ primitive workflows disable 01HXY...               # take it out of service now
+`);
+    // Server-side configuration is authored in TOML and applied with `config push`
+    // (issue #2645) — this group no longer carries config-setting flags.
+    workflows.addHelpText("after", `
+Configuration lives in workflows/<key>.toml:
+  $ primitive config fields workflow        # the file's keys, types and defaults
+  $ primitive config create workflow <key>  # scaffold one locally
+  $ primitive config set workflow/<key> <path>=<value>
+  $ primitive config push --only workflow/<key>
+
+Which body runs is configuration too: name it with activeConfigName in
+workflows/<key>.toml (bodies live in workflows/<key>.configs/<name>.toml) and
+apply it with 'config push'. There is no other write path: revisions are
+read-only history from the retired legacy path (issue #2768).
+`);
+    // Per-subject analytics have one home, the `analytics` noun (issue #2766) —
+    // this group's own analytics commands were removed, so say where they went.
+    workflows.addHelpText("after", `
+Analytics live under the analytics noun:
+  $ primitive analytics workflows --window-days 7 --limit 10
+
+That is the single home for per-subject analytics (workflows, prompts,
+integrations). It ranks workflows by runs; there is no app-wide workflow
+overview — the endpoint behind the retired overview command never existed.
 `);
     // List workflows
     workflows
@@ -23,14 +306,22 @@ Examples:
         .description("List workflows")
         .argument("[app-id]", "App ID (uses current app if not specified)")
         .option("--app <app-id>", "App ID")
-        .option("--status <status>", "Filter by status: draft, active, archived")
+        .option("--status <status>", "Filter by status: active, inactive, archived")
         .option("--json", "Output as JSON")
         .action(async (appId, options) => {
         const resolvedAppId = resolveAppId(appId, options);
         const client = new ApiClient();
         try {
+            // #2803 — `draft` is gone; refuse it by name rather than returning the
+            // empty list a canonical filter would produce. #2887 added `archived`:
+            // the delete flow writes one now, so asking for the retired rows is a
+            // question with an answer rather than a 400.
             const { items } = await client.listWorkflows(resolvedAppId, {
-                status: options.status,
+                status: parseStatusFilter("workflows", options.status, [
+                    "active",
+                    "inactive",
+                    "archived",
+                ]),
             });
             if (options.json) {
                 json(items);
@@ -53,83 +344,10 @@ Examples:
             process.exit(1);
         }
     });
-    // Create workflow
-    workflows
-        .command("create")
-        .description("Create a new workflow")
-        .argument("[app-id]", "App ID (uses current app if not specified)")
-        .option("--app <app-id>", "App ID")
-        .option("--key <key>", "Workflow key (unique identifier)")
-        .option("--name <name>", "Display name")
-        .option("--description <desc>", "Description")
-        .option("--from-file <path>", "Load workflow from TOML file")
-        .option("--requires-client-apply <bool>", "Require client-side apply: true or false")
-        .option("--json", "Output as JSON")
-        .action(async (appId, options) => {
-        const resolvedAppId = resolveAppId(appId, options);
-        const client = new ApiClient();
-        let payload;
-        if (options.fromFile) {
-            try {
-                const content = readFileSync(options.fromFile, "utf-8");
-                const tomlData = TOML.parse(content);
-                const workflow = tomlData.workflow || tomlData;
-                payload = {
-                    workflowKey: workflow.key || workflow.workflowKey,
-                    name: workflow.name,
-                    description: workflow.description,
-                    steps: tomlData.steps || [],
-                    inputSchema: workflow.inputSchema ? JSON.parse(workflow.inputSchema) : undefined,
-                    outputSchema: workflow.outputSchema ? JSON.parse(workflow.outputSchema) : undefined,
-                    perUserMaxRunning: workflow.perUserMaxRunning,
-                    perUserMaxQueued: workflow.perUserMaxQueued,
-                    perAppMaxRunning: workflow.perAppMaxRunning,
-                    perAppMaxQueued: workflow.perAppMaxQueued,
-                    queueTtlSeconds: workflow.queueTtlSeconds,
-                    dequeueOrder: workflow.dequeueOrder,
-                    requiresClientApply: workflow.requiresClientApply,
-                };
-            }
-            catch (err) {
-                error(`Failed to read TOML file: ${err.message}`);
-                process.exit(1);
-            }
-        }
-        else {
-            if (!options.key || !options.name) {
-                error("Required: --key, --name (or use --from-file)");
-                process.exit(1);
-            }
-            payload = {
-                workflowKey: options.key,
-                name: options.name,
-                description: options.description,
-                steps: [],
-            };
-        }
-        if (options.requiresClientApply !== undefined) {
-            payload.requiresClientApply = options.requiresClientApply === "true";
-        }
-        try {
-            const result = await client.createWorkflow(resolvedAppId, payload);
-            if (options.json) {
-                json(result);
-                return;
-            }
-            success(`Workflow created: ${result.workflow?.name || options.name}`);
-            keyValue("Workflow ID", result.workflow?.workflowId);
-            keyValue("Key", result.workflow?.workflowKey);
-        }
-        catch (err) {
-            error(err.message);
-            process.exit(1);
-        }
-    });
-    // Get workflow
     workflows
         .command("get")
         .description("Get workflow details")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow-id>", "Workflow ID or key")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--json", "Output as JSON")
         .action(async (workflowId, options) => {
@@ -142,20 +360,36 @@ Examples:
                 return;
             }
             const wf = result.workflow;
-            keyValue("Workflow ID", wf.workflowId);
-            keyValue("Key", wf.workflowKey);
-            keyValue("Name", wf.name);
-            keyValue("Description", wf.description);
-            keyValue("Status", formatStatus(wf.status));
-            keyValue("Active Config", wf.activeConfigId || "-");
-            keyValue("Latest Revision", wf.latestRevision || "-");
-            keyValue("Client Apply", wf.requiresClientApply !== false ? "yes" : "no");
+            printResult("Workflow ID", wf.workflowId);
+            printResult("Key", wf.workflowKey);
+            printResult("Name", wf.name);
+            printResult("Description", wf.description);
+            // #2803 — ONE availability row. The server returns the canonical
+            // value, so a workflow an operator disabled reads `inactive` here and
+            // there is no second row left to disagree with it.
+            printResult("Status", formatStatus(wf.status));
+            printResult("Active Config", wf.activeConfigId || "-");
+            printResult("Latest Revision", wf.latestRevision || "-");
+            printResult("Client Apply", wf.requiresClientApply !== false ? "yes" : "no");
+            printResult("Sync Callable", wf.syncCallable === true ? "yes" : "no");
+            // #1233 — runAs unset means the engine defaults to "caller"; show that
+            // default rather than a blank so the deployed principal mode is visible.
+            printResult("Run As", wf.runAs || "caller");
+            printResult("Capabilities", Array.isArray(wf.capabilities) && wf.capabilities.length > 0
+                ? wf.capabilities.join(", ")
+                : "-");
+            // #2652: a caller workflow with no rule denies every non-admin start,
+            // so the row is always printed — the sentinel names the consequence,
+            // and a `runAs: "system"` workflow gets its own exempt wording.
+            printResult("Access Rule", formatAccessRuleForDisplay(wf.accessRule, "workflow", {
+                runAs: wf.runAs,
+            }));
             divider();
             info("Queue Settings:");
-            keyValue("  Per User Max Running", wf.perUserMaxRunning);
-            keyValue("  Per User Max Queued", wf.perUserMaxQueued);
-            keyValue("  Per App Max Running", wf.perAppMaxRunning);
-            keyValue("  Dequeue Order", wf.dequeueOrder);
+            printResult("  Per User Max Running", wf.perUserMaxRunning);
+            printResult("  Per User Max Queued", wf.perUserMaxQueued);
+            printResult("  Per App Max Running", wf.perAppMaxRunning);
+            printResult("  Dequeue Order", wf.dequeueOrder);
             if (wf.inputSchema) {
                 divider();
                 info("Input Schema:");
@@ -196,163 +430,136 @@ Examples:
             process.exit(1);
         }
     });
-    // Update workflow
     workflows
-        .command("update")
-        .description("Update workflow metadata")
-        .argument("<workflow-id>", "Workflow ID")
-        .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--name <name>", "Display name")
-        .option("--description <desc>", "Description")
-        .option("--status <status>", "Status: draft, active, archived")
-        .option("--per-user-max-running <n>", "Max running per user")
-        .option("--per-user-max-queued <n>", "Max queued per user")
-        .option("--dequeue-order <order>", "Dequeue order: fifo, lifo")
-        .option("--requires-client-apply <bool>", "Require client-side apply: true or false")
-        .option("--json", "Output as JSON")
-        .action(async (workflowId, options) => {
-        const resolvedAppId = resolveAppId(undefined, options);
-        const payload = {};
-        if (options.name)
-            payload.name = options.name;
-        if (options.description)
-            payload.description = options.description;
-        if (options.status)
-            payload.status = options.status;
-        if (options.perUserMaxRunning)
-            payload.perUserMaxRunning = parseInt(options.perUserMaxRunning);
-        if (options.perUserMaxQueued)
-            payload.perUserMaxQueued = parseInt(options.perUserMaxQueued);
-        if (options.dequeueOrder)
-            payload.dequeueOrder = options.dequeueOrder;
-        if (options.requiresClientApply !== undefined) {
-            payload.requiresClientApply = options.requiresClientApply === "true";
-        }
-        if (Object.keys(payload).length === 0) {
-            error("No update options specified.");
-            process.exit(1);
-        }
-        const client = new ApiClient();
+        .command("expand")
+        .description("Expand a workflow TOML's include fragments and print the result")
+        .argument("<file>", "Path to the workflow TOML file")
+        .option("--format <format>", "Output format: json (default) or toml", "json")
+        .action((file, options) => {
         try {
-            const result = await client.updateWorkflow(resolvedAppId, workflowId, payload);
-            if (options.json) {
-                json(result);
-                return;
+            const expanded = expandWorkflow(file);
+            if (options.format === "toml") {
+                // Strip the `include` key (already removed by expander) and emit
+                // valid TOML so the result is round-trippable.
+                console.log(stringifyConfigToml(expanded));
+            }
+            else {
+                console.log(JSON.stringify(expanded, null, 2));
             }
-            success("Workflow updated.");
         }
         catch (err) {
-            error(err.message);
+            error(err?.message ?? String(err));
             process.exit(1);
         }
     });
-    // Delete workflow
+    // #2645 criterion 9 / #2803 — the operational verbs.
+    //
+    // Configuration is authored in TOML and applied with `config push`, which would
+    // otherwise mean an operator needs a repo checkout to stop a misbehaving
+    // workflow. It does not: taking something out of service is an OPERATION, and
+    // these two verbs are the ONLY writers of `status`, which is server-owned and
+    // no longer a TOML key. The committed configuration says nothing about
+    // availability, so `config diff` reports the server's status as operator
+    // state rather than as drift.
     workflows
-        .command("delete")
-        .description("Delete or archive a workflow")
-        .argument("<workflow-id>", "Workflow ID")
+        .command("disable")
+        .description("Take a workflow out of service now. Operational, not configuration: `status` is server-owned and 'config push' can neither set nor clear it.")
+        .argument("<workflow-id>", "Workflow ID or key")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--hard", "Permanently delete instead of archive")
-        .option("-y, --yes", "Skip confirmation prompt")
+        .option("--json", "Output as JSON")
         .action(async (workflowId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
-        if (!options.yes) {
-            const action = options.hard ? "permanently delete" : "archive";
-            const inquirer = await import("inquirer");
-            const { confirm } = await inquirer.default.prompt([
-                {
-                    type: "confirm",
-                    name: "confirm",
-                    message: `Are you sure you want to ${action} workflow ${workflowId}?`,
-                    default: false,
-                },
-            ]);
-            if (!confirm) {
-                info("Cancelled.");
-                return;
-            }
-        }
         const client = new ApiClient();
         try {
-            if (options.hard) {
-                await client.deleteWorkflow(resolvedAppId, workflowId);
-                success("Workflow deleted.");
-            }
-            else {
-                await client.updateWorkflow(resolvedAppId, workflowId, { status: "archived" });
-                success("Workflow archived.");
+            const result = await client.disableWorkflow(resolvedAppId, workflowId);
+            if (options.json) {
+                json(result);
+                return;
             }
+            success(`Workflow disabled: ${result.workflowKey || workflowId}`);
+            info("New runs are refused. Configuration is unchanged — `primitive workflows enable` puts it back.");
         }
         catch (err) {
             error(err.message);
             process.exit(1);
         }
     });
-    // Draft subcommand
-    const draft = workflows.command("draft").description("Manage workflow draft");
-    // Update draft
-    draft
-        .command("update")
-        .description("Update workflow draft steps")
-        .argument("<workflow-id>", "Workflow ID")
+    workflows
+        .command("enable")
+        .description("Put a workflow back in service.")
