youmd 0.8.29 → 0.9.1

package/dist/lib/orchestrator/loop.js

@@ -0,0 +1,248 @@
+"use strict";
+// loop.ts — the iterative agent loop for the You agent orchestrator.
+//
+// This is the one genuinely net-new piece the audit identified: every existing path in the
+// CLI is single-hop (route -> run -> template-summarize -> stop). This closes the loop:
+// the model picks a tool, we run it, feed the result back, and repeat until the model calls
+// `finish`. It is model-agnostic (the model caller is injected) and tool-agnostic (tools are
+// a registry of typed handlers), so the same loop drives orchestrator tools, brain tools, or
+// remote-machine tools without change.
+//
+// The loop deliberately stays small and uses a JSON tool-call convention over plain chat
+// completions, so it works against the existing /api/v1/chat proxy without needing native
+// provider tool_use. That keeps the You agent harness/model-agnostic by construction.
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.parseToolCall = parseToolCall;
+exports.runAgentLoop = runAgentLoop;
+const FINISH_TOOL = "finish";
+function buildSystemPrompt(goal, tools, context) {
+    const toolDocs = tools
+        .map((t) => {
+        const params = Object.entries(t.parameters)
+            .map(([k, v]) => `      - ${k}: ${v}`)
+            .join("\n");
+        return `  - ${t.name}: ${t.description}${params ? `\n${params}` : ""}`;
+    })
+        .join("\n");
+    return [
+        "You are U — Houston's personal master ORCHESTRATOR agent (U = the ultimate orchestrator, a role, not a person).",
+        "You do NOT write code or do the work yourself. You are model- and harness-agnostic: you",
+        "route work to the best worker agent (Claude Code for coding, etc.), launch and monitor",
+        "them across machines and projects, and report back. You are project-goal and task oriented.",
+        context ? `\nContext:\n${context}` : "",
+        "\nYou work in a loop. On EACH turn you call exactly one tool by replying with ONLY a JSON",
+        "object (no prose, no markdown fences) of the form:",
+        '  {"thought": "<one short sentence of reasoning>", "tool": "<tool_name>", "args": { ... }}',
+        "\nAvailable tools:",
+        toolDocs,
+        `  - ${FINISH_TOOL}: end the loop and report back. args: { summary: "what you did / found, for Houston" }`,
+        "\nRules:",
+        "- Reply with ONE json object and nothing else.",
+        "- Prefer delegating to a worker harness over doing work yourself.",
+        `- When the goal is met (or blocked), call ${FINISH_TOOL} with a clear summary.`,
+        `- Do not loop forever; if stuck, call ${FINISH_TOOL} and explain what is blocked.`,
+        `\nGoal: ${goal}`,
+    ]
+        .filter(Boolean)
+        .join("\n");
+}
+function toolCallFromObject(obj) {
+    if (typeof obj.tool !== "string")
+        return null;
+    const args = obj.args && typeof obj.args === "object" ? obj.args : {};
+    return {
+        thought: typeof obj.thought === "string" ? obj.thought : undefined,
+        tool: obj.tool,
+        args,
+    };
+}
+/**
+ * Extract a tool call from a model reply, tolerating prose around/between JSON, code fences, and
+ * multiple objects. Scans for every balanced `{…}` span and returns the first that parses to an
+ * object with a string `tool` — so "Sure! {…prose…} {\"tool\":…}" still works (the naive
+ * first-`{`-to-last-`}` slice would fail on that).
+ */
+/** Reject absurdly large candidate JSON before parsing — a sane tool call is small, and a
+ * multi-hundred-KB span is either junk or a runaway response we don't want to parse/echo. */
+const MAX_TOOL_CALL_CHARS = 256000;
+function parseToolCall(raw) {
+    if (!raw)
+        return null;
+    let text = raw.trim();
+    const fence = text.match(/```(?:json)?\s*([\s\S]*?)```/i);
+    if (fence)
+        text = fence[1].trim();
+    // Whole-string fast path.
+    if (text.length <= MAX_TOOL_CALL_CHARS) {
+        try {
+            const obj = JSON.parse(text);
+            const call = toolCallFromObject(obj);
+            if (call)
+                return call;
+        }
+        catch {
+            /* fall through to span scan */
+        }
+    }
+    // Scan for balanced {…} spans (string-aware) and try each.
+    let depth = 0;
+    let startIdx = -1;
+    let inStr = false;
+    let escaped = false;
+    for (let i = 0; i < text.length; i++) {
+        const ch = text[i];
+        if (inStr) {
+            if (escaped)
+                escaped = false;
+            else if (ch === "\\")
+                escaped = true;
+            else if (ch === '"')
+                inStr = false;
+            continue;
+        }
+        if (ch === '"')
+            inStr = true;
+        else if (ch === "{") {
+            if (depth === 0)
+                startIdx = i;
+            depth++;
+        }
+        else if (ch === "}") {
+            if (depth > 0) {
+                depth--;
+                if (depth === 0 && startIdx !== -1) {
+                    const candidate = text.slice(startIdx, i + 1);
+                    if (candidate.length > MAX_TOOL_CALL_CHARS) {
+                        startIdx = -1;
+                        continue;
+                    }
+                    try {
+                        const obj = JSON.parse(candidate);
+                        const call = toolCallFromObject(obj);
+                        if (call)
+                            return call;
+                    }
+                    catch {
+                        /* try next span */
+                    }
+                    startIdx = -1;
+                }
+            }
+        }
+    }
+    return null;
+}
+function truncate(s, max = 4000) {
+    return s.length > max ? s.slice(0, max) + `\n…[truncated ${s.length - max} chars]` : s;
+}
+/**
+ * Run the orchestrator loop until the model calls `finish` or maxSteps is hit.
