@elizaos/agent 2.0.0-alpha.144 → 2.0.0-alpha.151

package/packages/typescript/src/features/orchestrator/services/stall-classifier.js

@@ -1,446 +0,0 @@
-/**
- * Stall classification subsystem — determines what a "stalled" task agent
- * session is doing (finished, waiting for input, still working, or errored).
- *
- * Extracted as standalone functions that receive dependencies as parameters,
- * making them easy to test without coupling to PTYService.
- *
- * @module services/stall-classifier
- */
-import { ModelType } from "@elizaos/core";
-import { buildTaskCompletionTimeline, extractTaskCompletionTraceRecords, } from "pty-manager";
-import { stripAnsi } from "./ansi-utils.js";
-import { withTrajectoryContext } from "./trajectory-context.js";
-const STATUS_NOISE_LINE = /messages to be submitted after next tool call|working \(\d+s .*esc to interrupt\)|\b\d+% left\b|context left|use \/skills to list available skills/i;
-const STATUS_PATH_LINE = /(\/private\/|\/var\/folders\/|\/Users\/|\/tmp\/)/;
-const SPINNER_FRAGMENT_TOKEN = /^(?:w|wo|wor|work|worki|workin|working|orking|rking|king|ing|ng|g|\d+|[•·])$/i;
-function normalizeForComparison(value) {
-    return stripAnsi(value).replace(/\s+/g, " ").trim().toLowerCase();
-}
-function looksLikeSpinnerFragments(line) {
-    const tokens = line
-        .replace(/[^\w/%@.:\-/ ]+/g, " ")
-        .split(/\s+/)
-        .filter(Boolean);
-    if (tokens.length === 0)
-        return false;
-    const fragmentTokens = tokens.filter((token) => SPINNER_FRAGMENT_TOKEN.test(token));
-    return (fragmentTokens.length >= 4 &&
-        fragmentTokens.length >= Math.ceil(tokens.length * 0.6));
-}
-function isStatusNoiseLine(line) {
-    const compact = line.replace(/\s+/g, " ").trim();
-    if (!compact)
-        return true;
-    if (compact.startsWith("› "))
-        return true;
-    if (STATUS_NOISE_LINE.test(compact))
-        return true;
-    if (looksLikeSpinnerFragments(compact))
-        return true;
-    if (STATUS_PATH_LINE.test(compact) && /\b\d+% left\b/i.test(compact))
-        return true;
-    if (STATUS_PATH_LINE.test(compact) && looksLikeSpinnerFragments(compact))
-        return true;
-    return false;
-}
-function sanitizeOutputForClassification(output, lastSentInput) {
-    const normalizedInput = lastSentInput
-        ? normalizeForComparison(lastSentInput)
-        : "";
-    let removedEchoLines = 0;
-    let removedStatusLines = 0;
-    const sanitized = stripAnsi(output)
-        .split("\n")
-        .map((line) => line.replace(/\s+/g, " ").trim())
-        .filter((line) => {
-        if (!line)
-            return false;
-        const normalizedLine = line.toLowerCase();
-        if (normalizedInput &&
-            normalizedLine.length >= 12 &&
-            normalizedInput.includes(normalizedLine)) {
-            removedEchoLines += 1;
-            return false;
-        }
-        if (isStatusNoiseLine(line)) {
-            removedStatusLines += 1;
-            return false;
-        }
-        return true;
-    })
-        .join("\n")
-        .trim();
-    return { sanitized, removedEchoLines, removedStatusLines };
-}
-function promptLooksLikeFalseBlockedNoise(prompt, lastSentInput) {
-    if (!prompt)
-        return false;
-    const normalizedPrompt = normalizeForComparison(prompt);
-    if (!normalizedPrompt)
-        return false;
-    if (lastSentInput) {
-        const normalizedInput = normalizeForComparison(lastSentInput);
-        if (normalizedPrompt.length >= 12 &&
-            normalizedInput.includes(normalizedPrompt)) {
-            return true;
-        }
-    }
-    return isStatusNoiseLine(prompt) || looksLikeSpinnerFragments(prompt);
-}
-/**
- * Build the LLM system prompt used to classify stalled output.