+        .argument("<workflow-id>", "Workflow ID or key")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--from-file <path>", "Load steps from TOML file")
         .option("--json", "Output as JSON")
         .action(async (workflowId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
-        if (!options.fromFile) {
-            error("--from-file is required");
-            process.exit(1);
-        }
-        let payload;
-        try {
-            const content = readFileSync(options.fromFile, "utf-8");
-            const tomlData = TOML.parse(content);
-            const workflow = tomlData.workflow || tomlData;
-            payload = {
-                steps: tomlData.steps || [],
-                inputSchema: workflow.inputSchema ? JSON.parse(workflow.inputSchema) : undefined,
-                outputSchema: workflow.outputSchema ? JSON.parse(workflow.outputSchema) : undefined,
-            };
-        }
-        catch (err) {
-            error(`Failed to read TOML file: ${err.message}`);
-            process.exit(1);
-        }
         const client = new ApiClient();
         try {
-            const result = await client.updateWorkflowDraft(resolvedAppId, workflowId, payload);
+            const result = await client.enableWorkflow(resolvedAppId, workflowId);
             if (options.json) {
                 json(result);
                 return;
             }
-            success("Draft updated.");
-            keyValue("Steps", result.draft?.steps?.length || 0);
+            success(`Workflow enabled: ${result.workflowKey || workflowId}`);
         }
         catch (err) {
             error(err.message);
             process.exit(1);
         }
     });
-    // Publish workflow
+    // #2907 — the delete flow's CLI spelling. Not a second writer of `status`:
+    // this is the plain `DELETE` #2887 made retire-with-history, which until now
+    // only the admin API and the console could reach. Named `archive`, not
+    // `delete`, because a configuration noun gets no `delete` verb
+    // (docs/cli-design.md) and prune-by-push stays the only CLI hard-delete path.
     workflows
-        .command("publish")
-        .description("Publish the current draft as a new revision")
-        .argument("<workflow-id>", "Workflow ID")
+        .command("archive")
+        .description("Retire a workflow (soft delete). Delete-lifecycle, not `disable`: the definition is kept so its history still resolves, it goes on holding its key, and there is no un-archive.")
+        .argument("<workflow-id>", "Workflow ID or key")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
+        .option("-y, --yes", "Skip confirmation prompt")
         .option("--json", "Output as JSON")
+        .addHelpText("after", `
+What archiving a workflow costs (retire-with-history):
+  New runs are refused, and so is every \`workflow.call\` into it.
+  Its runs, revisions, configurations and test cases keep resolving what they
+  point at — that is what the tombstone exists for.
+  The archived row still holds its workflowKey.
+  \`primitive workflows enable\` refuses an archived workflow; there is no
+  un-archive.
+
+Reclaiming the key means hard-deleting the row and its cascade:
+  $ rm workflows/<key>.toml
+  $ primitive config push --prune     # confirmed; hard-deletes the row
+  # or hard-delete it in the console, then re-add the file and push.
+`)
         .action(async (workflowId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
+        if (!options.yes) {
+            let confirmed;
+            try {
+                confirmed = await confirmPrompt(`Archive workflow ${workflowId}? New runs are refused, its history is ` +
+                    `kept, it goes on holding its key, and there is no un-archive.`);
+            }
+            catch (err) {
+                error(err.message);
+                process.exit(1);
+            }
+            if (!confirmed) {
+                info("Cancelled.");
+                return;
+            }
+        }
         const client = new ApiClient();
         try {
-            const result = await client.publishWorkflow(resolvedAppId, workflowId);
+            // The soft delete: no `hard`, so the #1238 cascade does not run and the
+            // definition row stays queryable.
+            const result = await client.deleteWorkflow(resolvedAppId, workflowId);
             if (options.json) {
                 json(result);
                 return;
             }
-            success("Workflow published.");
-            keyValue("Revision ID", result.revision?.revisionId);
-            keyValue("Published At", formatDate(result.revision?.publishedAt));
+            success(`Workflow archived: ${workflowId}`);
+            info("Runs, revisions and test cases keep resolving. The row still holds its key — reclaim it by removing its file and running a confirmed `primitive config push --prune`.");
         }
         catch (err) {
             error(err.message);
@@ -363,14 +570,20 @@ Examples:
     workflows
         .command("preview")
         .description("Run a preview execution of the workflow")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow-id>", "Workflow ID or key")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--config <config-id>", "Use specific configuration (uses draft if not specified)")
+        .option("--config <config-id>", "Use specific configuration")
+        .option("--draft", "Force preview of the draft version, even if active is newer")
+        .option("--active", "Force preview of the active config, even if a newer draft exists (issue #687)")
         .option("--input <json>", "Root input as JSON")
         .option("--wait", "Wait for completion and show result")
         .option("--json", "Output as JSON")
         .action(async (workflowId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
+        if (options.draft && options.active) {
+            error("--draft and --active are mutually exclusive");
+            process.exit(1);
+        }
         let rootInput;
         if (options.input) {
             try {
@@ -386,7 +599,15 @@ Examples:
             const result = await client.previewWorkflow(resolvedAppId, workflowId, {
                 rootInput,
                 configId: options.config,
+                useDraft: options.draft || false,
+                // Issue #687: --active forces the active config even when a newer
+                // draft exists (the inverse of --draft).
+                preferActive: options.active || false,
             });
+            // Display warning from the server (e.g. "previewing draft because newer than active")
+            if (result.warning) {
+                warn(result.warning);
+            }
             if (!options.wait) {
                 if (options.json) {
                     json(result);
@@ -394,26 +615,77 @@ Examples:
                 }
                 success("Preview started.");
                 keyValue("Instance ID", result.instanceId);
+                // Issue #687: name the side we ran so the user always knows which
+                // version produced the output (the "print the source" pattern).
+                const source = result.source;
+                if (source === "draft") {
+                    info("Previewing draft version.");
+                }
+                else if (source === "active") {
+                    // Issue #687 (review feedback): when configName is missing,
+                    // fall back to a short ID prefix so users staging multiple
+                    // unnamed configs can disambiguate.
+                    const label = formatConfigSlotLabel("active config", result.configName, result.configId);
+                    info(`Previewing ${label}.`);
+                }
+                else if (source === "config") {
+                    const label = formatConfigSlotLabel("the requested config", result.configName, result.configId);
+                    info(`Previewing ${label}.`);
+                }
                 info("Use 'workflows runs status' to check progress.");
                 return;
             }
-            // Poll for completion
+            // Poll for completion. Issue #941: track whether the loop broke on a
+            // terminal state vs. ran to exhaustion, so a timed-out wait is
+            // distinguishable from a workflow that legitimately reported "running".
             info("Waiting for completion...");
             let status;
-            const maxAttempts = 60;
+            // Fixed ~60s cap (60 × 1s sleep + RTT). PRIMITIVE_PREVIEW_WAIT_ATTEMPTS
+            // overrides the iteration count for tests that exercise the
+            // poll-loop-exhaustion path without a full 60s wait; unset in normal use.
+            const maxAttempts = (() => {
+                const raw = process.env.PRIMITIVE_PREVIEW_WAIT_ATTEMPTS;
+                const parsed = raw ? parseInt(raw, 10) : NaN;
+                return Number.isInteger(parsed) && parsed > 0 ? parsed : 60;
+            })();
+            let reachedTerminal = false;
             for (let i = 0; i < maxAttempts; i++) {
                 await new Promise((r) => setTimeout(r, 1000));
                 const statusResult = await client.getPreviewStatus(resolvedAppId, workflowId, result.instanceId);
                 status = statusResult.status;
-                if (status?.status === "completed" || status?.status === "failed") {
+                const s = status?.status;
+                // Issue #941 (Codex P2): an aborted preview reports
+                // status:"terminated" from the admin preview-status endpoint, which
+                // is a real terminal state — the wait did NOT time out. The shared
+                // predicate includes it so the loop breaks and `waitTimedOut` stays
+                // false (vs. the prior check that omitted "terminated" and let the
+                // loop run to the cap, misreporting an abort as a timeout).
+                if (isPreviewWaitTerminalStatus(s)) {
+                    reachedTerminal = true;
                     break;
                 }
             }
+            // Issue #941: on poll-loop exhaustion with a still-non-terminal run,
+            // emit an explicit timeout signal and exit non-zero (124, the GNU
+            // `timeout` convention) instead of silently returning the bare
+            // non-terminal status as if the wait succeeded.
+            const waitTimedOut = !reachedTerminal;
             if (options.json) {
-                json({ instanceId: result.instanceId, status });
+                // Additive field only — `status` is preserved so existing parsers
+                // don't break; `waitTimedOut` lets automation tell "still running,
+                // wait gave up" from "finished".
+                json({ instanceId: result.instanceId, status, waitTimedOut });
+                if (waitTimedOut) {
+                    process.exit(124);
+                }
                 return;
             }
-            if (status?.status === "completed") {
+            // #2348: this render ladder used to re-derive terminality (and matched
+            // the raw "complete"/"errored" spellings the server no longer emits).
+            // It now branches on the same canonical values the poll loop stops on,
+            // via the shared sets in lib/run-status.ts.
+            const waitOutcome = previewWaitOutcome(status?.status);
+            if (waitOutcome === "completed") {
                 success("Preview completed.");
                 if (status.output) {
                     console.log("\nOutput:");
@@ -427,15 +699,45 @@ Examples:
                     });
                 }
             }
-            else if (status?.status === "failed") {
+            else if (waitOutcome === "failed") {
                 error("Preview failed.");
                 if (status.error) {
                     console.log("\nError:");
                     console.log(JSON.stringify(status.error, null, 2));
                 }
             }
+            else if (waitOutcome === "terminated") {
+                // Issue #941 (Codex P2): the preview was aborted (e.g. via
+                // `workflows preview ... abort`, or an external terminate). This is a
+                // real terminal state — NOT a timeout — so render it distinctly and
+                // exit 0 (the wait succeeded in observing a terminal result).
+                warn("Preview was terminated (aborted before completion).");
+            }
+            else if (waitOutcome === "skipped") {
+                // #2636 — the preview's declarative lock was held and the definition
+                // declared `onContention: "ignore"`, so it never ran. That is an
+                // outcome, not a timeout: report it and exit 0.
+                warn(`Preview did not run — lock held${status?.skipReason ? ` (${status.skipReason})` : ""}.`);
+            }
+            else if (waitOutcome === "apply") {
+                // #2348: the run finished and is parked waiting for a client to claim
+                // and apply its result. It will not advance on its own, so the wait
+                // stopping here is a success, not a timeout.
+                warn(`Preview finished and is waiting to be applied (status: ${status.status}).`);
+                if (status.output) {
+                    console.log("\nOutput:");
+                    console.log(JSON.stringify(status.output, null, 2));
+                }
+            }
+            else if (waitOutcome === "missing") {
+                // #2348: no live instance to observe — reporting a timeout here would
+                // suggest the preview is still running, which it is not.
+                error("Preview instance not found (status: missing).");
+            }
             else {
-                info("Preview still running. Check status with 'workflows runs status'.");
+                // Timed out without ever reaching a terminal state (issue #941).
+                error(`Wait timed out after ${maxAttempts}s; preview still running. Check status with 'workflows runs status'.`);
+                process.exit(124);
             }
         }
         catch (err) {
@@ -443,51 +745,400 @@ Examples:
             process.exit(1);
         }
     });
+    // #1367 — hidden queue-latency measurement tool. Fires a batch of runs of a
+    // (trivial, no-op) workflow, polls until each has an executionStartedAt, and
+    // reports the observed queueDelayMs / createCallDurationMs distribution. This
+    // is the post-deploy verification vehicle: local miniflare does not exhibit
+    // real Cloudflare scheduling delay, so run it against an agents/alpha scratch
+    // app. Creates are paced (default 10/s) to stay clear of CF's ~100/s shared
+    // instance-creation limit; a create failure (e.g. 429) is recorded, not fatal.