+ * Each step: ask the model for a tool call → run it → append the result → repeat.
+ */
+async function runAgentLoop(options) {
+    const { goal, tools, callModel, context } = options;
+    const maxSteps = options.maxSteps ?? 12;
+    const maxModelRetries = options.maxModelRetries ?? 2;
+    const sleep = options.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
+    const toolByName = new Map(tools.map((t) => [t.name, t]));
+    // Absorb transient model-call failures (timeouts, 5xx, socket resets) with bounded
+    // exponential backoff. Real LLM/proxy endpoints hiccup; a single blip shouldn't abandon
+    // the whole goal mid-run. Throws the last error only after every attempt fails.
+    const callModelWithRetry = async (msgs) => {
+        let lastErr;
+        for (let attempt = 0; attempt <= maxModelRetries; attempt++) {
+            try {
+                return await callModel(msgs);
+            }
+            catch (err) {
+                lastErr = err;
+                if (attempt < maxModelRetries)
+                    await sleep(Math.min(250 * 2 ** attempt, 4000));
+            }
+        }
+        throw lastErr instanceof Error ? lastErr : new Error(String(lastErr));
+    };
+    const messages = [
+        { role: "system", content: buildSystemPrompt(goal, tools, context) },
+        { role: "user", content: `Begin. Goal: ${goal}` },
+    ];
+    const steps = [];
+    let consecutiveParseFailures = 0;
+    const MAX_PARSE_FAILURES = 2;
+    // Safety net for high maxSteps: each step appends 1-2 messages, so a long/stuck run could
+    // grow the context unboundedly. Stop cleanly well before that becomes a token problem.
+    const MAX_MESSAGES = 256;
+    for (let i = 0; i < maxSteps; i++) {
+        if (messages.length > MAX_MESSAGES) {
+            return {
+                finished: false,
+                summary: `Stopped: conversation grew past ${MAX_MESSAGES} messages without finishing (likely stuck). Narrow the goal or run again.`,
+                steps,
+            };
+        }
+        let reply;
+        try {
+            reply = await callModelWithRetry(messages);
+        }
+        catch (err) {
+            return {
+                finished: false,
+                summary: `Model call failed at step ${i + 1} after ${maxModelRetries + 1} attempt(s): ${err.message}`,
+                steps,
+            };
+        }
+        const call = parseToolCall(reply);
+        if (!call) {
+            // Model didn't emit a valid tool call — nudge, but bail after a couple of failures so a
+            // model that can't produce JSON doesn't silently burn the whole step budget.
+            consecutiveParseFailures++;
+            if (consecutiveParseFailures >= MAX_PARSE_FAILURES) {
+                return {
+                    finished: false,
+                    summary: `Stopped: the model did not return a valid tool call after ${MAX_PARSE_FAILURES} tries. Last reply: ${reply.slice(0, 200)}`,
+                    steps,
+                };
+            }
+            messages.push({ role: "assistant", content: reply });
+            messages.push({
+                role: "user",
+                content: 'That was not a valid tool call. Reply with ONLY a json object: {"thought":"...","tool":"...","args":{...}}.',
+            });
+            continue;
+        }
+        consecutiveParseFailures = 0;
+        messages.push({ role: "assistant", content: JSON.stringify(call) });
+        if (call.tool === FINISH_TOOL) {
+            const summary = typeof call.args.summary === "string" && call.args.summary.trim()
+                ? call.args.summary
+                : "Done.";
+            return { finished: true, summary, steps };
+        }
+        const tool = toolByName.get(call.tool);
+        let result;
+        if (!tool) {
+            result = `error: unknown tool "${call.tool}". Available: ${tools.map((t) => t.name).join(", ")}, ${FINISH_TOOL}.`;
