@sema-agent/server 1.241.0 → 1.243.0

package/dist/http/server.js

@@ -25,6 +25,7 @@ import { RateLimiter } from "../observability/rate-limit.js";
 import { withPrincipal } from "../observability/principal-context.js";
 import { projectEvents, runSummary, mapTraceEvent, toolStartEventData, toolEndEventData, taskProgressEventData, taskNotificationEventData, compactedEventData, diagnosticsEventData, brainStatusEventData, appendModelUsageDelta, appendPromptManifest, attachModelUsage } from "../trace/project.js";
 import { createLedgerSink } from "../trace/ledger-sink.js";
+import { cacheFamilyOfMirror } from "../budget.js";
 import { projectArtifacts } from "../trace/artifacts.js";
 import { composeSupervisorCost, infraCost, infraUsageFromEvents, hasInfraPricing } from "../finance/cost-taxonomy.js";
 import { redactSecrets, redactDeep, redactedPreview } from "../trace/redact.js";
@@ -329,6 +330,10 @@ export function createHttpServer(deps) {
                 port: deps.config.port,
                 ...(deps.config.directDoorActive ? { directDoorActive: true } : {}),
                 ...(restart ? { restartRequired: true, restart: { reasons: restart.reasons, version: restart.version, since: restart.since } } : {}),
+                ...(() => {
+                    const stuck = deps.planeDeferredState?.();
+                    return stuck ? { modelPlaneDeferred: { version: stuck.version, since: stuck.since, noHandoff: true, ...(stuck.blocked ? { blockedReasons: stuck.blocked } : {}) } } : {};
+                })(),
             });
             return;
         }
@@ -659,6 +664,8 @@ export function createHttpServer(deps) {
                 subagentSteer: Boolean(deps.subagentSteerRegistry) && Boolean(deps.runStore),
                 subagentResume: Boolean(deps.subagentSteerRegistry) && Boolean(deps.runStore),
                 taskSettings: { permissions: true, permissionMode: true, model: true, outputStyle: true, env: false, hooks: deps.config.requirePrincipal !== true },
+                appendSystemPrompt: true,
+                compactionModel: true,
                 taskAgents: deps.config.requirePrincipal !== true,
                 retainBackgroundProcesses: deps.config.requirePrincipal !== true,
                 interactiveTools: true,
@@ -670,6 +677,7 @@ export function createHttpServer(deps) {
                 mcpElicitation: Boolean(deps.elicitation),
                 askUserQuestion: Boolean(deps.question),
                 toolApproval: Boolean(deps.toolApproval),
+                sideQuery: true,
                 sessionList: Boolean(deps.sessionStorage?.listSessions ?? deps.runStore?.listSessions),
                 sessionSearch: Boolean(deps.sessionStorage?.listSessions ?? deps.runStore?.listSessions),
                 sessionFork: Boolean(deps.sessionStorage?.fork),
@@ -708,6 +716,77 @@ export function createHttpServer(deps) {
             sendJson(res, 200, { models, default: deps.config.model.id, defaultModel: { id: deps.config.model.id, ...(defaultName ? { name: defaultName } : {}) } });
             return;
         }
+        if (req.method === "POST" && url === "/v1/side-query") {
+            const ac = new AbortController();
+            const onClose = () => ac.abort();
+            res.on("close", onClose);
+            try {
+                if (rateLimited(req, res) || quotaExceeded(req, res) || (await leaseDenied(req, res)))
+                    return;
+                const principal = gatedPrincipal(req, deps.config);
+                if (deps.config.requirePrincipal && !principal) {
+                    sendJson(res, 401, { error: `missing principal header '${deps.config.principalHeader}'` });
+                    return;
+                }
+                let body;
+                try {
+                    const raw = await readJson(req);
+                    if (raw === null || typeof raw !== "object" || Array.isArray(raw))
+                        throw new HttpError(400, "body must be a JSON object (SideQuerySpec shape)");
+                    body = raw;
+                }
+                catch (e) {
+                    const he = e instanceof HttpError ? e : new HttpError(400, msg(e));
+                    sendJson(res, he.status, { error: he.message });
+                    return;
+                }
+                if (!Array.isArray(body.messages) || body.messages.length === 0) {
+                    sendJson(res, 400, { error: "messages must be a non-empty array (SideQueryMessage[])" });
+                    return;
+                }
+                if (body.thinking !== undefined && !isThinkingLevel(body.thinking)) {
+                    sendJson(res, 400, { error: "invalid thinking level" });
+                    return;
+                }
+                if (ac.signal.aborted)
+                    return;
+                const spec = {
+                    messages: body.messages,
+                    ...(typeof body.model === "string" ? { model: body.model } : {}),
+                    ...(typeof body.modelRole === "string" ? { modelRole: body.modelRole } : {}),
+                    ...(isThinkingLevel(body.thinking) ? { thinking: body.thinking } : {}),
+                    ...(typeof body.systemPrompt === "string" ? { systemPrompt: body.systemPrompt } : {}),
+                    ...(Array.isArray(body.tools) ? { tools: body.tools } : {}),
+                    ...(typeof body.maxOutputTokens === "number" && Number.isFinite(body.maxOutputTokens) && body.maxOutputTokens > 0 ? { maxOutputTokens: Math.floor(body.maxOutputTokens) } : {}),
+                    signal: ac.signal,
+                };
+                try {