+    workflows
+        .command("queue-latency-test", { hidden: true })
+        .description("(hidden) Fire N no-op runs and report the queue-delay distribution (post-deploy tool)")
+        .argument("<workflow-id>", "Workflow ID or key of a trivial no-op workflow")
+        .option("--app <app-id>", "App ID (uses current app if not specified)")
+        .option("--count <n>", "Number of runs to fire", "50")
+        .option("--rate <r>", "Max creates per second (<=10 recommended)", "10")
+        .option("--input <json>", "Root input as JSON")
+        .option("--timeout <sec>", "Seconds to wait for all runs to start", "120")
+        .option("--verbose", "Print a per-run row")
+        .option("--json", "Output as JSON")
+        .action(async (workflowId, options) => {
+        const resolvedAppId = resolveAppId(undefined, options);
+        const client = new ApiClient();
+        const count = Math.max(1, parseInt(options.count, 10) || 50);
+        const rate = Math.max(1, parseInt(options.rate, 10) || 10);
+        const timeoutSec = Math.max(1, parseInt(options.timeout, 10) || 120);
+        const intervalMs = Math.ceil(1000 / rate);
+        const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
+        let rootInput;
+        if (options.input) {
+            try {
+                rootInput = JSON.parse(options.input);
+            }
+            catch {
+                error("Invalid JSON in --input");
+                process.exit(1);
+            }
+        }
+        const fired = [];
+        if (!options.json) {
+            info(`Firing ${count} run(s) of ${workflowId} at up to ${rate}/s (interval ${intervalMs}ms)...`);
+        }
+        for (let i = 0; i < count; i++) {
+            try {
+                const res = await client.previewWorkflow(resolvedAppId, workflowId, {
+                    rootInput,
+                });
+                fired.push({
+                    index: i,
+                    instanceId: res?.instanceId,
+                    createFailed: false,
+                    started: false,
+                });
+            }
+            catch (err) {
+                // A create failure (e.g. HTTP 500 from CF's shared-class rate limit) is
+                // a recorded datapoint, not a reason to abort the whole batch.
+                fired.push({
+                    index: i,
+                    createFailed: true,
+                    createError: err?.message ? String(err.message) : String(err),
+                    started: false,
+                });
+            }
+            if (i < count - 1)
+                await sleep(intervalMs);
+        }
+        // Poll the run list until every non-failed run has executionStartedAt set
+        // (queueDelayMs present) or we hit the timeout.
+        const byInstance = new Map();
+        for (const f of fired) {
+            if (f.instanceId)
+                byInstance.set(f.instanceId, f);
+        }
+        const deadline = Date.now() + timeoutSec * 1000;
+        const pending = () => fired.filter((f) => !f.createFailed && !f.started).length;
+        // The admin run-list endpoint caps each page at 200, so a single large
+        // request cannot cover a batch of more than 200 runs — the earliest ones
+        // would never be matched and would be falsely reported as not started.
+        // Page through with the cursor (runs come back newest-first, so our batch
+        // is at the front) until every pending run is matched or we run out of
+        // pages. Bound the scan so we never walk the whole run history.
+        const PAGE_LIMIT = 200;
+        const scanBudget = count + PAGE_LIMIT;
+        while (pending() > 0 && Date.now() < deadline) {
+            await sleep(1000);
+            try {
+                let cursor;
+                let scanned = 0;
+                do {
+                    const { items, cursor: nextCursor } = await client.listWorkflowRuns(resolvedAppId, workflowId, { limit: PAGE_LIMIT, cursor: cursor ?? undefined });
+                    for (const run of items || []) {
+                        const f = run?.instanceId ? byInstance.get(run.instanceId) : undefined;
+                        if (f && !f.started && run.executionStartedAt) {
+                            f.started = true;
+                            f.queueDelayMs =
+                                typeof run.queueDelayMs === "number" ? run.queueDelayMs : null;
+                            f.createCallDurationMs =
+                                typeof run.createCallDurationMs === "number"
+                                    ? run.createCallDurationMs
+                                    : null;
+                        }
+                    }
+                    scanned += items?.length || 0;
+                    cursor = nextCursor;
+                } while (pending() > 0 && cursor && scanned < scanBudget);
+            }
+            catch {
+                // Transient list error — keep polling until the deadline.
+            }
+        }
+        const createFailures = fired.filter((f) => f.createFailed).length;
+        const startedRuns = fired.filter((f) => f.started);
+        const queueDelays = startedRuns.map((f) => f.queueDelayMs);
+        const createDurations = startedRuns.map((f) => f.createCallDurationMs);
+        const queueDist = summarizeLatencyDistribution(queueDelays);
+        const createDist = summarizeLatencyDistribution(createDurations);
+        const notStarted = fired.filter((f) => !f.createFailed && !f.started).length;
+        if (options.json) {
+            json({
+                workflowId,
+                requested: count,
+                created: count - createFailures,
+                createFailures,
+                started: startedRuns.length,
+                notStartedWithinTimeout: notStarted,
+                queueDelayMs: queueDist,
+                createCallDurationMs: createDist,
+                runs: options.verbose
+                    ? fired.map((f) => ({
+                        index: f.index,
+                        instanceId: f.instanceId ?? null,
+                        createFailed: f.createFailed,
+                        createError: f.createError ?? null,
+                        started: f.started,
+                        queueDelayMs: f.queueDelayMs ?? null,
+                        createCallDurationMs: f.createCallDurationMs ?? null,
+                    }))
+                    : undefined,
+            });
+            return;
+        }
+        divider();
+        keyValue("Requested", count);
+        keyValue("Created", count - createFailures);
+        keyValue("Create failures", createFailures);
+        keyValue("Started", startedRuns.length);
+        keyValue("Not started (timeout)", notStarted);
+        divider();
+        if (queueDist) {
+            info("Queue delay (executionStartedAt - requested):");
+            keyValue("  min", formatDuration(queueDist.min));
+            keyValue("  p50", formatDuration(queueDist.p50));
+            keyValue("  p95", formatDuration(queueDist.p95));
+            keyValue("  max", formatDuration(queueDist.max));
+        }
+        else {
+            warn("No runs started — no queue-delay samples collected.");
+        }
+        if (createDist) {
+            divider();
+            info("Create-call duration:");
+            keyValue("  min", formatDuration(createDist.min));
+            keyValue("  p50", formatDuration(createDist.p50));
+            keyValue("  p95", formatDuration(createDist.p95));
+            keyValue("  max", formatDuration(createDist.max));
+        }
+        if (options.verbose) {
+            divider();
+            console.log(formatTable(fired.map((f) => ({
+                run: f.index,
+                instanceId: f.instanceId ?? "-",
+                status: f.createFailed ? "create-failed" : f.started ? "started" : "pending",
+                queueDelayMs: f.queueDelayMs,
+                createCallDurationMs: f.createCallDurationMs,
+            })), [
+                { header: "RUN", key: "run" },
+                { header: "INSTANCE", key: "instanceId", format: formatId },
+                { header: "STATE", key: "status" },
+                { header: "DELAY", key: "queueDelayMs", format: formatDuration },
+                { header: "CREATE", key: "createCallDurationMs", format: formatDuration },
+            ]));
+        }
+    });
     // Runs subcommand
     const runs = workflows.command("runs").description("Manage workflow runs");
     // List runs
+    const runsTableColumns = [
+        { header: "RUN ID", key: "runId", format: formatId },
+        { header: "STATUS", key: "status", format: formatStatus },
+        { header: "STARTED", key: "startedAt", format: formatDate },
+        // #1367 — queue delay (requested → execution start).
+        { header: "DELAY", key: "queueDelayMs", format: formatDuration },
+        { header: "ENDED", key: "endedAt", format: formatDate },
+        { header: "PREVIEW", key: "isPreview", format: (v) => (v ? "yes" : "") },
+    ];
     runs
         .command("list")
         .description("List workflow runs")
-        .argument("<workflow-id>", "Workflow ID")
+        // #1967 — the workflow id is optional when `--user-id` is given: that
+        // form lists one user's runs across every workflow.
+        .argument("[workflow-id]", "Workflow ID or key (optional with --user-id)")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--status <status>", "Filter by status: pending, running, completed, failed")
+        // #2348: the canonical run-status vocabulary. The old text advertised
+        // "pending", which no server path ever writes, and omitted "terminated".
+        // An unrecognised value is now a 400 from the server, not an empty page.
+        .option("--status <status>", `Filter by status: ${RUN_STATUS_LIST}`)
+        .option("--user-id <user-id>", "List runs started by this user, across every workflow")
         .option("--limit <n>", "Number of runs to show", "20")
+        .option("--cursor <cursor>", "Pagination cursor from a previous page")
+        // #1969 — shared tailing flags. `--watch` re-renders the current list each
+        // interval (snapshot); `--follow` appends new/changed runs (tail). Mutually
+        // exclusive; `--interval` floor is 1s (default 2s).
+        .option("--watch", "Re-render the current run list every interval (snapshot)")
+        .option("--follow", "Append runs as they start or change (tail)")
+        .option("--interval <seconds>", "Poll interval in seconds (min 1, default 2)")
         .option("--json", "Output as JSON")
         .action(async (workflowId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
-        try {
-            const { items } = await client.listWorkflowRuns(resolvedAppId, workflowId, {
-                status: options.status,
-                limit: parseInt(options.limit),
-            });
-            if (options.json) {
-                json(items);
-                return;
+        const limit = parseInt(options.limit);
+        if (options.watch && options.follow) {
+            error("--watch and --follow are mutually exclusive; pass only one.");
+            process.exit(1);
+        }
+        // #1967 — one of the two selectors is required; `list` never enumerates
+        // every run in the app.
+        if (!workflowId && !options.userId) {
+            error("Provide a workflow ID, or --user-id to list one user's runs across every workflow.");
+            process.exit(1);
+        }
+        // The user view reads the runs-by-user index, which has no resume
+        // contract — `--follow`'s checkpoint is keyed on the by-workflow index.
+        if (options.userId && options.follow) {
+            error("--follow is not supported with --user-id (it reads a different index). Use --watch instead.");
+            process.exit(1);
+        }
+        // Both selectors resolve through the same page shape, so the rest of the
+        // command is identical either way.
+        const fetchPage = (params) => options.userId
+            ? client.listUserWorkflowRuns(resolvedAppId, options.userId, {
+                ...params,
+                // With both a workflow id and --user-id, narrow to that workflow.
+                workflowId: workflowId || undefined,
+            })
+            : client.listWorkflowRuns(resolvedAppId, workflowId, params);
+        // One-shot list (no tailing).
+        if (!options.watch && !options.follow) {
+            try {
+                const page = await fetchPage({
+                    status: options.status,
+                    limit,
+                    cursor: options.cursor,
+                });
+                const { items } = page;
+                if (options.json) {
+                    // #1316 envelope with the shared inspection item shape — never a
+                    // bare array — preserving nextCursor/hasMore + the deprecated
+                    // `cursor` dual-emit alias (behaviors 6, 7).
+                    json(buildWorkflowRunEnvelope(items, page));
+                    return;
+                }
+                // The cursor hint prints whether or not this page had rows: a
+                // filtered read can come back empty with more of the index still to
+                // walk, and a bare "No runs found." would read as "this user has no
+                // matching runs" (#1967 review).
+                const printCursorHint = () => {
+                    if (!page.nextCursor)
+                        return;
+                    console.log();
+                    dim(`More runs. Use --cursor ${page.nextCursor} for the next page.`);
+                };
+                if (!items || items.length === 0) {
+                    // #2237 — the honest empty state: a bounded filtered read can stop
+                    // before it reaches a match, so the message says how far it got and
+                    // how to continue instead of asserting there are none. `runs list`
+                    // deliberately does NOT auto-resume like `runs failures` — it is a
+                    // pager, and its `--limit` is a page size.
+                    info(describeEmptyRunPage(page, !!options.cursor));
+                    return;
+                }
+                console.log(formatTable(items, runsTableColumns));
+                printCursorHint();
             }
-            if (!items || items.length === 0) {
-                info("No runs found.");
-                return;
+            catch (err) {
+                error(err.message);
+                process.exit(1);
             }
-            console.log(formatTable(items, [
-                { header: "RUN ID", key: "runId", format: formatId },
-                { header: "STATUS", key: "status", format: formatStatus },
-                { header: "STARTED", key: "startedAt", format: formatDate },
-                { header: "ENDED", key: "endedAt", format: formatDate },
-                { header: "PREVIEW", key: "isPreview", format: (v) => v ? "yes" : "" },
-            ]));
+            return;
+        }
+        // Tailing (`--watch` / `--follow`). SIGINT → AbortController installed
+        // here at the command level; the handler is removed in `finally` so it
+        // never leaks past this command (unlike the persistent process.on).