+        }
+        else {
+            try {
+                result = await tool.run(call.args);
+            }
+            catch (err) {
+                result = `error running ${call.tool}: ${err.message}`;
+            }
+        }
+        const step = { index: i, tool: call.tool, args: call.args, result, thought: call.thought };
+        steps.push(step);
+        options.onStep?.(step);
+        messages.push({
+            role: "user",
+            content: `Result of ${call.tool}:\n${truncate(result)}\n\nChoose the next tool (or ${FINISH_TOOL}).`,
+        });
+    }
+    return {
+        finished: false,
+        summary: `Reached the ${maxSteps}-step limit without finishing. Last activity is in the step log; consider a narrower goal or running again.`,
+        steps,
+    };
+}
+//# sourceMappingURL=loop.js.map

package/dist/lib/orchestrator/loop.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"loop.js","sourceRoot":"","sources":["../../../src/lib/orchestrator/loop.ts"],"names":[],"mappings":";AAAA,qEAAqE;AACrE,EAAE;AACF,2FAA2F;AAC3F,wFAAwF;AACxF,4FAA4F;AAC5F,6FAA6F;AAC7F,6FAA6F;AAC7F,uCAAuC;AACvC,EAAE;AACF,yFAAyF;AACzF,0FAA0F;AAC1F,sFAAsF;;AA+GtF,sCAwDC;AAUD,oCAkHC;AA3OD,MAAM,WAAW,GAAG,QAAQ,CAAC;AAE7B,SAAS,iBAAiB,CAAC,IAAY,EAAE,KAAiB,EAAE,OAAgB;IAC1E,MAAM,QAAQ,GAAG,KAAK;SACnB,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;QACT,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC;aACxC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,WAAW,CAAC,KAAK,CAAC,EAAE,CAAC;aACrC,IAAI,CAAC,IAAI,CAAC,CAAC;QACd,OAAO,OAAO,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,WAAW,GAAG,MAAM,CAAC,CAAC,CAAC,KAAK,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;IACzE,CAAC,CAAC;SACD,IAAI,CAAC,IAAI,CAAC,CAAC;IAEd,OAAO;QACL,iHAAiH;QACjH,yFAAyF;QACzF,wFAAwF;QACxF,6FAA6F;QAC7F,OAAO,CAAC,CAAC,CAAC,eAAe,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE;QACvC,2FAA2F;QAC3F,oDAAoD;QACpD,4FAA4F;QAC5F,oBAAoB;QACpB,QAAQ;QACR,OAAO,WAAW,wFAAwF;QAC1G,UAAU;QACV,gDAAgD;QAChD,mEAAmE;QACnE,6CAA6C,WAAW,wBAAwB;QAChF,yCAAyC,WAAW,+BAA+B;QACnF,WAAW,IAAI,EAAE;KAClB;SACE,MAAM,CAAC,OAAO,CAAC;SACf,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC;AAED,SAAS,kBAAkB,CAAC,GAA4B;IACtD,IAAI,OAAO,GAAG,CAAC,IAAI,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC;IAC9C,MAAM,IAAI,GAAG,GAAG,CAAC,IAAI,IAAI,OAAO,GAAG,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAE,GAAG,CAAC,IAAgC,CAAC,CAAC,CAAC,EAAE,CAAC;IACnG,OAAO;QACL,OAAO,EAAE,OAAO,GAAG,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS;QAClE,IAAI,EAAE,GAAG,CAAC,IAAI;QACd,IAAI;KACL,CAAC;AACJ,CAAC;AAED;;;;;GAKG;AACH;6FAC6F;AAC7F,MAAM,mBAAmB,GAAG,MAAO,CAAC;AAEpC,SAAgB,aAAa,CAAC,GAAW;IACvC,IAAI,CAAC,GAAG;QAAE,OAAO,IAAI,CAAC;IACtB,IAAI,IAAI,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC;IACtB,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,+BAA+B,CAAC,CAAC;IAC1D,IAAI,KAAK;QAAE,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAElC,0BAA0B;IAC1B,IAAI,IAAI,CAAC,MAAM,IAAI,mBAAmB,EAAE,CAAC;QACvC,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAA4B,CAAC;YACxD,MAAM,IAAI,GAAG,kBAAkB,CAAC,GAAG,CAAC,CAAC;YACrC,IAAI,IAAI;gBAAE,OAAO,IAAI,CAAC;QACxB,CAAC;QAAC,MAAM,CAAC;YACP,+BAA+B;QACjC,CAAC;IACH,CAAC;IAED,2DAA2D;IAC3D,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,IAAI,QAAQ,GAAG,CAAC,CAAC,CAAC;IAClB,IAAI,KAAK,GAAG,KAAK,CAAC;IAClB,IAAI,OAAO,GAAG,KAAK,CAAC;IACpB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACrC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QACnB,IAAI,KAAK,EAAE,CAAC;YACV,IAAI,OAAO;gBAAE,OAAO,GAAG,KAAK,CAAC;iBACxB,IAAI,EAAE,KAAK,IAAI;gBAAE,OAAO,GAAG,IAAI,CAAC;iBAChC,IAAI,EAAE,KAAK,GAAG;gBAAE,KAAK,GAAG,KAAK,CAAC;YACnC,SAAS;QACX,CAAC;QACD,IAAI,EAAE,KAAK,GAAG;YAAE,KAAK,GAAG,IAAI,CAAC;aACxB,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;YACpB,IAAI,KAAK,KAAK,CAAC;gBAAE,QAAQ,GAAG,CAAC,CAAC;YAC9B,KAAK,EAAE,CAAC;QACV,CAAC;aAAM,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;YACtB,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;gBACd,KAAK,EAAE,CAAC;gBACR,IAAI,KAAK,KAAK,CAAC,IAAI,QAAQ,KAAK,CAAC,CAAC,EAAE,CAAC;oBACnC,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;oBAC9C,IAAI,SAAS,CAAC,MAAM,GAAG,mBAAmB,EAAE,CAAC;wBAC3C,QAAQ,GAAG,CAAC,CAAC,CAAC;wBACd,SAAS;oBACX,CAAC;oBACD,IAAI,CAAC;wBACH,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAA4B,CAAC;wBAC7D,MAAM,IAAI,GAAG,kBAAkB,CAAC,GAAG,CAAC,CAAC;wBACrC,IAAI,IAAI;4BAAE,OAAO,IAAI,CAAC;oBACxB,CAAC;oBAAC,MAAM,CAAC;wBACP,mBAAmB;oBACrB,CAAC;oBACD,QAAQ,GAAG,CAAC,CAAC,CAAC;gBAChB,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,QAAQ,CAAC,CAAS,EAAE,GAAG,GAAG,IAAI;IACrC,OAAO,CAAC,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,GAAG,iBAAiB,CAAC,CAAC,MAAM,GAAG,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC;AACzF,CAAC;AAED;;;GAGG;AACI,KAAK,UAAU,YAAY,CAAC,OAAuB;IACxD,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,EAAE,GAAG,OAAO,CAAC;IACpD,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,IAAI,EAAE,CAAC;IACxC,MAAM,eAAe,GAAG,OAAO,CAAC,eAAe,IAAI,CAAC,CAAC;IACrD,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,CAAC,CAAC,EAAU,EAAE,EAAE,CAAC,IAAI,OAAO,CAAO,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;IAC7F,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAE1D,mFAAmF;IACnF,wFAAwF;IACxF,gFAAgF;IAChF,MAAM,kBAAkB,GAAG,KAAK,EAAE,IAAmB,EAAmB,EAAE;QACxE,IAAI,OAAgB,CAAC;QACrB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,eAAe,EAAE,OAAO,EAAE,EAAE,CAAC;YAC5D,IAAI,CAAC;gBACH,OAAO,MAAM,SAAS,CAAC,IAAI,CAAC,CAAC;YAC/B,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,OAAO,GAAG,GAAG,CAAC;gBACd,IAAI,OAAO,GAAG,eAAe;oBAAE,MAAM,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC,IAAI,OAAO,EAAE,IAAI,CAAC,CAAC,CAAC;YACjF,CAAC;QACH,CAAC;QACD,MAAM,OAAO,YAAY,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC;IACxE,CAAC,CAAC;IAEF,MAAM,QAAQ,GAAkB;QAC9B,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,iBAAiB,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,EAAE;QACpE,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,gBAAgB,IAAI,EAAE,EAAE;KAClD,CAAC;IAEF,MAAM,KAAK,GAAe,EAAE,CAAC;IAC7B,IAAI,wBAAwB,GAAG,CAAC,CAAC;IACjC,MAAM,kBAAkB,GAAG,CAAC,CAAC;IAC7B,0FAA0F;IAC1F,uFAAuF;IACvF,MAAM,YAAY,GAAG,GAAG,CAAC;IAEzB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,EAAE,CAAC,EAAE,EAAE,CAAC;QAClC,IAAI,QAAQ,CAAC,MAAM,GAAG,YAAY,EAAE,CAAC;YACnC,OAAO;gBACL,QAAQ,EAAE,KAAK;gBACf,OAAO,EAAE,mCAAmC,YAAY,2EAA2E;gBACnI,KAAK;aACN,CAAC;QACJ,CAAC;QAED,IAAI,KAAa,CAAC;QAClB,IAAI,CAAC;YACH,KAAK,GAAG,MAAM,kBAAkB,CAAC,QAAQ,CAAC,CAAC;QAC7C,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO;gBACL,QAAQ,EAAE,KAAK;gBACf,OAAO,EAAE,6BAA6B,CAAC,GAAG,CAAC,UAAU,eAAe,GAAG,CAAC,gBAAiB,GAAa,CAAC,OAAO,EAAE;gBAChH,KAAK;aACN,CAAC;QACJ,CAAC;QAED,MAAM,IAAI,GAAG,aAAa,CAAC,KAAK,CAAC,CAAC;QAClC,IAAI,CAAC,IAAI,EAAE,CAAC;YACV,wFAAwF;YACxF,6EAA6E;YAC7E,wBAAwB,EAAE,CAAC;YAC3B,IAAI,wBAAwB,IAAI,kBAAkB,EAAE,CAAC;gBACnD,OAAO;oBACL,QAAQ,EAAE,KAAK;oBACf,OAAO,EAAE,6DAA6D,kBAAkB,uBAAuB,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE;oBACpI,KAAK;iBACN,CAAC;YACJ,CAAC;YACD,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC;YACrD,QAAQ,CAAC,IAAI,CAAC;gBACZ,IAAI,EAAE,MAAM;gBACZ,OAAO,EACL,6GAA6G;aAChH,CAAC,CAAC;YACH,SAAS;QACX,CAAC;QACD,wBAAwB,GAAG,CAAC,CAAC;QAE7B,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAEpE,IAAI,IAAI,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YAC9B,MAAM,OAAO,GACX,OAAO,IAAI,CAAC,IAAI,CAAC,OAAO,KAAK,QAAQ,IAAI,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE;gBAC/D,CAAC,CAAE,IAAI,CAAC,IAAI,CAAC,OAAkB;gBAC/B,CAAC,CAAC,OAAO,CAAC;YACd,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QAC5C,CAAC;QAED,MAAM,IAAI,GAAG,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACvC,IAAI,MAAc,CAAC;QACnB,IAAI,CAAC,IAAI,EAAE,CAAC;YACV,MAAM,GAAG,wBAAwB,IAAI,CAAC,IAAI,iBAAiB,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,WAAW,GAAG,CAAC;QACpH,CAAC;aAAM,CAAC;YACN,IAAI,CAAC;gBACH,MAAM,