- */
-export function buildStallClassificationPrompt(agentType, sessionId, output) {
-    return (`You are Eliza, an AI orchestrator managing task-agent sessions. ` +
-        `A ${agentType} task agent (session: ${sessionId}) appears to have stalled — ` +
-        `it has stopped producing output while in a busy state.\n\n` +
-        `Here is the recent terminal output:\n` +
-        `---\n${output.slice(-1500)}\n---\n\n` +
-        `Classify what's happening. Read the output carefully and choose the MOST specific match:\n\n` +
-        `1. "task_complete" — The agent FINISHED its task and returned to its idle prompt. ` +
-        `Strong indicators: a summary of completed work ("Done", "All done", "Here's what was completed"), ` +
-        `timing info ("Baked for", "Churned for", "Crunched for", "Cooked for", "Worked for"), ` +
-        `or the agent's main prompt symbol (❯) appearing AFTER completion output. ` +
-        `If the output contains evidence of completed work followed by an idle prompt, this is ALWAYS task_complete, ` +
-        `even though the agent is technically "waiting" — it is waiting for a NEW task, not asking a question.\n\n` +
-        `2. "waiting_for_input" — The agent is MID-TASK and blocked on a specific question or permission prompt. ` +
-        `The agent has NOT finished its work — it needs a response to continue. ` +
-        `Examples: Y/n confirmation, file permission dialogs, "Do you want to proceed?", ` +
-        `tool approval prompts, or interactive menus. ` +
-        `This is NOT the same as the agent sitting at its idle prompt after finishing work.\n\n` +
-        `3. "still_working" — The agent is actively processing (API call, compilation, thinking, etc.) ` +
-        `and has not produced final output yet. No prompt or completion summary visible.\n\n` +
-        `4. "error" — The agent hit an error state (crash, unrecoverable error, stack trace).\n\n` +
-        `5. "tool_running" — The agent is using an external tool (browser automation, ` +
-        `MCP tool, etc.). Indicators: "Claude in Chrome", "javascript_tool", ` +
-        `"computer_tool", "screenshot", "navigate", tool execution output. ` +
-        `The agent is actively working but the terminal may be quiet.\n\n` +
-        `IMPORTANT: If you see BOTH completed work output AND an idle prompt (❯), choose "task_complete". ` +
-        `Only choose "waiting_for_input" if the agent is clearly asking a question mid-task. ` +
-        `Ignore echoed user input, copied prior transcripts, spinner fragments, and status rows like ` +
-        `"Working (12s • esc to interrupt)" or "97% left" — those mean the agent is still working, not blocked.\n\n` +
-        `If "waiting_for_input", also provide:\n` +
-        `- "prompt": the text of what it's asking\n` +
-        `- "suggestedResponse": what to type/send. Use "keys:enter" for TUI menu confirmation, ` +
-        `"keys:down,enter" to select a non-default option, or plain text like "y" for text prompts.\n\n` +
-        `Respond with ONLY a JSON object:\n` +
-        `{"state": "...", "prompt": "...", "suggestedResponse": "..."}`);
-}
-/**
- * Write a debug snapshot to ~/.eliza/debug/ for offline stall analysis.