+        let intervalSeconds;
+        try {
+            intervalSeconds = parseIntervalSeconds(options.interval);
         }
         catch (err) {
             error(err.message);
             process.exit(1);
         }
+        const intervalMs = intervalSeconds * 1000;
+        const controller = new AbortController();
+        const onSigint = () => controller.abort();
+        process.on("SIGINT", onSigint);
+        try {
+            if (options.follow) {
+                const source = {
+                    fetchAfter: async (resumeAfter) => {
+                        if (resumeAfter === undefined) {
+                            // Bootstrap: newest position, descending, one row. followLoop
+                            // discards the items and keeps only the baseline checkpoint.
+                            const page = await client.listWorkflowRuns(resolvedAppId, workflowId, {
+                                follow: true,
+                                forward: false,
+                                limit: 1,
+                                status: options.status,
+                            });
+                            return { items: page.items, hasMore: false, resumeAfter: page.resumeAfter };
+                        }
+                        const page = await client.listWorkflowRuns(resolvedAppId, workflowId, {
+                            follow: true,
+                            forward: true,
+                            after: resumeAfter,
+                            limit,
+                            status: options.status,
+                        });
+                        return {
+                            items: page.items,
+                            hasMore: !!page.hasMore,
+                            resumeAfter: page.resumeAfter,
+                        };
+                    },
+                };
+                if (!options.json) {
+                    info("Following runs (Ctrl-C to stop)…");
+                }
+                await followLoop(source, {
+                    intervalMs,
+                    signal: controller.signal,
+                    render: (items) => {
+                        if (options.json) {
+                            // One JSON object per line (a tail has no closing bracket to
+                            // wait for), each in the shared inspection item shape — the
+                            // same items the one-shot envelope carries (behavior 7).
+                            for (const run of items)
+                                jsonLine(toWorkflowRunInspectionRow(run));
+                            return;
+                        }
+                        for (const run of items) {
+                            console.log(`${formatDate(run.startedAt)}  ${formatStatus(run.status)}  ${formatId(run.runId)}`);
+                        }
+                    },
+                });
+            }
+            else {
+                // --watch: periodic snapshot redraw. The most recent page's
+                // pagination fields are kept so the `--json` redraw emits the same
+                // envelope the one-shot path does, not a bare array (behavior 7).
+                let lastPage = {};
+                await watchLoop(async () => {
+                    const page = await fetchPage({
+                        status: options.status,
+                        limit,
+                        cursor: options.cursor,
+                    });
+                    lastPage = page;
+                    return page.items;
+                }, {
+                    intervalMs,
+                    signal: controller.signal,
+                    render: (items) => {
+                        if (options.json) {
+                            json(buildWorkflowRunEnvelope(items, lastPage));
+                            return;
+                        }
+                        // Renderer owns the redraw: clear the screen, then print.
+                        process.stdout.write("\x1b[2J\x1b[H");
+                        if (!items || items.length === 0) {
+                            console.log("No runs found.");
+                            return;
+                        }
+                        console.log(formatTable(items, runsTableColumns));
+                    },
+                });
+            }
+        }
+        finally {
+            process.removeListener("SIGINT", onSigint);
+        }
     });
     // Run status
     runs
         .command("status")
         .description("Get status of a workflow run")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow-id>", "Workflow ID or key")
         .argument("<run-id>", "Run ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--json", "Output as JSON")
@@ -496,89 +1147,137 @@ Examples:
         const client = new ApiClient();
         try {
             const result = await client.getWorkflowRunStatus(resolvedAppId, workflowId, runId);
+            // #2771 — a Cloudflare instance reports `stepResults` only once the run
+            // is terminal, so a live run used to answer with `status: running` and
+            // nothing else for its whole duration. Step rows are now written as the
+            // run goes, so fall back to them: the steps finished so far plus the one
+            // currently executing. Resolved BEFORE the output branches — a script
+            // polling `--json` needs the in-flight trace as much as a human does.
+            // Best-effort: a lookup failure still leaves the run details intact.
+            const instanceStepResults = result.instanceStatus?.stepResults;
+            let persistedStepRuns = [];
+            if (!instanceStepResults || instanceStepResults.length === 0) {
+                try {
+                    const stepRuns = await client.getWorkflowStepRuns(resolvedAppId, workflowId, runId);
+                    persistedStepRuns = stepRuns?.items || [];
+                }
+                catch {
+                    persistedStepRuns = [];
+                }
+            }
             if (options.json) {
-                json(result);
+                // Purely additive: `stepRuns` appears only when the persisted rows are
+                // what carries the trace, so a terminal run's JSON is unchanged.
+                json(persistedStepRuns.length > 0 ? { ...result, stepRuns: persistedStepRuns } : result);
                 return;
             }
             const run = result.run;
             keyValue("Run ID", run.runId);
             keyValue("Status", formatStatus(run.status));
-            keyValue("Started", formatDate(run.startedAt));
+            // #2636 — the machine-readable outcome next to the status: why a
+            // failed run failed, or why an elided run never ran.
+            if (run.errorCode)
+                keyValue("Error code", run.errorCode);
+            if (run.skipReason)
+                keyValue("Skip reason", run.skipReason);
+            // #1367 — `startedAt` is the request time; `executionStartedAt` is when
+            // Cloudflare actually started running the instance. Queue delay is the
+            // gap. A still-queued run shows "-" for execution start / queue delay.
+            keyValue("Requested", formatDate(run.startedAt));
+            keyValue("Execution started", formatDate(run.executionStartedAt));
+            keyValue("Queue delay", formatDuration(run.queueDelayMs));
+            keyValue("Create call", formatDuration(run.createCallDurationMs));
             keyValue("Ended", formatDate(run.endedAt));
             keyValue("Preview", run.isPreview ? "yes" : "no");
-            if (result.instanceStatus) {
-                const status = result.instanceStatus;
-                if (status.output) {
-                    divider();
-                    info("Output:");
-                    console.log(JSON.stringify(status.output, null, 2));
-                }
-                if (status.stepResults && status.stepResults.length > 0) {
-                    divider();
-                    info("Step Results:");
-                    status.stepResults.forEach((step) => {
-                        console.log(`  ${step.id}: ${formatDuration(step.durationMs)} ${step.skipped ? "(skipped)" : ""}`);
-                    });
-                }
-                if (status.error) {
-                    divider();
-                    error("Error:");
-                    console.log(JSON.stringify(status.error, null, 2));
+            const stepResultLines = renderRunStatusStepResults(stepResultsForRunStatus(instanceStepResults, persistedStepRuns));
+            if (result.instanceStatus?.output) {
+                divider();
+                info("Output:");
+                console.log(JSON.stringify(result.instanceStatus.output, null, 2));
+            }
+            if (stepResultLines.length > 0) {
+                divider();
+                info("Step Results:");
+                for (const line of stepResultLines) {
+                    console.log(line);
                 }
             }
+            if (result.instanceStatus?.error) {
+                divider();
+                error("Error:");
+                console.log(JSON.stringify(result.instanceStatus.error, null, 2));
+            }
         }
         catch (err) {
             error(err.message);
             process.exit(1);
         }
     });
-    // ============================================
-    // CONFIGS SUBCOMMAND
-    // ============================================
-    const configs = workflows
-        .command("configs")
-        .description("Manage workflow configurations")
-        .addHelpText("after", `
-Examples:
-  $ primitive workflows configs list <workflow-id>
-  $ primitive workflows configs get <workflow-id> <config-id>
-  $ primitive workflows configs create <workflow-id> --name "v2" --from-file workflow.toml
-  $ primitive workflows configs activate <workflow-id> <config-id>
-  $ primitive workflows configs duplicate <workflow-id> <config-id> --name "v2-copy"
-`);
-    // List configs
-    configs
-        .command("list")
-        .description("List configurations for a workflow")
-        .argument("<workflow-id>", "Workflow ID")
-        .option("--app <app-id>", "App ID (uses current app if not specified)")
+    // Steps for a run
+    runs
+        .command("steps")
+        .description("Show step-level details for a workflow run")
+        .argument("<workflow-id>", "Workflow ID or key")
+        .argument("<run-id>", "Run ID")
+        .option("--app <app-id>", "App ID")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, options) => {
+        .action(async (workflowId, runId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
-            const { items } = await client.listWorkflowConfigs(resolvedAppId, workflowId);
+            const { items } = await client.getWorkflowStepRuns(resolvedAppId, workflowId, runId);
             if (options.json) {
-                json(items);
+                // Steps are unpaginated (one run's full trace) — the shared
+                // inspection item shape wrapped in `{items}`, no cursor (behavior 6).
+                json(buildWorkflowStepEnvelope(items));
                 return;
             }
             if (!items || items.length === 0) {
-                info("No configurations found.");
+                info("No step runs found.");
                 return;
             }
-            // Get workflow to find active config
-            const { workflow } = await client.getWorkflow(resolvedAppId, workflowId);
-            const activeConfigId = workflow?.activeConfigId;
-            console.log(formatTable(items, [
-                { header: "ID", key: "configId", format: formatId },
-                { header: "NAME", key: "configName" },
-                { header: "STATUS", key: "status", format: formatStatus },
+            // #1367 — inter-step gap: idle time between one step ending and the
+            // next starting. Only surfaced above a 1s threshold so normal
+            // back-to-back steps stay quiet; a large gap flags a mid-run stall.
+            const STEP_GAP_THRESHOLD_MS = 1000;
+            // #2771 — an in-flight run also returns the step executing right now
+            // (status `running`, duration not written yet). Show its elapsed time
+            // so a step stuck retrying is visible as such.
+            const rows = annotateStepGaps(items).map((row) => ({
+                ...row,
+                displayDurationMs: stepRunDisplayDurationMs(row),
+            }));
+            console.log(formatTable(rows, [
+                { header: "STEP", key: "stepId" },
+                { header: "KIND", key: "stepKind" },
                 {
-                    header: "ACTIVE",
-                    key: "configId",
-                    format: (v) => v === activeConfigId ? "★" : "",
+                    header: "STATUS",
+                    key: "status",
+                    format: (v) => v === "completed" ? chalk.green(v) :
+                        v === "failed" ? chalk.red(v) :
+                            v === "skipped" ? chalk.gray(v) :
+                                v === "error_captured" ? chalk.yellow(v) :
+                                    // #2771 — the step executing right now.