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACrC,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,MAAM,GAAG,iBAAiB,IAAI,CAAC,IAAI,KAAM,GAAa,CAAC,OAAO,EAAE,CAAC;YACnE,CAAC;QACH,CAAC;QAED,MAAM,IAAI,GAAa,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC;QACrG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACjB,OAAO,CAAC,MAAM,EAAE,CAAC,IAAI,CAAC,CAAC;QAEvB,QAAQ,CAAC,IAAI,CAAC;YACZ,IAAI,EAAE,MAAM;YACZ,OAAO,EAAE,aAAa,IAAI,CAAC,IAAI,MAAM,QAAQ,CAAC,MAAM,CAAC,gCAAgC,WAAW,IAAI;SACrG,CAAC,CAAC;IACL,CAAC;IAED,OAAO;QACL,QAAQ,EAAE,KAAK;QACf,OAAO,EAAE,eAAe,QAAQ,6GAA6G;QAC7I,KAAK;KACN,CAAC;AACJ,CAAC"}

package/dist/lib/orchestrator/supervisor.d.ts

@@ -0,0 +1,80 @@
+export type WorkerHarness = "claude" | "codex" | "cursor" | "custom";
+export type WorkerStatus = "running" | "exited" | "failed" | "stopped" | "unknown";
+export interface WorkerRecord {
+    id: string;
+    harness: WorkerHarness;
+    /** You.md project name / slug this worker is operating on (for reporting). */
+    project?: string;
+    /** Working directory the harness was launched in. */
+    cwd: string;
+    /** The task prompt handed to the worker. */
+    goal: string;
+    pid?: number;
+    status: WorkerStatus;
+    exitCode?: number;
+    startedAt: string;
+    endedAt?: string;
+    logFile: string;
+    /** Host that launched this worker (for cross-machine fleet views). */
+    host: string;
+    /** True once this worker's terminal transition has been reported (watch loop). */
+    reported?: boolean;
+}
+/** Terminal worker states — the worker is no longer doing work. */
+export declare const TERMINAL_WORKER_STATUSES: WorkerStatus[];
+export declare function isTerminalWorkerStatus(status: WorkerStatus): boolean;
+export declare function loadWorkers(): WorkerRecord[];
+/** True if a pid is alive (signal 0 probe). */
+export declare function isPidAlive(pid: number | undefined): boolean;
+/** Reconcile registry status against live pids (a running worker whose pid died → exited). */
+export declare function refreshWorkers(workers?: WorkerRecord[]): WorkerRecord[];
+export interface SpawnWorkerInput {
+    harness: WorkerHarness;
+    goal: string;
+    cwd: string;
+    project?: string;
+    host: string;
+    /** Required when harness === "custom": full argv, with {prompt} substituted in any element. */
+    customCommand?: string[];
+    /**
+     * Environment for the worker process. Defaults to the parent env for LOCAL spawns. Remote
+     * (bus-triggered) spawns MUST pass a minimized env so a remotely-launched agent does not
+     * inherit the daemon's secrets (you.md token, OPENROUTER_API_KEY, vault/folder keys).
+     */
+    env?: NodeJS.ProcessEnv;
+}
+export interface SpawnWorkerResult {
+    ok: boolean;
+    worker?: WorkerRecord;
+    error?: string;
+}
+/**
+ * Launch a worker harness detached, streaming its output to a per-worker log file, and record
+ * it in the registry. Detached + unref so the worker outlives this CLI process (an orchestrator
+ * dispatches and moves on; it does not block on the worker).
+ */
+export declare function spawnWorker(input: SpawnWorkerInput): SpawnWorkerResult;
+export declare function getWorker(id: string): WorkerRecord | undefined;
+/** Tail the last `lines` of a worker's captured output. */
+export declare function getWorkerOutput(id: string, lines?: number): {
+    ok: boolean;
+    output?: string;
+    error?: string;
+};
+/** Stop a running worker (SIGTERM the process group). */
+export declare function stopWorker(id: string): {
+    ok: boolean;
+    error?: string;
+};
+/**
+ * Reconcile status, then return workers that just reached a terminal state and have NOT yet been
+ * reported. Does NOT flip the `reported` flag — the caller marks each id reported only AFTER a
+ * successful post (via markReported), so a failed/transient post is retried next pass instead of
+ * silently dropping the completion. The "always on, report back" primitive for the watch loop.