- */
-export async function writeStallSnapshot(sessionId, agentType, recentOutput, effectiveOutput, buffers, traceEntries, log) {
-    try {
-        const fs = await import("node:fs");
-        const os = await import("node:os");
-        const path = await import("node:path");
-        const snapshotDir = path.join(os.homedir(), ".eliza", "debug");
-        fs.mkdirSync(snapshotDir, { recursive: true });
-        const ourBuffer = buffers.get(sessionId);
-        const ourTail = ourBuffer
-            ? ourBuffer.slice(-100).join("\n")
-            : "(no buffer)";
-        void ourTail; // used in snapshot context but not directly printed
-        let traceTimeline = "(no trace entries)";
-        try {
-            const records = extractTaskCompletionTraceRecords(traceEntries);
-            const timeline = buildTaskCompletionTimeline(records, {
-                adapterType: agentType,
-            });
-            traceTimeline = JSON.stringify(timeline, null, 2);
-        }
-        catch (e) {
-            traceTimeline = `(trace error: ${e})`;
-        }
-        const snapshot = [
-            `=== STALL SNAPSHOT @ ${new Date().toISOString()} ===`,
-            `Session: ${sessionId} | Agent: ${agentType}`,
-            `recentOutput length: ${recentOutput.length} | effectiveOutput length: ${effectiveOutput.length}`,
-            ``,
-            `--- effectiveOutput (what LLM sees) ---`,
-            effectiveOutput.slice(-1500),
-            ``,
-            `--- trace timeline ---`,
-            traceTimeline,
-            ``,
-            `--- raw trace entries (last 20 of ${traceEntries.length}) ---`,
-            traceEntries.slice(-20).join("\n"),
-            ``,
-        ].join("\n");
-        const snapshotPath = path.join(snapshotDir, `stall-snapshot-${sessionId}.txt`);
-        fs.writeFileSync(snapshotPath, snapshot);
-        log(`Stall snapshot → ${snapshotPath}`);
-    }
-    catch (_) {
-        /* best-effort */
-    }
-}
-/**
- * Main stall classification logic. Determines what a stalled session is doing
- * by checking the buffer, building a prompt, and asking the LLM.
- */
-export async function classifyStallOutput(ctx) {
-    const { sessionId, recentOutput, agentType, buffers, traceEntries, runtime, metricsTracker, log, } = ctx;
-    metricsTracker.incrementStalls(agentType);
-    // Use our own buffer if pty-manager's recentOutput is empty or too short.
-    let effectiveOutput = recentOutput;
-    if (!recentOutput || recentOutput.trim().length < 200) {
-        const ourBuffer = buffers.get(sessionId);
-        if (ourBuffer && ourBuffer.length > 0) {
-            const rawTail = ourBuffer.slice(-100).join("\n");
-            const stripped = stripAnsi(rawTail);
-            if (stripped.length > effectiveOutput.length) {
-                effectiveOutput = stripped;
-                log(`Using own buffer for stall classification (${effectiveOutput.length} chars after stripping, pty-manager had ${recentOutput.length})`);
-            }
-        }
-    }
-    const { sanitized: sanitizedOutput, removedEchoLines, removedStatusLines, } = sanitizeOutputForClassification(effectiveOutput, ctx.lastSentInput);
-    if (removedEchoLines > 0 || removedStatusLines > 0) {
-        log(`Sanitized stall output for ${sessionId}: removed ${removedEchoLines} echoed lines and ${removedStatusLines} status lines`);
-    }
-    if (!sanitizedOutput && removedEchoLines + removedStatusLines > 0) {
-        log(`Stall classification short-circuit for ${sessionId}: only echoed input / status noise remained`);
-        return { state: "still_working" };
-    }
-    const systemPrompt = buildStallClassificationPrompt(agentType, sessionId, sanitizedOutput || effectiveOutput);
-    // Dump debug snapshot for offline analysis (opt-in via PTYServiceConfig.debug)
-    if (ctx.debugSnapshots) {
-        await writeStallSnapshot(sessionId, agentType, recentOutput, effectiveOutput, buffers, traceEntries, log);
-    }
-    try {
-        log(`Stall detected for ${sessionId}, asking LLM to classify...`);
-        const result = await withTrajectoryContext(runtime, {
-            source: "orchestrator",
-            decisionType: "stall-classification",
-            sessionId,
-        }, () => runtime.useModel(ModelType.TEXT_SMALL, { prompt: systemPrompt }));
-        const jsonMatch = result.match(/\{[\s\S]*\}/);
-        if (!jsonMatch) {
-            log(`Stall classification: no JSON in LLM response`);
-            return null;
-        }
-        const parsed = JSON.parse(jsonMatch[0]);
-        const validStates = [
-            "waiting_for_input",
-            "still_working",
-            "task_complete",
-            "error",
-            "tool_running",
-        ];
-        if (!validStates.includes(parsed.state)) {
-            log(`Stall classification: invalid state "${parsed.state}"`);
-            return null;
-        }
-        // Map tool_running → still_working (StallClassification doesn't have tool_running).