+                                    v === "running" ? chalk.cyan(v) : v,
+                },
+                {
+                    header: "GAP",
+                    key: "gapMs",
+                    format: (v) => typeof v === "number" && v >= STEP_GAP_THRESHOLD_MS
+                        ? chalk.yellow(formatDuration(v))
+                        : chalk.dim("-"),
+                },
+                { header: "DURATION", key: "displayDurationMs", format: formatDuration },
+                {
+                    header: "TOKENS",
+                    key: "totalTokens",
+                    format: (v) => v ? String(v).replace(/\B(?=(\d{3})+(?!\d))/g, ",") : "-",
+                },
+                {
+                    header: "ERROR",
+                    key: "error",
+                    format: (v) => v ? (v.length > 60 ? v.slice(0, 57) + "..." : v) : "",
                 },
-                { header: "MODIFIED", key: "modifiedAt", format: formatDate },
             ]));
         }
         catch (err) {
@@ -586,220 +1285,524 @@ Examples:
             process.exit(1);
         }
     });
-    // Get config
-    configs
-        .command("get")
-        .description("Get configuration details")
-        .argument("<workflow-id>", "Workflow ID")
-        .argument("<config-id>", "Configuration ID")
-        .option("--app <app-id>", "App ID (uses current app if not specified)")
+    // Step detail
+    runs
+        .command("step-detail")
+        .description("Show full details for a single step")
+        .argument("<workflow-id>", "Workflow ID or key")
+        .argument("<run-id>", "Run ID")
+        .argument("<step-id>", "Step ID")
+        .option("--app <app-id>", "App ID")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, configId, options) => {
+        .action(async (workflowId, runId, stepId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
-            const config = await client.getWorkflowConfig(resolvedAppId, workflowId, configId);
+            const { items } = await client.getWorkflowStepRuns(resolvedAppId, workflowId, runId);
+            const step = items.find((s) => s.stepId === stepId);
+            if (!step) {
+                error(`Step "${stepId}" not found in run ${runId}`);
+                process.exit(1);
+            }
             if (options.json) {
-                json(config);
+                json(step);
                 return;
             }
-            keyValue("Config ID", config.configId);
-            keyValue("Name", config.configName);
-            keyValue("Description", config.description || "-");
-            keyValue("Status", formatStatus(config.status));
-            keyValue("Created", formatDate(config.createdAt));
-            keyValue("Modified", formatDate(config.modifiedAt));
-            if (config.steps && config.steps.length > 0) {
+            keyValue("Step", step.stepId);
+            keyValue("Kind", step.stepKind);
+            keyValue("Status", formatStatus(step.status));
+            keyValue("Duration", formatDuration(step.durationMs));
+            keyValue("Started", formatDate(step.startedAt));
+            keyValue("Ended", formatDate(step.endedAt));
+            if (step.totalTokens) {
+                keyValue("Tokens", `${step.inputTokens || 0} in / ${step.outputTokens || 0} out / ${step.totalTokens} total`);
+            }
+            if (step.retryCount) {
+                keyValue("Retries", String(step.retryCount));
+            }
+            if (step.config) {
                 divider();
-                info(`Steps (${config.steps.length}):`);
-                config.steps.forEach((step, idx) => {
-                    console.log(`  ${idx + 1}. ${step.id} (${step.kind})`);
+                info("Config (rendered):");
+                console.log(JSON.stringify(step.config, null, 2));
+            }
+            if (step.rawConfig) {
+                divider();
+                info("Config (raw):");
+                console.log(JSON.stringify(step.rawConfig, null, 2));
+            }
+            if (step.input) {
+                divider();
+                info("Input:");
+                console.log(JSON.stringify(step.input, null, 2));
+            }
+            if (step.output) {
+                divider();
+                info("Output:");
+                console.log(JSON.stringify(step.output, null, 2));
+            }
+            if (step.error) {
+                divider();
+                error("Error: " + step.error);
+                if (step.errorDetails) {
+                    console.log(step.errorDetails);
+                }
+            }
+            if (step.templateWarnings && step.templateWarnings.length > 0) {
+                divider();
+                warn("Template Warnings:");
+                step.templateWarnings.forEach((w) => {
+                    console.log(`  ${w.expression}: ${w.message}`);
                 });
             }
+            if (step.context) {
+                divider();
+                info("Context Snapshot:");
+                console.log(JSON.stringify(step.context, null, 2));
+            }
         }
         catch (err) {
             error(err.message);
             process.exit(1);
         }
     });
-    // Create config
-    configs
-        .command("create")
-        .description("Create a new configuration")
-        .argument("<workflow-id>", "Workflow ID")
+    // Integration call logs for a run (cross-pivot from workflow-run to
+    // integrations.logs — see issue #699).
+    runs
+        .command("logs")
+        .description("List integration calls made by a specific workflow run (cross-pivot of `integrations logs`)")
+        .argument("<run-id>", "Workflow run ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--name <name>", "Configuration name (required)")
-        .option("--description <desc>", "Description")
-        .option("--from-file <path>", "Load steps from TOML file")
+        .option("--limit <n>", "Number of logs to show", "100")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, options) => {
+        .action(async (runId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
-        if (!options.name) {
-            error("--name is required");
+        const client = new ApiClient();
+        try {
+            const logs = await client.listWorkflowRunIntegrationLogs(resolvedAppId, runId, { limit: parseInt(options.limit) });
+            if (options.json) {
+                json(logs);
+                return;
+            }
+            if (!logs || logs.length === 0) {
+                info("No integration calls found for this run.");
+                return;
+            }
+            console.log(formatTable(logs, [
+                { header: "TIME", key: "timestamp", format: formatDate },
+                { header: "STEP", key: "stepId", format: (v) => v || "" },
+                {
+                    header: "INTEGRATION",
+                    key: "integrationKey",
+                    format: (v) => v || "",
+                },
+                { header: "METHOD", key: "method" },
+                { header: "PATH", key: "path" },
+                { header: "STATUS", key: "status" },
+                { header: "DURATION", key: "durationMs", format: formatDuration },
+                { header: "TRACE", key: "traceId", format: formatId },
+            ]));
+        }
+        catch (err) {
+            error(err.message);
             process.exit(1);
         }
-        let steps = [];
-        if (options.fromFile) {
-            try {
-                const content = readFileSync(options.fromFile, "utf-8");
-                const tomlData = TOML.parse(content);
-                steps = tomlData.steps || [];
+    });
+    // Error summary
+    runs
+        .command("error")
+        .description("Show error details for a failed workflow run")
+        .argument("<workflow-id>", "Workflow ID or key")
+        .argument("<run-id>", "Run ID")
+        .option("--app <app-id>", "App ID")
+        .option("--json", "Output as JSON")
+        .action(async (workflowId, runId, options) => {
+        const resolvedAppId = resolveAppId(undefined, options);
+        const client = new ApiClient();
+        try {
+            const [runResult, stepsResult] = await Promise.all([
+                client.getWorkflowRunStatus(resolvedAppId, workflowId, runId),
+                client.getWorkflowStepRuns(resolvedAppId, workflowId, runId),
+            ]);
+            // #2074 — one selection rule shared with the STEP column of
+            // `runs failures`: prefer the run's own denormalized attribution, and
+            // fall back to the lowest-`stepIndex` failed row (never "whatever the
+            // query returned first") so the two views can't disagree.
+            const failedStep = selectFailedStepRow(stepsResult.items, runResult.run?.failedStepId);
+            const errorInfo = {
+                runId,
+                status: runResult.run?.status,
+                // #2636 — the platform's classification of the failure, when it owns
+                // one (a lost declarative-lock race, an exhausted lock wait).
+                errorCode: runResult.run?.errorCode ?? null,
+                errorMessage: runResult.run?.errorMessage || runResult.instanceStatus?.error?.message,
+                failedStep: failedStep ? {
+                    stepId: failedStep.stepId,
+                    stepKind: failedStep.stepKind,
+                    error: failedStep.error,
+                    errorDetails: failedStep.errorDetails,
+                    input: failedStep.input,
+                    config: failedStep.config,
+                } : null,
+            };
+            if (options.json) {
+                json(errorInfo);
+                return;
             }
-            catch (err) {
-                error(`Failed to read TOML file: ${err.message}`);
-                process.exit(1);
+            keyValue("Run ID", runId);
+            keyValue("Status", formatStatus(errorInfo.status));
+            if (errorInfo.errorCode) {
+                keyValue("Code", errorInfo.errorCode);
             }
+            if (errorInfo.errorMessage) {
+                divider();
+                error("Error: " + errorInfo.errorMessage);
+            }
+            if (failedStep) {
+                divider();
+                info(`Failed Step: ${failedStep.stepId} (${failedStep.stepKind})`);
+                if (failedStep.error) {
+                    error("  " + failedStep.error);
+                }
+                if (failedStep.errorDetails) {
+                    console.log(failedStep.errorDetails);
+                }
+                if (failedStep.input) {
+                    divider();
+                    info("Step Input:");
+                    console.log(JSON.stringify(failedStep.input, null, 2));
+                }
+                if (failedStep.config) {
+                    divider();
+                    info("Step Config:");
+                    console.log(JSON.stringify(failedStep.config, null, 2));
+                }
+            }
+            else {
+                info("No failed step found in step runs.");
+            }
+        }
+        catch (err) {
+            error(err.message);
+            process.exit(1);
         }
+    });
+    // Failures list
+    runs
+        .command("failures")
+        .description("List recent workflow run failures")
+        .argument("<workflow-id>", "Workflow ID or key")
+        .option("--app <app-id>", "App ID")
+        // #2237 — `--limit` is the number of FAILURES to collect, as the help text
+        // always claimed. It used to be passed straight through as the server's
+        // bounded page limit, so a workflow with a long success streak answered
+        // "No failures found." while failures sat just past the scan budget.
+        .option("--limit <n>", "Number of failures to collect", "10")
+        .option("--max-scan <n>", `Stop after examining this many runs, to the nearest request (each examines up to ${SERVER_SCAN_ROWS_PER_REQUEST}; 0 = no cap; a sweep issues at most ${FAILURE_SWEEP_MAX_REQUESTS} requests either way)`, String(DEFAULT_FAILURE_SWEEP_MAX_SCAN))
+        // `--cursor` seeds the sweep's starting position; the sweep then resumes
+        // itself, so it is only needed to continue an earlier capped search.
+        .option("--cursor <cursor>", "Resume the search from a previous cursor")
+        .option("--json", "Output as JSON")
+        .action(async (workflowId, options) => {
+        const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
+        const limit = parseInt(options.limit, 10);
+        const maxScan = parseInt(options.maxScan, 10);
+        if (!Number.isFinite(limit) || limit < 1) {
+            error("--limit must be a positive integer.");
+            process.exit(1);
+        }
+        if (!Number.isFinite(maxScan) || maxScan < 0) {
+            error("--max-scan must be a non-negative integer (0 = no cap).");
+            process.exit(1);
+        }
+        // Ctrl-C during a multi-second sweep prints what was collected plus the
+        // resume cursor rather than dropping the work on the floor. The handler
+        // is removed in `finally` so it never leaks past this command.
+        //
+        // Registering a listener replaces Node's terminate-on-SIGINT default, and
+        // `shouldStop` is only read between requests — so a first Ctrl-C during a
+        // slow in-flight request would otherwise appear to do nothing and leave
+        // the command unkillable. A second one exits immediately (#2237 review).
+        let interrupted = false;
+        let sigints = 0;
+        const onSigint = () => {
+            interrupted = true;
+            if (++sigints > 1)
+                process.exit(130);
+        };
+        process.on("SIGINT", onSigint);
+        // Progress on stderr, TTY only, never in `--json` mode: a silent
+        // multi-second command invites the Ctrl-C it is trying to survive.
+        const showProgress = !options.json && !!process.stderr.isTTY;
         try {
-            const config = await client.createWorkflowConfig(resolvedAppId, workflowId, {
-                configName: options.name,
-                description: options.description,
-                steps,
+            const sweep = await collectWorkflowFailures(({ cursor, limit: pageLimit }) => client.listWorkflowRuns(resolvedAppId, workflowId, {
+                status: "failed",
+                limit: pageLimit,
+                cursor,
+            }), {
+                limit,
+                maxScan,
+                startCursor: options.cursor,
+                shouldStop: () => interrupted,
+                onProgress: showProgress
+                    ? (scanned, collected) => progress(`Searching… ${scanned} runs examined, ${collected} failure${collected === 1 ? "" : "s"} found`)
+                    : undefined,
             });
+            if (showProgress && sweep.requests >= 1) {
+                // Clear the progress line so it does not sit above the table. One
+                // request is the common case — `onProgress` fires after EVERY request
+                // including the first, so guarding on `> 1` left the line on screen
+                // with the cursor parked at its end (#2237 review).
+                process.stderr.write("\r\x1b[K");
+            }
+            // A SIGINT that arrived after the sweep returned (while the table is
+            // rendering) did not truncate anything, so only a sweep that actually
+            // stopped on the interrupt reports 130. `exitCode` rather than
+            // `process.exit()` so a piped stdout is not truncated mid-write.
+            const exitInterrupted = sweep.stoppedBecause === "interrupted";
+            const items = sweep.items;
             if (options.json) {
-                json(config);
+                // #2074 — the shared inspection item shape inside the endpoint's own
+                // #1316 envelope, never a bare array. #2237 — `items` is the
+                // concatenation across every resumed page, `scanned` their sum, and
+                // `nextCursor` the last one observed.
+                json(buildWorkflowRunEnvelope(items, {
+                    nextCursor: sweep.nextCursor,
+                    hasMore: !!sweep.nextCursor,
+                    ...(sweep.scannedKnown ? { scanned: sweep.scanned } : {}),
+                }));
+                if (exitInterrupted)
+                    process.exitCode = 130;
                 return;
             }
-            success(`Configuration created: ${config.configName}`);
-            keyValue("Config ID", config.configId);
+            if (items.length > 0) {
+                console.log(formatTable(items, [
+                    { header: "RUN ID", key: "runId", format: formatId },
+                    // STEP sits before ERROR so the truncated message stays the last
+                    // wide column and the step id is never the thing that gets cut.
+                    // `-` is a routine outcome, not a bug: launch-time failures
+                    // (create/schema/R2 aborts, reclaimed runs) legitimately have no
+                    // failed step, and runs that predate #2074 carry no attribution.
+                    { header: "STEP", key: "failedStepId", format: formatFailureStepCell },
+                    // #2636 — the platform's classification, next to the message an
+                    // operator would otherwise have had to pattern-match.