+ */
+export declare function collectUnreportedCompletions(): WorkerRecord[];
+/** Mark the given worker ids as reported (call only after the completion was successfully posted). */
+export declare function markReported(ids: string[]): void;
+/** Drop exited/stopped workers older than `maxAgeMs` from the registry (keeps running ones). */
+export declare function pruneWorkers(maxAgeMs?: number): number;
+//# sourceMappingURL=supervisor.d.ts.map

package/dist/lib/orchestrator/supervisor.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"supervisor.d.ts","sourceRoot":"","sources":["../../../src/lib/orchestrator/supervisor.ts"],"names":[],"mappings":"AAeA,MAAM,MAAM,aAAa,GAAG,QAAQ,GAAG,OAAO,GAAG,QAAQ,GAAG,QAAQ,CAAC;AACrE,MAAM,MAAM,YAAY,GAAG,SAAS,GAAG,QAAQ,GAAG,QAAQ,GAAG,SAAS,GAAG,SAAS,CAAC;AAEnF,MAAM,WAAW,YAAY;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,aAAa,CAAC;IACvB,8EAA8E;IAC9E,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,qDAAqD;IACrD,GAAG,EAAE,MAAM,CAAC;IACZ,4CAA4C;IAC5C,IAAI,EAAE,MAAM,CAAC;IACb,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,YAAY,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,CAAC;IAChB,sEAAsE;IACtE,IAAI,EAAE,MAAM,CAAC;IACb,kFAAkF;IAClF,QAAQ,CAAC,EAAE,OAAO,CAAC;CACpB;AAED,mEAAmE;AACnE,eAAO,MAAM,wBAAwB,EAAE,YAAY,EAAoC,CAAC;AAExF,wBAAgB,sBAAsB,CAAC,MAAM,EAAE,YAAY,GAAG,OAAO,CAEpE;AAmCD,wBAAgB,WAAW,IAAI,YAAY,EAAE,CAiB5C;AAaD,+CAA+C;AAC/C,wBAAgB,UAAU,CAAC,GAAG,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,CAS3D;AAED,8FAA8F;AAC9F,wBAAgB,cAAc,CAAC,OAAO,GAAE,YAAY,EAAkB,GAAG,YAAY,EAAE,CAWtF;AASD,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,aAAa,CAAC;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,MAAM,CAAC;IACZ,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,+FAA+F;IAC/F,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB;;;;OAIG;IACH,GAAG,CAAC,EAAE,MAAM,CAAC,UAAU,CAAC;CACzB;AAED,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,OAAO,CAAC;IACZ,MAAM,CAAC,EAAE,YAAY,CAAC;IACtB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED;;;;GAIG;AACH,wBAAgB,WAAW,CAAC,KAAK,EAAE,gBAAgB,GAAG,iBAAiB,CAoFtE;AAED,wBAAgB,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,YAAY,GAAG,SAAS,CAE9D;AAED,2DAA2D;AAC3D,wBAAgB,eAAe,CAAC,EAAE,EAAE,MAAM,EAAE,KAAK,SAAK,GAAG;IAAE,EAAE,EAAE,OAAO,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,CAUxG;AAED,yDAAyD;AACzD,wBAAgB,UAAU,CAAC,EAAE,EAAE,MAAM,GAAG;IAAE,EAAE,EAAE,OAAO,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,CAqBtE;AAED;;;;;GAKG;AACH,wBAAgB,4BAA4B,IAAI,YAAY,EAAE,CAG7D;AAED,sGAAsG;AACtG,wBAAgB,YAAY,CAAC,GAAG,EAAE,MAAM,EAAE,GAAG,IAAI,CAYhD;AAED,gGAAgG;AAChG,wBAAgB,YAAY,CAAC,QAAQ,SAAsB,GAAG,MAAM,CAWnE"}

package/dist/lib/orchestrator/supervisor.js

@@ -0,0 +1,315 @@
+"use strict";
+// supervisor.ts — process supervisor for the You agent orchestrator.
+//
+// The You agent is a MASTER ORCHESTRATOR, not a worker/coding agent. It launches, monitors,
+// and stops other agentic harnesses (Claude Code, Codex, Cursor, etc.) — each best at a
+// given task — in real project directories, captures their output, and reports back. This
+// module is the deterministic substrate for that: spawn / list / tail / stop, persisted in a
+// registry so an always-on daemon (or the next CLI invocation) can see the whole fleet.
+//
+// No LLM in this file. It is pure, testable process management; the loop (loop.ts) drives it.