-        //
-        // Also downgrade task_complete → still_working. The stall classifier LLM
-        // guesses from raw stripped-ANSI terminal buffer text, which cannot
-        // reliably distinguish "agent is truly finished" from "shell prompt
-        // showed up between tool calls" or "intermediate command exited with a
-        // summary". Treating this LLM's task_complete guess as a completion
-        // signal causes the coordinator's turn-complete pipeline to fire
-        // mid-work and re-inject the original prompt, creating an infinite
-        // retry loop for long open-ended tasks.
-        //
-        // Task completion is signaled authoritatively by the agent's own hook
-        // system (routed through pty-service.handleHookEvent) and by the
-        // jsonl-based completion watcher in the eliza package. The stall
-        // classifier is still useful for detecting waiting_for_input and error
-        // states, which is why we don't short-circuit those paths.
-        let mappedState;
-        if (parsed.state === "tool_running" || parsed.state === "task_complete") {
-            mappedState = "still_working";
-            if (parsed.state === "task_complete") {
-                log(`Stall classification for ${sessionId}: LLM said task_complete — downgrading to still_working (authoritative completion comes from hooks, not buffer guessing)`);
-            }
-        }
-        else {
-            mappedState = parsed.state;
-        }
-        const classification = {
-            state: mappedState,
-            prompt: parsed.prompt,
-            suggestedResponse: parsed.suggestedResponse,
-        };
-        if (classification.state === "waiting_for_input" &&
-            promptLooksLikeFalseBlockedNoise(classification.prompt, ctx.lastSentInput)) {
-            log(`Stall classification override for ${sessionId}: prompt looked like echoed input / status noise`);
-            return { state: "still_working" };
-        }
-        log(`Stall classification for ${sessionId}: ${classification.state}${classification.suggestedResponse ? ` → "${classification.suggestedResponse}"` : ""}`);
-        return classification;
-    }
-    catch (err) {
-        log(`Stall classification failed: ${err}`);
-        return null;
-    }
-}
-/**
- * Build a combined prompt that classifies the stall AND decides how to respond,
- * merging stall classification with coordinator decision guidelines.
- *
- * Used for coordinator-managed sessions in autonomous mode to eliminate the
- * redundant second LLM call in the coordinator's handleBlocked path.
- */
-export function buildCombinedClassifyDecidePrompt(agentType, sessionId, output, taskContext, decisionHistory) {
-    const historySection = decisionHistory.length > 0
-        ? `\nPrevious decisions for this session:\n${decisionHistory
-            .slice(-5)
-            .map((d, i) => `  ${i + 1}. [${d.event}] prompt="${d.promptText}" → ${d.action}${d.response ? ` ("${d.response}")` : ""} — ${d.reasoning}`)
-            .join("\n")}\n`
-        : "";
-    return (`You are Eliza, an AI orchestrator managing task-agent sessions. ` +
-        `A ${agentType} task agent (session: ${sessionId}) appears to have stalled — ` +
-        `it has stopped producing output while in a busy state.\n\n` +
-        `Original task: "${taskContext.originalTask}"\n` +
-        `Working directory: ${taskContext.workdir}\n` +
-        `Repository: ${taskContext.repo ?? "none (scratch directory)"}\n` +
-        historySection +
-        `\nHere is the recent terminal output:\n` +
-        `---\n${output.slice(-1500)}\n---\n\n` +
-        `Classify what's happening AND decide how to respond. Read the output carefully.\n\n` +
-        `Classification states:\n\n` +
-        `1. "task_complete" — The agent FINISHED its task and returned to its idle prompt. ` +
-        `Strong indicators: a summary of completed work, timing info, ` +
-        `or the agent's main prompt symbol (❯) appearing AFTER completion output.\n\n` +
-        `2. "waiting_for_input" — The agent is MID-TASK and blocked on a specific question or permission prompt. ` +
-        `Examples: Y/n confirmation, file permission dialogs, tool approval prompts, interactive menus.\n\n` +
-        `3. "still_working" — The agent is actively processing (API call, compilation, thinking). ` +
-        `No prompt or completion summary visible.\n\n` +
-        `4. "error" — The agent hit an error state (crash, unrecoverable error, stack trace).\n\n` +
-        `5. "tool_running" — The agent is using an external tool (browser automation, MCP tool, etc.).\n\n` +
-        `Ignore echoed user input, copied prior transcripts, spinner fragments, and status rows like ` +
-        `"Working (12s • esc to interrupt)" or "97% left" — those indicate active work, not a live prompt.\n\n` +