+                    { header: "CODE", key: "errorCode", format: formatFailureCodeCell },
+                    { header: "ERROR", key: "errorMessage", format: formatFailureErrorCell },
+                    { header: "STARTED", key: "startedAt", format: formatDate },
+                    { header: "ENDED", key: "endedAt", format: formatDate },
+                ]));
+            }
+            // #2237 — replaces the unconditional "No failures found.": every form
+            // either states the search was complete or says how far it got and how
+            // to continue. A triage command must never assert an absence it did
+            // not establish.
+            const summary = describeFailureSweep(sweep, limit);
+            if (summary) {
+                if (items.length > 0)
+                    console.log();
+                info(summary);
+            }
+            if (exitInterrupted)
+                process.exitCode = 130;
         }
         catch (err) {
             error(err.message);
             process.exit(1);
         }
+        finally {
+            process.removeListener("SIGINT", onSigint);
+        }
     });
-    // Update config
-    configs
-        .command("update")
-        .description("Update a configuration")
-        .argument("<workflow-id>", "Workflow ID")
-        .argument("<config-id>", "Configuration ID")
+    // ── Iterations: inspect / reset iterate-users singleton iterations (#1209) ──
+    const iterations = workflows
+        .command("iterations")
+        .description("Inspect and reset iterate-users iterations");
+    iterations
+        .command("list")
+        .description("List iterate-users iterations for the app")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--name <name>", "Configuration name")
-        .option("--description <desc>", "Description")
-        .option("--from-file <path>", "Update steps from TOML file")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, configId, options) => {
+        .action(async (options) => {
         const resolvedAppId = resolveAppId(undefined, options);
-        const payload = {};
-        if (options.name)
-            payload.configName = options.name;
-        if (options.description !== undefined)
-            payload.description = options.description;
-        if (options.fromFile) {
-            try {
-                const content = readFileSync(options.fromFile, "utf-8");
-                const tomlData = TOML.parse(content);
-                payload.steps = tomlData.steps || [];
+        const client = new ApiClient();
+        try {
+            const { items } = await client.listIterations(resolvedAppId);
+            if (options.json) {
+                json(items);
+                return;
             }
-            catch (err) {
-                error(`Failed to read TOML file: ${err.message}`);
-                process.exit(1);
+            if (!items || items.length === 0) {
+                info("No iterations found.");
+                return;
             }
+            console.log(formatTable(items, [
+                { header: "ITERATION", key: "iterationName" },
+                { header: "STATUS", key: "status", format: formatStatus },
+                { header: "ACQUIRED", key: "lastAcquireMode", format: (v) => v || "" },
+                { header: "PROCESSED", key: "totalProcessed" },
+                { header: "FAILED", key: "failedCount" },
+                { header: "LAST RUN", key: "lastRunAt", format: formatDate },
+            ]));
         }
-        if (Object.keys(payload).length === 0) {
-            error("No update options specified. Use --name, --description, or --from-file.");
+        catch (err) {
+            error(err.message);
             process.exit(1);
         }
+    });
+    iterations
+        .command("get")
+        .description("Show details for a single iterate-users iteration")
+        .argument("<name>", "Iteration name")
+        .option("--app <app-id>", "App ID (uses current app if not specified)")
+        .option("--json", "Output as JSON")
+        .action(async (name, options) => {
+        const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
-            const config = await client.updateWorkflowConfig(resolvedAppId, workflowId, configId, payload);
+            const it = await client.getIteration(resolvedAppId, name);
             if (options.json) {
-                json(config);
+                json(it);
                 return;
             }
-            success("Configuration updated.");
-            keyValue("Config ID", config.configId);
-            keyValue("Name", config.configName);
+            keyValue("Iteration", it.iterationName);
+            keyValue("Status", formatStatus(it.status));
+            keyValue("Acquired as", it.lastAcquireMode || "(unknown)");
+            keyValue("Run ID", it.runId);
+            keyValue("Instance ID", it.instanceId);
+            keyValue("Started", formatDate(it.startedAt));
+            keyValue("Last run", formatDate(it.lastRunAt));
+            keyValue("Lock expires", formatDate(it.lockExpiresAt));
+            keyValue("Processed", String(it.totalProcessed ?? 0));
+            keyValue("Succeeded", String(it.succeededCount ?? 0));
+            keyValue("Failed (total)", String(it.failedCount ?? 0));
+            keyValue("Skipped", String(it.skippedCount ?? 0));
+            if (it.lastError) {
+                keyValue("Last error", it.lastError);
+            }
+            const sample = Array.isArray(it.failedUserIdsSample)
+                ? it.failedUserIdsSample
+                : [];
+            if (sample.length > 0) {
+                const label = it.failedUserIdsTruncated
+                    ? `Failed user IDs (sample, truncated; ${it.failedCount} total)`
+                    : `Failed user IDs (sample)`;
+                keyValue(label, sample.join(", "));
+            }
         }
         catch (err) {
             error(err.message);
             process.exit(1);
         }
     });
-    // Activate config
-    configs
-        .command("activate")
-        .description("Set a configuration as the active (default) configuration")
-        .argument("<workflow-id>", "Workflow ID")
-        .argument("<config-id>", "Configuration ID")
+    iterations
+        .command("reset")
+        .description("Reset a terminal (completed/failed) iteration so the next trigger runs fresh")
+        .argument("<name>", "Iteration name")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, configId, options) => {
+        .action(async (name, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
-            const result = await client.activateWorkflowConfig(resolvedAppId, workflowId, configId);
+            const res = await client.resetIteration(resolvedAppId, name);
             if (options.json) {
-                json(result);
+                json(res);
                 return;
             }
-            success("Configuration activated.");
-            keyValue("Active Config ID", result.activeConfigId || configId);
+            success(`Reset iteration "${name}" (was ${res.previousStatus}). The next trigger will run fresh.`);
         }
         catch (err) {
+            // A running iteration is refused with 409 + lock metadata — surface a
+            // clear message rather than a raw error.
             error(err.message);
             process.exit(1);
         }
     });
-    // Duplicate config
+    // ============================================
+    // CONFIGS SUBCOMMAND
+    // ============================================
+    const configs = workflows
+        .command("configs")
+        .description("Manage workflow configurations")
+        .addHelpText("after", `
+Examples:
+  $ primitive workflows configs list <workflow-id>
+  $ primitive workflows configs get <workflow-id> <config-id>
+
+Config bodies are authored as files, not verbs: write
+workflows/<key>.configs/<name>.toml and name the live one with
+activeConfigName in workflows/<key>.toml, then run 'primitive config push'.
+`);
+    // List configs
     configs
-        .command("duplicate")
-        .description("Duplicate a configuration")
-        .argument("<workflow-id>", "Workflow ID")
-        .argument("<config-id>", "Configuration ID to duplicate")
+        .command("list")
+        .description("List configurations for a workflow")
+        .argument("<workflow-id>", "Workflow ID or key")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--name <name>", "Name for the new configuration")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, configId, options) => {
+        .action(async (workflowId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
-            const config = await client.duplicateWorkflowConfig(resolvedAppId, workflowId, configId, {
-                configName: options.name,
-            });
+            const { items } = await client.listWorkflowConfigs(resolvedAppId, workflowId);
             if (options.json) {
-                json(config);
+                json(items);
                 return;
             }
-            success(`Configuration duplicated: ${config.configName}`);
-            keyValue("New Config ID", config.configId);
+            if (!items || items.length === 0) {
+                info("No configurations found.");
+                return;
+            }
+            // Get workflow to find active config
+            const { workflow } = await client.getWorkflow(resolvedAppId, workflowId);
+            const activeConfigId = workflow?.activeConfigId;
+            console.log(formatTable(items, [
+                { header: "ID", key: "configId", format: formatId },
+                { header: "NAME", key: "configName" },
+                { header: "STATUS", key: "status", format: formatStatus },
+                {
+                    header: "ACTIVE",
+                    key: "configId",
+                    format: (v) => v === activeConfigId ? "★" : "",
+                },
+                { header: "MODIFIED", key: "modifiedAt", format: formatDate },
+            ]));
         }
         catch (err) {
             error(err.message);
             process.exit(1);
         }
     });
-    // Archive config
+    // Get config
     configs
-        .command("archive")
-        .description("Archive a configuration")
-        .argument("<workflow-id>", "Workflow ID")
+        .command("get")
+        .description("Get configuration details")
+        .argument("<workflow-id>", "Workflow ID or key")
         .argument("<config-id>", "Configuration ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("-y, --yes", "Skip confirmation prompt")
+        .option("--json", "Output as JSON")
         .action(async (workflowId, configId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
-        if (!options.yes) {
-            const inquirer = await import("inquirer");
-            const { confirm } = await inquirer.default.prompt([
-                {
-                    type: "confirm",
-                    name: "confirm",
-                    message: `Are you sure you want to archive configuration ${configId}?`,
-                    default: false,
-                },
-            ]);
-            if (!confirm) {
-                info("Cancelled.");
-                return;
-            }
-        }
         const client = new ApiClient();
         try {
-            await client.archiveWorkflowConfig(resolvedAppId, workflowId, configId);
-            success("Configuration archived.");
+            const config = await client.getWorkflowConfig(resolvedAppId, workflowId, configId);
+            if (options.json) {
+                json(config);
+                return;
+            }
+            printResult("Config ID", config.configId);
+            printResult("Name", config.configName);
+            printResult("Description", config.description || "-");
+            printResult("Status", formatStatus(config.status));
+            printResult("Created", formatDate(config.createdAt));
+            printResult("Modified", formatDate(config.modifiedAt));
+            if (config.steps && config.steps.length > 0) {
+                divider();
+                info(`Steps (${config.steps.length}):`);
+                config.steps.forEach((step, idx) => {
+                    console.log(`  ${idx + 1}. ${step.id} (${step.kind})`);
+                });
+            }
         }
         catch (err) {
             error(err.message);
@@ -811,26 +1814,36 @@ Examples:
     // ============================================
     const tests = workflows
         .command("tests")
-        .description("Manage and run workflow test cases")
+        .description("Run and inspect workflow test cases")
         .addHelpText("after", `
+Test cases are authored in TOML at workflows/<key>.tests/<name>.toml and applied
+with \`primitive config push\` (issue #2769); \`config fields workflow\` lists the
+fields. Deleting a case is removing its file and running \`config push --prune\`.
+
+Every verb here takes the workflow's KEY or its ID (#3249). An identifier
+that names no workflow in the app fails; "No test cases found." only ever
+means the workflow exists and has no cases.
+
 Examples:
-  $ primitive workflows tests list <workflow-id>
-  $ primitive workflows tests create <workflow-id> --name "Basic test" --vars '{"input":"hello"}'
-  $ primitive workflows tests run <workflow-id> <test-case-id>
-  $ primitive workflows tests run-all <workflow-id>
-  $ primitive workflows tests run-all <workflow-id> --test-cases 01ABC,01DEF,01GHI
+  $ primitive workflows tests list <workflow>
+  $ primitive workflows tests run <workflow> <test-case-id>
+  $ primitive workflows tests run-all <workflow>
+  $ primitive workflows tests run-all <workflow> --test-cases 01ABC,01DEF,01GHI
 `);
-    // List test cases
     tests
         .command("list")
         .description("List test cases for a workflow")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow>", "Workflow key or ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, options) => {
+        .action(async (workflow, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
+            // #3249 — resolve BEFORE listing: the test-case endpoint answers an
+            // unknown block with an empty page, so an unresolved identifier would
+            // print "No test cases found." and exit 0.