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.TERMINAL_WORKER_STATUSES = void 0;
+exports.isTerminalWorkerStatus = isTerminalWorkerStatus;
+exports.loadWorkers = loadWorkers;
+exports.isPidAlive = isPidAlive;
+exports.refreshWorkers = refreshWorkers;
+exports.spawnWorker = spawnWorker;
+exports.getWorker = getWorker;
+exports.getWorkerOutput = getWorkerOutput;
+exports.stopWorker = stopWorker;
+exports.collectUnreportedCompletions = collectUnreportedCompletions;
+exports.markReported = markReported;
+exports.pruneWorkers = pruneWorkers;
+const child_process = __importStar(require("child_process"));
+const fs = __importStar(require("fs"));
+const path = __importStar(require("path"));
+const config_1 = require("../config");
+/** Terminal worker states — the worker is no longer doing work. */
+exports.TERMINAL_WORKER_STATUSES = ["exited", "stopped", "failed"];
+function isTerminalWorkerStatus(status) {
+    return exports.TERMINAL_WORKER_STATUSES.includes(status);
+}
+const HARNESS_SPECS = {
+    // Claude Code headless print mode.
+    claude: { bin: "claude", args: (p) => ["-p", p], overrideEnv: "YOU_HARNESS_CLAUDE" },
+    // Codex non-interactive exec mode.
+    codex: { bin: "codex", args: (p) => ["exec", p], overrideEnv: "YOU_HARNESS_CODEX" },
+    // Cursor agent CLI headless mode.
+    cursor: { bin: "cursor-agent", args: (p) => ["-p", p], overrideEnv: "YOU_HARNESS_CURSOR" },
+};
+function orchestratorDir() {
+    const dir = path.join((0, config_1.getWritableHomeBundleDir)(), "orchestrator");
+    fs.mkdirSync(dir, { recursive: true });
+    return dir;
+}
+function logsDir() {
+    const dir = path.join(orchestratorDir(), "logs");
+    fs.mkdirSync(dir, { recursive: true });
+    return dir;
+}
+function registryPath() {
+    return path.join(orchestratorDir(), "workers.json");
+}
+function loadWorkers() {
+    try {
+        const raw = fs.readFileSync(registryPath(), "utf-8");
+        const parsed = JSON.parse(raw);
+        if (!Array.isArray(parsed))
+            return [];
+        // Drop malformed records (partial write, hand-edit, schema drift) instead of letting a
+        // single bad entry throw downstream — a corrupt row shouldn't hide every healthy worker.
+        return parsed.filter((w) => !!w &&
+            typeof w === "object" &&
+            typeof w.id === "string" &&
+            typeof w.status === "string");
+    }
+    catch {
+        return [];
+    }
+}
+function saveWorkers(workers) {
+    // Atomic write (tmp + rename) so a concurrent reader never sees a half-written file.
+    // NOTE: this does not prevent lost updates under concurrent writers (the registry is a
+    // single JSON file with no lock); spawn/watch/remote runs are low-frequency so the window
+    // is small, but a future hardening is per-record files or a lockfile.
+    const target = registryPath();
+    const tmp = `${target}.${process.pid}.tmp`;
+    fs.writeFileSync(tmp, JSON.stringify(workers, null, 2));
+    fs.renameSync(tmp, target);
+}
+/** True if a pid is alive (signal 0 probe). */
+function isPidAlive(pid) {
+    if (!pid || pid <= 0)
+        return false;
+    try {
+        process.kill(pid, 0);
+        return true;
+    }
+    catch (err) {
+        // ESRCH = no such process; EPERM = exists but not ours (still alive).
+        return err.code === "EPERM";
+    }
+}
+/** Reconcile registry status against live pids (a running worker whose pid died → exited). */
+function refreshWorkers(workers = loadWorkers()) {
+    let changed = false;
+    for (const w of workers) {
+        if (w.status === "running" && !isPidAlive(w.pid)) {
+            w.status = "exited";
+            w.endedAt = w.endedAt ?? new Date().toISOString();
+            changed = true;
+        }
+    }
+    if (changed)
+        saveWorkers(workers);
+    return workers;
+}
+function shortId() {
+    // ULID-ish without external deps: time + random suffix. Not crypto; just a unique handle.
+    const rand = Math.floor(Math.random() * 0xffffff).toString(16).padStart(6, "0");
+    const stamp = Date.now().toString(36);
+    return `w_${stamp}${rand}`;
+}
+/**
+ * Launch a worker harness detached, streaming its output to a per-worker log file, and record
+ * it in the registry. Detached + unref so the worker outlives this CLI process (an orchestrator
+ * dispatches and moves on; it does not block on the worker).
+ */
+function spawnWorker(input) {
+    const { harness, goal, cwd, project, host } = input;
+    if (!fs.existsSync(cwd) || !fs.statSync(cwd).isDirectory()) {
+        return { ok: false, error: `working directory does not exist: ${cwd}` };
+    }
+    let bin;
+    let argv;
+    if (harness === "custom") {
+        if (!input.customCommand || input.customCommand.length === 0) {
+            return { ok: false, error: "custom harness requires customCommand argv" };
+        }
+        const subst = input.customCommand.map((part) => part.replace(/\{prompt\}/g, goal));
+        bin = subst[0];
+        argv = subst.slice(1);
+    }
+    else {
+        const spec = HARNESS_SPECS[harness];
+        const override = process.env[spec.overrideEnv];
+        bin = override && override.trim() ? override.trim() : spec.bin;
+        argv = spec.args(goal);
+    }
+    const id = shortId();
+    const logFile = path.join(logsDir(), `${id}.log`);
+    let pid;
+    try {
+        const out = fs.openSync(logFile, "a");
+        fs.writeSync(out, `# you orchestrator worker ${id}\n# harness=${harness} project=${project ?? "-"} cwd=${cwd}\n# goal: ${goal}\n# started: ${new Date().toISOString()}\n\n`);
+        const child = child_process.spawn(bin, argv, {
+            cwd,
+            detached: true,
+            stdio: ["ignore", out, out],
+            env: input.env ?? process.env,
+        });
+        pid = child.pid;
+        // A missing/unlaunchable binary surfaces ASYNCHRONOUSLY via 'error' (ENOENT), after spawn
+        // returns. Without this handler Node throws it as an uncaught exception, and the registry
+        // would keep a phantom "running" record. Mark the worker failed and never let it crash us.