+                    const catalogSnapshot = Object.values(deps.runner.agentCatalog?.models ?? {}).map((m) => ({ id: m?.id, api: m?.api, params: m?.params?.promptCacheFamily ? { promptCacheFamily: m.params.promptCacheFamily } : undefined }));
+                    const result = await deps.runner.sideQuery(spec);
+                    const rr = result;
+                    const modelId = typeof rr.model === "string" ? rr.model : "unknown";
+                    const byId = catalogSnapshot.filter((m) => m.id === modelId);
+                    const families = new Set(byId.map((m) => cacheFamilyOfMirror(m)));
+                    const family = families.size === 1 ? [...families][0] : undefined;
+                    deps.sideQueryAccounting?.(principal, { model: modelId, ...(family ? { family } : {}), usage: rr.usage });
+                    deps.metrics?.inc?.("side_query_total", { result: "ok" });
+                    if (!res.writableEnded)
+                        sendJson(res, 200, result);
+                }
+                catch (e) {
+                    deps.metrics?.inc?.("side_query_total", { result: "error" });
+                    if (!res.writableEnded && !ac.signal.aborted) {
+                        const m = msg(e);
+                        const isValidation = /^Unknown model ref |^No model for role |requires a non-empty messages array/.test(m);
+                        sendJson(res, isValidation ? 400 : 500, { error: redactSecrets(m) });
+                    }
+                }
+            }
+            finally {
+                res.off("close", onClose);
+            }
+            return;
+        }
         if (req.method === "POST" && (url === "/v1/tasks" || url === "/v1/tasks/stream")) {
             const rawIdem = headerStr(req.headers["idempotency-key"]);
             const idemKey = rawIdem ? scopedIdempotencyKey(rawIdem, source, gatedPrincipal(req, deps.config)) : undefined;
@@ -4090,6 +4169,16 @@ export function createHttpServer(deps) {
             sendJson(res, 400, { error: `systemPrompt must be at most ${MAX_SYSTEM_PROMPT_CHARS} characters` });
             return null;
         }
+        if (body.appendSystemPrompt !== undefined) {
+            if (typeof body.appendSystemPrompt !== "string" || body.appendSystemPrompt.length === 0) {
+                sendJson(res, 400, { error: "appendSystemPrompt must be a non-empty string" });
+                return null;
+            }
+            if (body.appendSystemPrompt.length > MAX_SYSTEM_PROMPT_CHARS) {
+                sendJson(res, 400, { error: `appendSystemPrompt must be at most ${MAX_SYSTEM_PROMPT_CHARS} characters` });
+                return null;
+            }
+        }
         if (body.cwd !== undefined && !isValidCwd(body.cwd)) {
             sendJson(res, 400, { error: "cwd must be a non-empty absolute host path" });
             return null;
@@ -4121,6 +4210,13 @@ export function createHttpServer(deps) {
                 sendJson(res, 400, { error: `settings.outputStyle must be at most ${MAX_SETTINGS_OUTPUT_STYLE_CHARS} characters` });
                 return null;
             }
+            if (typeof st.outputStyle === "string" &&
+                st.outputStyle.length > 0 &&
+                typeof body.appendSystemPrompt === "string" &&
+                body.appendSystemPrompt.length + 2 + st.outputStyle.length > MAX_SYSTEM_PROMPT_CHARS) {
+                sendJson(res, 400, { error: `appendSystemPrompt + settings.outputStyle fold into one system-prompt block — combined they must be at most ${MAX_SYSTEM_PROMPT_CHARS} characters (including the 2-char joiner)` });
+                return null;
+            }
             const perms = st.permissions;
             if (perms !== null && typeof perms === "object" && !Array.isArray(perms)) {
                 for (const k of ["allow", "deny", "ask"]) {
@@ -4352,6 +4448,21 @@ export function createHttpServer(deps) {
                 return null;
             }
         }
+        if (body.compactionModel !== undefined) {
+            if (typeof body.compactionModel !== "string" || body.compactionModel.length === 0) {
+                sendJson(res, 400, { error: "compactionModel must be a non-empty string (a configured catalog name, tier word, or model id)" });
+                return null;
+            }
+            const bare = deps.config.models ?? {};
+            const catalog = expandTiers(bare, deps.config.tiers ?? {}) ?? bare;
+            if (matchCatalogModel(body.compactionModel, catalog) === undefined) {
+                sendJson(res, 400, {
+                    error: `unknown compactionModel "${body.compactionModel.slice(0, 120)}" — not in the configured catalog (name, tier word, or id)`,
+                    available: Object.keys(catalog).filter((n) => n !== "default"),
+                });
+                return null;
+            }
+        }
         if (body.attachments !== undefined) {
             const a = body.attachments;
             if (typeof a !== "object" || a === null || Array.isArray(a)) {
@@ -4394,7 +4505,7 @@ export function createHttpServer(deps) {
         }
         try {
             const auth = deps.authorize ? await deps.authorize({ req, body }) : undefined;
-            return { spec: await deps.resolveSpec(body, req, auth), auth, verify, cascade, jobId: body.jobId, body };
+            return { spec: await deps.resolveSpec(body, req, auth, { leg: "fresh" }), auth, verify, cascade, jobId: body.jobId, body };
         }
         catch (err) {
             if (err instanceof HttpError) {
@@ -5598,7 +5709,8 @@ function headerStr(v) {
     return Array.isArray(v) ? v[0] : v;
 }
 function isBillableSubmitPath(url) {
-    return (url === "/v1/tasks" ||
+    return (url === "/v1/side-query" ||
+        url === "/v1/tasks" ||
         url === "/v1/tasks/stream" ||
         url === "/v1/runs" ||
         url === "/v1/leader" ||