+            const workflowId = await resolveWorkflowId(client, resolvedAppId, workflow);
             const { items } = await client.listTestCases(resolvedAppId, "workflow", workflowId);
             if (options.json) {
                 json(items);
@@ -851,97 +1864,33 @@ Examples:
             process.exit(1);
         }
     });
-    // Create test case
-    tests
-        .command("create")
-        .description("Create a test case for a workflow")
-        .argument("<workflow-id>", "Workflow ID")
-        .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--name <name>", "Test case name (required)")
-        .option("--vars <json>", "Input variables as JSON (required)")
-        .option("--pattern <regex>", "Expected output pattern (regex)")
-        .option("--contains <json>", "Expected strings to contain (JSON array)")
-        .option("--json-subset <json>", "Expected JSON subset to match")
-        .option("--json", "Output as JSON")
-        .action(async (workflowId, options) => {
-        const resolvedAppId = resolveAppId(undefined, options);
-        if (!options.name || !options.vars) {
-            error("Required: --name and --vars");
-            process.exit(1);
-        }
-        let inputVariables;
-        try {
-            inputVariables = JSON.parse(options.vars);
-        }
-        catch {
-            error("Invalid JSON in --vars");
-            process.exit(1);
-        }
-        let expectedOutputContains;
-        if (options.contains) {
-            try {
-                expectedOutputContains = JSON.parse(options.contains);
-            }
-            catch {
-                error("Invalid JSON in --contains");
-                process.exit(1);
-            }
-        }
-        let expectedJsonSubset;
-        if (options.jsonSubset) {
-            try {
-                expectedJsonSubset = JSON.parse(options.jsonSubset);
-            }
-            catch {
-                error("Invalid JSON in --json-subset");
-                process.exit(1);
-            }
-        }
-        const client = new ApiClient();
-        try {
-            const result = await client.createTestCase(resolvedAppId, "workflow", workflowId, {
-                name: options.name,
-                inputVariables,
-                expectedOutputPattern: options.pattern,
-                expectedOutputContains,
-                expectedJsonSubset,
-            });
-            if (options.json) {
-                json(result);
-                return;
-            }
-            success(`Test case created: ${result.name}`);
-            keyValue("Test Case ID", result.testCaseId);
-        }
-        catch (err) {
-            error(err.message);
-            process.exit(1);
-        }
-    });
     // Get test case
     tests
         .command("get")
         .description("Get test case details")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow>", "Workflow key or ID")
         .argument("<test-case-id>", "Test Case ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, testCaseId, options) => {
+        .action(async (workflow, testCaseId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
+            const workflowId = await resolveWorkflowId(client, resolvedAppId, workflow);
             const result = await client.getTestCase(resolvedAppId, "workflow", workflowId, testCaseId);
             if (options.json) {
                 json(result);
                 return;
             }
-            keyValue("Test Case ID", result.testCaseId);
-            keyValue("Name", result.name);
-            keyValue("Created", formatDate(result.createdAt));
+            printResult("Test Case ID", result.testCaseId);
+            printResult("Name", result.name);
+            printResult("Created", formatDate(result.createdAt));
             divider();
             info("Input Variables:");
             try {
-                const vars = JSON.parse(result.inputVariables || "{}");
+                const vars = typeof result.inputVariables === 'string'
+                    ? JSON.parse(result.inputVariables || "{}")
+                    : (result.inputVariables || {});
                 console.log(JSON.stringify(vars, null, 2));
             }
             catch {
@@ -949,13 +1898,16 @@ Examples:
             }
             if (result.expectedOutputPattern) {
                 divider();
-                keyValue("Expected Pattern", result.expectedOutputPattern);
+                printResult("Expected Pattern", result.expectedOutputPattern);
             }
             if (result.expectedOutputContains) {
                 divider();
                 info("Expected Contains:");
                 try {
-                    console.log(JSON.stringify(JSON.parse(result.expectedOutputContains), null, 2));
+                    const contains = typeof result.expectedOutputContains === 'string'
+                        ? JSON.parse(result.expectedOutputContains)
+                        : result.expectedOutputContains;
+                    console.log(JSON.stringify(contains, null, 2));
                 }
                 catch {
                     console.log(result.expectedOutputContains);
@@ -965,7 +1917,10 @@ Examples:
                 divider();
                 info("Expected JSON Subset:");
                 try {
-                    console.log(JSON.stringify(JSON.parse(result.expectedJsonSubset), null, 2));
+                    const subset = typeof result.expectedJsonSubset === 'string'
+                        ? JSON.parse(result.expectedJsonSubset)
+                        : result.expectedJsonSubset;
+                    console.log(JSON.stringify(subset, null, 2));
                 }
                 catch {
                     console.log(result.expectedJsonSubset);
@@ -977,140 +1932,25 @@ Examples:
             process.exit(1);
         }
     });
-    // Update test case
-    tests
-        .command("update")
-        .description("Update a test case")
-        .argument("<workflow-id>", "Workflow ID")
-        .argument("<test-case-id>", "Test Case ID")
-        .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--name <name>", "Test case name")
-        .option("--vars <json>", "Input variables as JSON")
-        .option("--pattern <regex>", "Expected output pattern (regex)")
-        .option("--contains <json>", "Expected strings to contain (JSON array)")
-        .option("--json-subset <json>", "Expected JSON subset to match")
-        .option("--clear-pattern", "Clear expected pattern")
-        .option("--clear-contains", "Clear expected contains")
-        .option("--clear-json-subset", "Clear expected JSON subset")
-        .option("--json", "Output as JSON")
-        .action(async (workflowId, testCaseId, options) => {
-        const resolvedAppId = resolveAppId(undefined, options);
-        const payload = {};
-        if (options.name)
-            payload.name = options.name;
-        if (options.vars) {
-            try {
-                payload.inputVariables = JSON.parse(options.vars);
-            }
-            catch {
-                error("Invalid JSON in --vars");
-                process.exit(1);
-            }
-        }
-        if (options.clearPattern) {
-            payload.expectedOutputPattern = null;
-        }
-        else if (options.pattern) {
-            payload.expectedOutputPattern = options.pattern;
-        }
-        if (options.clearContains) {
-            payload.expectedOutputContains = null;
-        }
-        else if (options.contains) {
-            try {
-                payload.expectedOutputContains = JSON.parse(options.contains);
-            }
-            catch {
-                error("Invalid JSON in --contains");
-                process.exit(1);
-            }
-        }
-        if (options.clearJsonSubset) {
-            payload.expectedJsonSubset = null;
-        }
-        else if (options.jsonSubset) {
-            try {
-                payload.expectedJsonSubset = JSON.parse(options.jsonSubset);
-            }
-            catch {
-                error("Invalid JSON in --json-subset");
-                process.exit(1);
-            }
-        }
-        if (Object.keys(payload).length === 0) {
-            error("No update options specified.");
-            process.exit(1);
-        }
-        const client = new ApiClient();
-        try {
-            const result = await client.updateTestCase(resolvedAppId, "workflow", workflowId, testCaseId, payload);
-            if (options.json) {
-                json(result);
-                return;
-            }
-            success("Test case updated.");
-        }
-        catch (err) {
-            error(err.message);
-            process.exit(1);
-        }
-    });
-    // Delete test case
-    tests
-        .command("delete")
-        .description("Delete a test case")
-        .argument("<workflow-id>", "Workflow ID")
-        .argument("<test-case-id>", "Test Case ID")
-        .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("-y, --yes", "Skip confirmation prompt")
-        .action(async (workflowId, testCaseId, options) => {
-        const resolvedAppId = resolveAppId(undefined, options);
-        if (!options.yes) {
-            const inquirer = await import("inquirer");
-            const { confirm } = await inquirer.default.prompt([
-                {
-                    type: "confirm",
-                    name: "confirm",
-                    message: `Are you sure you want to delete test case ${testCaseId}?`,
-                    default: false,
-                },
-            ]);
-            if (!confirm) {
-                info("Cancelled.");
-                return;
-            }
-        }
-        const client = new ApiClient();
-        try {
-            await client.deleteTestCase(resolvedAppId, "workflow", workflowId, testCaseId);
-            success("Test case deleted.");
-        }
-        catch (err) {
-            error(err.message);
-            process.exit(1);
-        }
-    });
     // Run a single test case
     tests
         .command("run")
         .description("Run a single test case")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow>", "Workflow key or ID")
         .argument("<test-case-id>", "Test Case ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, testCaseId, options) => {
+        .action(async (workflow, testCaseId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
+            const workflowId = await resolveWorkflowId(client, resolvedAppId, workflow);
             // Get test case to retrieve its variables
             const testCase = await client.getTestCase(resolvedAppId, "workflow", workflowId, testCaseId);
-            let variables = {};
-            try {
-                variables = JSON.parse(testCase.inputVariables || "{}");
-            }
-            catch {
-                // Use empty object if parsing fails
-            }
+            // #3158 — the case's stored input IS the run's input, and the server
+            // hands it back parsed; `JSON.parse` on that object threw and the run
+            // went ahead with `{}`, disagreeing with `run-all` over the same case.
+            const variables = requireTestCaseInputVariables(testCase.inputVariables, testCaseId);
             const result = await client.executeBlockTest(resolvedAppId, "workflow", workflowId, {
                 variables,
                 testCaseId,
@@ -1163,14 +2003,17 @@ Examples:
     tests
         .command("run-all")
         .description("Run all test cases for a workflow, or a specific subset")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow>", "Workflow key or ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--test-cases <ids>", "Comma-separated list of test case IDs to run (runs only these)")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, options) => {
+        .action(async (workflow, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
+            // #3249 — resolve first, so `Total Tests: 0` can only ever mean the
+            // workflow exists and has no cases, never that the identifier was wrong.
+            const workflowId = await resolveWorkflowId(client, resolvedAppId, workflow);
             // Parse test case IDs if provided
             let testCaseIds;
             if (options.testCases) {
@@ -1227,15 +2070,16 @@ Examples:
     tests
         .command("runs")
         .description("List recent test runs for a workflow")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow>", "Workflow key or ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--limit <n>", "Maximum number of runs to show", "20")
         .option("--group <group>", "Filter by comparison group")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, options) => {
+        .action(async (workflow, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
+            const workflowId = await resolveWorkflowId(client, resolvedAppId, workflow);
             const { items } = await client.listTestRuns(resolvedAppId, "workflow", workflowId, {
                 limit: parseInt(options.limit),
                 comparisonGroup: options.group,
@@ -1277,23 +2121,24 @@ Examples:
         .description("Run tests in parallel using workflow-based execution")
         .addHelpText("after", `
 Examples:
-  $ primitive workflows tests batch start <workflow-id>
-  $ primitive workflows tests batch start <workflow-id> --test-cases 01ABC,01DEF
-  $ primitive workflows tests batch status <workflow-id> <batch-id>
-  $ primitive workflows tests batch cancel <workflow-id> <batch-id>
+  $ primitive workflows tests batch start <workflow>
+  $ primitive workflows tests batch start <workflow> --test-cases 01ABC,01DEF
+  $ primitive workflows tests batch status <workflow> <batch-id>
+  $ primitive workflows tests batch cancel <workflow> <batch-id>
 `);
     // Start batch tests
     batch
         .command("start")
         .description("Start parallel batch test execution")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow>", "Workflow key or ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--test-cases <ids>", "Comma-separated list of test case IDs to run")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, options) => {
+        .action(async (workflow, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
+            const workflowId = await resolveWorkflowId(client, resolvedAppId, workflow);
             let testCaseIds;
             if (options.testCases) {
                 testCaseIds = options.testCases.split(",").map((id) => id.trim()).filter(Boolean);
@@ -1329,27 +2174,37 @@ Examples:
     batch
         .command("status")
         .description("Get status of a batch test execution")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow>", "Workflow key or ID")
         .argument("<batch-id>", "Batch ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--wait", "Wait for completion (polls every 2 seconds)")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, batchId, options) => {
+        .action(async (workflow, batchId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
+            const workflowId = await resolveWorkflowId(client, resolvedAppId, workflow);
             const fetchStatus = async () => {
                 return client.getBatchTestStatus(resolvedAppId, "workflow", workflowId, batchId);
             };
             let result = await fetchStatus();
             if (options.wait && result.status === "running") {
                 info("Waiting for batch completion...");
-                while (result.status === "running") {
-                    await new Promise((r) => setTimeout(r, 2000));
-                    result = await fetchStatus();
-                    process.stdout.write(`\r  Completed: ${result.completed}/${result.results?.length || 0}  `);
-                }
-                console.log();
+                // #1969 — shared poll primitive (delay-then-probe every 2s, honoring
+                // an AbortSignal). `onError: "throw"` preserves the prior loop's
+                // fail-fast behavior: a throwing fetchStatus (expired token, deleted
+                // batch → 401/404) propagates to the command-level catch below and
+                // exits non-zero, rather than spinning forever with a frozen progress
+                // line. (The follow/watch tail loops legitimately keep "continue".)
+                const controller = new AbortController();
+                result = await pollUntil(fetchStatus, (r) => r.status !== "running", {
+                    intervalMs: 2000,
+                    signal: controller.signal,
+                    onError: "throw",
+                    // Progress goes to stderr so it can't corrupt JSON on stdout under --json.