+        child.on("error", (err) => {
+            try {
+                fs.appendFileSync(logFile, `\n# launch error: ${err.message}\n`);
+            }
+            catch {
+                /* ignore */
+            }
+            const all = loadWorkers();
+            const rec = all.find((w) => w.id === id);
+            if (rec && rec.status === "running") {
+                rec.status = "failed";
+                rec.endedAt = new Date().toISOString();
+                saveWorkers(all);
+            }
+        });
+        child.unref();
+        fs.closeSync(out);
+    }
+    catch (err) {
+        return {
+            ok: false,
+            error: `failed to launch ${bin}: ${err.message} (is the harness installed and on PATH?)`,
+        };
+    }
+    const worker = {
+        id,
+        harness,
+        project,
+        cwd,
+        goal,
+        pid,
+        status: pid ? "running" : "failed",
+        startedAt: new Date().toISOString(),
+        logFile,
+        host,
+    };
+    const workers = loadWorkers();
+    workers.push(worker);
+    saveWorkers(workers);
+    return { ok: true, worker };
+}
+function getWorker(id) {
+    return refreshWorkers().find((w) => w.id === id);
+}
+/** Tail the last `lines` of a worker's captured output. */
+function getWorkerOutput(id, lines = 40) {
+    const worker = getWorker(id);
+    if (!worker)
+        return { ok: false, error: `no worker with id ${id}` };
+    try {
+        const raw = fs.readFileSync(worker.logFile, "utf-8");
+        const tail = raw.split(/\r?\n/).slice(-lines).join("\n");
+        return { ok: true, output: tail };
+    }
+    catch (err) {
+        return { ok: false, error: `could not read log: ${err.message}` };
+    }
+}
+/** Stop a running worker (SIGTERM the process group). */
+function stopWorker(id) {
+    const workers = refreshWorkers();
+    const worker = workers.find((w) => w.id === id);
+    if (!worker)
+        return { ok: false, error: `no worker with id ${id}` };
+    if (worker.status !== "running")
+        return { ok: true };
+    try {
+        if (worker.pid && isPidAlive(worker.pid)) {
+            // Negative pid → kill the detached process group.
+            try {
+                process.kill(-worker.pid, "SIGTERM");
+            }
+            catch {
+                process.kill(worker.pid, "SIGTERM");
+            }
+        }
+        worker.status = "stopped";
+        worker.endedAt = new Date().toISOString();
+        saveWorkers(workers);
+        return { ok: true };
+    }
+    catch (err) {
+        return { ok: false, error: `failed to stop: ${err.message}` };
+    }
+}
+/**
+ * Reconcile status, then return workers that just reached a terminal state and have NOT yet been
+ * reported. Does NOT flip the `reported` flag — the caller marks each id reported only AFTER a
+ * successful post (via markReported), so a failed/transient post is retried next pass instead of
+ * silently dropping the completion. The "always on, report back" primitive for the watch loop.
+ */
+function collectUnreportedCompletions() {
+    const workers = refreshWorkers();
+    return workers.filter((w) => isTerminalWorkerStatus(w.status) && !w.reported);
+}
+/** Mark the given worker ids as reported (call only after the completion was successfully posted). */
+function markReported(ids) {
+    if (ids.length === 0)
+        return;
+    const set = new Set(ids);
+    const workers = loadWorkers();
+    let changed = false;
+    for (const w of workers) {
+        if (set.has(w.id) && !w.reported) {
+            w.reported = true;
+            changed = true;
+        }
+    }
+    if (changed)
+        saveWorkers(workers);
+}
+/** Drop exited/stopped workers older than `maxAgeMs` from the registry (keeps running ones). */
+function pruneWorkers(maxAgeMs = 24 * 60 * 60 * 1000) {
+    const now = Date.now();
+    const workers = refreshWorkers();
+    const kept = workers.filter((w) => {
+        if (w.status === "running")
+            return true;
+        const ended = w.endedAt ? Date.parse(w.endedAt) : Date.parse(w.startedAt);
+        return now - ended < maxAgeMs;
+    });
+    const removed = workers.length - kept.length;
+    if (removed > 0)
+        saveWorkers(kept);
+    return removed;
+}
+//# sourceMappingURL=supervisor.js.map