-        `If "waiting_for_input", you must also decide how to respond. Guidelines:\n` +
-        `- IMPORTANT: If the prompt asks to approve access to files or directories OUTSIDE the working ` +
-        `directory (${taskContext.workdir}), DECLINE the request. Respond with "n" and tell the agent: ` +
-        `"That path is outside your workspace. Use ${taskContext.workdir} instead."\n` +
-        `- For tool approval prompts (file writes, shell commands), respond "y" or use "keys:enter".\n` +
-        `- For Y/n confirmations that align with the original task, respond "y".\n` +
-        `- For TUI menus, use "keys:enter" for default or "keys:down,enter" for non-default.\n` +
-        `- If the prompt asks for information NOT in the original task, set suggestedResponse to null ` +
-        `(this will escalate to the human).\n` +
-        `- If a PR was just created, the task is likely done — classify as "task_complete".\n\n` +
-        `Respond with ONLY a JSON object:\n` +
-        `{"state": "...", "prompt": "...", "suggestedResponse": "..."}`);
-}
-/**
- * Combined classify-and-decide for coordinator-managed autonomous sessions.
- *
- * Performs classification AND coordinator-quality response decision in a single
- * LLM call. The suggestedResponse is kept intact (not stripped), so pty-manager
- * auto-responds and the coordinator receives autoResponded: true — skipping
- * the second LLM call in handleBlocked().
- */
-export async function classifyAndDecideForCoordinator(ctx) {
-    const { sessionId, recentOutput, agentType, buffers, traceEntries, runtime, manager, metricsTracker, taskContext, decisionHistory = [], log, } = ctx;
-    metricsTracker.incrementStalls(agentType);
-    // Buffer fallback — same logic as classifyStallOutput
-    let effectiveOutput = recentOutput;
-    if (!recentOutput || recentOutput.trim().length < 200) {
-        const ourBuffer = buffers.get(sessionId);
-        if (ourBuffer && ourBuffer.length > 0) {
-            const rawTail = ourBuffer.slice(-100).join("\n");
-            const stripped = stripAnsi(rawTail);
-            if (stripped.length > effectiveOutput.length) {
-                effectiveOutput = stripped;
-                log(`Using own buffer for combined classify+decide (${effectiveOutput.length} chars after stripping, pty-manager had ${recentOutput.length})`);
-            }
-        }
-    }
-    const { sanitized: sanitizedOutput, removedEchoLines, removedStatusLines, } = sanitizeOutputForClassification(effectiveOutput, ctx.lastSentInput);
-    if (removedEchoLines > 0 || removedStatusLines > 0) {
-        log(`Sanitized combined stall output for ${sessionId}: removed ${removedEchoLines} echoed lines and ${removedStatusLines} status lines`);
-    }
-    if (!sanitizedOutput && removedEchoLines + removedStatusLines > 0) {
-        log(`Combined classify+decide short-circuit for ${sessionId}: only echoed input / status noise remained`);
-        return { state: "still_working" };
-    }
-    const systemPrompt = buildCombinedClassifyDecidePrompt(agentType, sessionId, sanitizedOutput || effectiveOutput, taskContext, decisionHistory);
-    if (ctx.debugSnapshots) {
-        await writeStallSnapshot(sessionId, agentType, recentOutput, effectiveOutput, buffers, traceEntries, log);
-    }
-    try {
-        log(`Stall detected for coordinator-managed ${sessionId}, combined classify+decide...`);
-        const result = await withTrajectoryContext(runtime, {
-            source: "orchestrator",
-            decisionType: "stall-classify-decide",
-            sessionId,
-            taskLabel: taskContext.label,
-            repo: taskContext.repo,
-            workdir: taskContext.workdir,
-            originalTask: taskContext.originalTask,
-        }, () => runtime.useModel(ModelType.TEXT_SMALL, { prompt: systemPrompt }));
-        const jsonMatch = result.match(/\{[\s\S]*\}/);
-        if (!jsonMatch) {
-            log(`Combined classify+decide: no JSON in LLM response`);
-            return null;
-        }
-        const parsed = JSON.parse(jsonMatch[0]);
-        const validStates = [
-            "waiting_for_input",
-            "still_working",
-            "task_complete",
-            "error",
-            "tool_running",
-        ];
-        if (!validStates.includes(parsed.state)) {
-            log(`Combined classify+decide: invalid state "${parsed.state}"`);
-            return null;
-        }
-        // Same downgrade rationale as classifyStallOutput: the LLM's task_complete
-        // guess from buffer text is unreliable on long multi-step tasks. Authoritative
-        // completion comes from the agent's hook system (pty-service.handleHookEvent)
-        // and the jsonl-based completion watcher in the eliza package.