+                    onValue: (r) => progress(`  Completed: ${r.completed}/${r.results?.length || 0}  `),
+                });
+                progressEnd();
             }
             if (options.json) {
                 json(result);
@@ -1385,23 +2240,22 @@ Examples:
     batch
         .command("cancel")
         .description("Cancel a running batch test execution")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow>", "Workflow key or ID")
         .argument("<batch-id>", "Batch ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("-y, --yes", "Skip confirmation prompt")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, batchId, options) => {
+        .action(async (workflow, batchId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         if (!options.yes) {
-            const inquirer = await import("inquirer");
-            const { confirm } = await inquirer.default.prompt([
-                {
-                    type: "confirm",
-                    name: "confirm",
-                    message: `Are you sure you want to cancel batch ${batchId}?`,
-                    default: false,
-                },
-            ]);
+            let confirm;
+            try {
+                confirm = await confirmPrompt(`Are you sure you want to cancel batch ${batchId}?`);
+            }
+            catch (err) {
+                error(err.message);
+                process.exit(1);
+            }
             if (!confirm) {
                 info("Cancelled.");
                 return;
@@ -1409,6 +2263,7 @@ Examples:
         }
         const client = new ApiClient();
         try {
+            const workflowId = await resolveWorkflowId(client, resolvedAppId, workflow);
             // First get current status to get instance IDs
             const status = await client.getBatchTestStatus(resolvedAppId, "workflow", workflowId, batchId);
             // Get instance IDs that are still running
@@ -1441,26 +2296,29 @@ Examples:
     // ============================================
     const attachments = tests
         .command("attachments")
-        .description("Manage test case file attachments")
+        .description("Inspect test case file attachments")
         .addHelpText("after", `
+Attachments are authored as files in workflows/<key>.tests/<name>/ and applied
+with \`primitive config push\` (#2769); removing one and running
+\`config push --prune\` deletes it server-side. These verbs read server state.
+
 Examples:
-  $ primitive workflows tests attachments list <workflow-id> <test-case-id>
-  $ primitive workflows tests attachments upload <workflow-id> <test-case-id> ./document.pdf
-  $ primitive workflows tests attachments download <workflow-id> <test-case-id> document.pdf
-  $ primitive workflows tests attachments delete <workflow-id> <test-case-id> document.pdf
+  $ primitive workflows tests attachments list <workflow> <test-case-id>
+  $ primitive workflows tests attachments download <workflow> <test-case-id> document.pdf
 `);
     // List attachments
     attachments
         .command("list")
         .description("List attachments for a test case")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow>", "Workflow key or ID")
         .argument("<test-case-id>", "Test Case ID")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
         .option("--json", "Output as JSON")
-        .action(async (workflowId, testCaseId, options) => {
+        .action(async (workflow, testCaseId, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
+            const workflowId = await resolveWorkflowId(client, resolvedAppId, workflow);
             const result = await client.listTestCaseAttachments(resolvedAppId, "workflow", workflowId, testCaseId);
             if (options.json) {
                 json(result.attachments);
@@ -1485,62 +2343,20 @@ Examples:
             process.exit(1);
         }
     });
-    // Upload attachment
-    attachments
-        .command("upload")
-        .description("Upload a file attachment to a test case")
-        .argument("<workflow-id>", "Workflow ID")
-        .argument("<test-case-id>", "Test Case ID")
-        .argument("<file-path>", "Path to the file to upload")
-        .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("--name <name>", "Override the filename")
-        .option("--json", "Output as JSON")
-        .action(async (workflowId, testCaseId, filePath, options) => {
-        const resolvedAppId = resolveAppId(undefined, options);
-        const client = new ApiClient();
-        try {
-            // Check file exists and get stats
-            const stats = statSync(filePath);
-            if (!stats.isFile()) {
-                error(`Not a file: ${filePath}`);
-                process.exit(1);
-            }
-            // Check file size (10 MB limit)
-            const TEN_MB = 10 * 1024 * 1024;
-            if (stats.size > TEN_MB) {
-                error(`File too large (max 10MB): ${formatFileSize(stats.size)}`);
-                process.exit(1);
-            }
-            const filename = options.name || basename(filePath);
-            const data = readFileSync(filePath);
-            const contentType = mimeLookup(filePath) || "application/octet-stream";
-            const result = await client.uploadTestCaseAttachment(resolvedAppId, "workflow", workflowId, testCaseId, filename, data, contentType);
-            if (options.json) {
-                json(result);
-                return;
-            }
-            success(`Uploaded: ${filename}`);
-            keyValue("Size", formatFileSize(result.attachment.size));
-            keyValue("Type", result.attachment.contentType);
-        }
-        catch (err) {
-            error(err.message);
-            process.exit(1);
-        }
-    });
     // Download attachment
     attachments
         .command("download")
         .description("Download a file attachment from a test case")
-        .argument("<workflow-id>", "Workflow ID")
+        .argument("<workflow>", "Workflow key or ID")
         .argument("<test-case-id>", "Test Case ID")
         .argument("<filename>", "Name of the file to download")
         .argument("[output-path]", "Output path (defaults to current directory)")
         .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .action(async (workflowId, testCaseId, filename, outputPath, options) => {
+        .action(async (workflow, testCaseId, filename, outputPath, options) => {
         const resolvedAppId = resolveAppId(undefined, options);
         const client = new ApiClient();
         try {
+            const workflowId = await resolveWorkflowId(client, resolvedAppId, workflow);
             const result = await client.downloadTestCaseAttachment(resolvedAppId, "workflow", workflowId, testCaseId, filename);
             const finalPath = outputPath || filename;
             writeFileSync(finalPath, result.data);
@@ -1553,36 +2369,129 @@ Examples:
             process.exit(1);
         }
     });
-    // Delete attachment
-    attachments
-        .command("delete")
-        .description("Delete a file attachment from a test case")
-        .argument("<workflow-id>", "Workflow ID")
-        .argument("<test-case-id>", "Test Case ID")
-        .argument("<filename>", "Name of the file to delete")
-        .option("--app <app-id>", "App ID (uses current app if not specified)")
-        .option("-y, --yes", "Skip confirmation prompt")
-        .action(async (workflowId, testCaseId, filename, options) => {
-        const resolvedAppId = resolveAppId(undefined, options);
-        if (!options.yes) {
-            const inquirer = await import("inquirer");
-            const { confirm } = await inquirer.default.prompt([
-                {
-                    type: "confirm",
-                    name: "confirm",
-                    message: `Are you sure you want to delete attachment "${filename}"?`,
-                    default: false,
-                },
-            ]);
-            if (!confirm) {
-                info("Cancelled.");
+    // ============================================
+    // Codegen command (issue #1442)
+    // ============================================
+    workflows
+        .command("codegen")
+        .description("Generate typed client signatures from the local workflows/*.toml schemas. TypeScript (default): one <key>.generated.ts per workflow (<Key>Input/<Key>Output types + a <key>(client) invoker factory). Swift (--lang swift): one <key>.generated.swift per workflow (<Key>Input/<Key>Output Codable types).\n\nFormatting: the emitted TypeScript is already formatted the way Prettier's DEFAULT config formats it, so a `codegen` -> `prettier --write` -> `codegen --check` pipeline stays green. If your project uses a NON-default Prettier config (a different printWidth, trailingComma, or quote style), Prettier will reformat the generated files and `--check` will then fail: add the generated directory to `.prettierignore` and commit the raw emission instead.")
+        .argument("[workflow-key]", "Generate for a single workflow (defaults to every workflows/*.toml found)")
+        .option("-o, --output <dir>", "Output directory for the generated files (defaults to <config-dir>/workflows/generated/)")
+        .option("--lang <lang>", "Target language for the generated types: 'ts' (default) or 'swift'.", "ts")
+        .option("--check", "Exit non-zero if generated output is out of date (CI guard); does not write. Compares byte-for-byte, so the file on disk must match the emission exactly.")
+        .option("--json", "Output the result summary as JSON")
+        .action(async (workflowKey, options) => {
+        try {
+            // 0. Validate --lang up front so an unknown value fails clearly rather
+            //    than silently defaulting to the TS emitter.
+            const lang = String(options.lang ?? "ts").toLowerCase();
+            if (lang !== "ts" && lang !== "swift") {
+                error(`Unknown --lang "${options.lang}". Use "ts" or "swift".`);
+                process.exit(1);
+            }
+            // 1. The single source workflows/ directory: the environment's, and
+            //    only ever the environment's (#1510, #3154).
+            const workflowsSourceDir = resolveCodegenSourceDir({
+                subdir: "workflows",
+            });
+            // 2. Collect the source .toml files (one per workflow), filtered to a
+            //    single workflow when an argument is given. Match on the file stem;
+            //    the generator resolves the real key from `[workflow].key`.
+            const inputs = [];
+            for (const fileName of readdirSync(workflowsSourceDir)) {
+                if (!fileName.endsWith(".toml"))
+                    continue;
+                const stem = fileName.slice(0, -".toml".length);
+                if (workflowKey && stem !== workflowKey)
+                    continue;
+                const tomlPath = path.join(workflowsSourceDir, fileName);
+                inputs.push({
+                    fileStem: stem,
+                    tomlPath,
+                    tomlContent: readFileSync(tomlPath, "utf-8"),
+                });
+            }
+            if (inputs.length === 0) {
+                error(workflowKey
+                    ? `No workflows/${workflowKey}.toml found under primitive/<env>/.`
+                    : "No workflows/*.toml files found to generate from.");
+                process.exit(1);
+            }
+            // 3. Resolve the output directory. Default: a `generated/` subdir of
+            //    the first source dir (keeps generated output beside the schema).
+            const outputDir = options.output
+                ? path.resolve(options.output)
+                : path.join(path.dirname(inputs[0].tomlPath), "generated");
+            // 4. Run codegen (or --check) with the selected emitter. Both the TS
+            //    and Swift generators return the same WriteOrCheckResult shape, so
+            //    the reporting below is emitter-agnostic. Fork 1 = A: only the
+            //    emitter differs; both read the same synced TOMLs and the same
+            //    drift-guarded schema descriptor.
+            const codegenResult = lang === "swift"
+                ? await generateWorkflowSwiftTypes({
+                    inputs,
+                    outputDir,
+                    check: !!options.check,
+                    singleWorkflow: !!workflowKey,
+                })
+                : await generateWorkflowTypes({
+                    inputs,
+                    outputDir,
+                    check: !!options.check,
+                    singleWorkflow: !!workflowKey,
+                });
+            if (options.check) {
+                if (codegenResult.mismatches.length > 0) {
+                    if (options.json) {
+                        json({ ok: false, mismatches: codegenResult.mismatches });
+                    }
+                    else {
+                        error(`Check failed: ${codegenResult.mismatches.length} file(s) out of date.`);
+                        for (const m of codegenResult.mismatches) {
+                            error(`  ${m.reason}: ${path.relative(process.cwd(), m.filePath)}`);
+                        }
+                        // Echo back the options that decide WHERE and WHAT was
+                        // checked. Suggesting a bare `workflows codegen` after a
+                        // `--check -o <dir>` sent people to regenerate into the
+                        // default <config-dir>/workflows/generated/ instead of the
+                        // directory that is actually out of date, so the files stayed
+                        // stale and the next check failed the same way.
+                        const regenerate = ["primitive workflows codegen"];
+                        if (workflowKey)
+                            regenerate.push(workflowKey);
+                        if (lang === "swift")
+                            regenerate.push("--lang swift");
+                        if (options.output)
+                            regenerate.push(`-o ${options.output}`);
+                        info(`Run \`${regenerate.join(" ")}\` to regenerate.`);
+                    }
+                    process.exit(1);
+                }
+                if (options.json) {
+                    json({ ok: true, checked: codegenResult.writtenFiles.length });
+                }
+                else {
+                    success(`Check passed: ${codegenResult.writtenFiles.length} file(s) up to date.`);
+                }
                 return;
             }
-        }
-        const client = new ApiClient();
-        try {
-            await client.deleteTestCaseAttachment(resolvedAppId, "workflow", workflowId, testCaseId, filename);
-            success(`Deleted: ${filename}`);
+            if (options.json) {
+                json({
+                    written: codegenResult.writtenFiles,
+                    deleted: codegenResult.deletedFiles,
+                });
+                return;
+            }
+            success(`Generated ${codegenResult.writtenFiles.length} file(s)` +
+                (codegenResult.deletedFiles.length > 0
+                    ? `, deleted ${codegenResult.deletedFiles.length} stale file(s).`
+                    : "."));
+            for (const f of codegenResult.writtenFiles) {
+                keyValue("  wrote", path.relative(process.cwd(), f));
+            }
+            for (const f of codegenResult.deletedFiles) {
+                keyValue("  deleted", path.relative(process.cwd(), f));
+            }
         }
         catch (err) {
             error(err.message);