-        let mappedState;
-        if (parsed.state === "tool_running" || parsed.state === "task_complete") {
-            mappedState = "still_working";
-            if (parsed.state === "task_complete") {
-                log(`Combined classify+decide for ${sessionId}: LLM said task_complete — downgrading to still_working (authoritative completion comes from hooks)`);
-            }
-        }
-        else {
-            mappedState = parsed.state;
-        }
-        // Deterministic safety guard: if the LLM approved access to a path
-        // outside the workspace, override with a decline. This runs before
-        // pty-manager auto-responds, so the unsafe approval never reaches the agent.
-        if (mappedState === "waiting_for_input" && parsed.suggestedResponse) {
-            const promptText = typeof parsed.prompt === "string" ? parsed.prompt : "";
-            const responseText = parsed.suggestedResponse.trim().toLowerCase();
-            const approving = ["y", "yes", "keys:enter", "keys:down,enter"].includes(responseText);
-            const hasAbsPath = /(?:^|[\s"'`])\/[^\s"'`]+/.test(promptText);
-            if (approving &&
-                hasAbsPath &&
-                !promptText.includes(taskContext.workdir)) {
-                log(`Combined classify+decide: overriding out-of-scope approval for ${sessionId}`);
-                parsed.suggestedResponse = `n — That path is outside your workspace. Use ${taskContext.workdir} instead.`;
-            }
-        }
-        const classification = {
-            state: mappedState,
-            prompt: parsed.prompt,
-            suggestedResponse: parsed.suggestedResponse,
-        };
-        if (classification.state === "waiting_for_input" &&
-            promptLooksLikeFalseBlockedNoise(classification.prompt, ctx.lastSentInput)) {
-            log(`Combined classify+decide override for ${sessionId}: prompt looked like echoed input / status noise`);
-            return { state: "still_working" };
-        }
-        log(`Combined classify+decide for ${sessionId}: ${classification.state}${classification.suggestedResponse ? ` → "${classification.suggestedResponse}"` : ""}`);
-        if (classification.state === "task_complete") {
-            const session = manager?.get(sessionId);
-            const durationMs = session?.startedAt
-                ? Date.now() - new Date(session.startedAt).getTime()
-                : 0;
-            metricsTracker.recordCompletion(agentType, "classifier", durationMs);
-        }
-        return classification;
-    }
-    catch (err) {
-        log(`Combined classify+decide failed: ${err}`);
-        return null;
-    }
-}

package/packages/typescript/src/features/orchestrator/services/swarm-coordinator-prompts.d.ts

@@ -1,97 +0,0 @@
-/**
- * Prompt construction and response parsing for the Swarm Coordinator's
- * LLM-driven coordination decisions.
- *
- * Pure functions — no side effects, easy to test.
- * Pattern follows stall-classifier.ts:buildStallClassificationPrompt().
- *
- * @module services/swarm-coordinator-prompts
- */
-/** Per-session task context provided to the LLM for decision-making. */
-export interface TaskContextSummary {
-    sessionId: string;
-    agentType: string;
-    label: string;
-    originalTask: string;
-    workdir: string;
-    repo?: string;
-}
-/** A previous coordination decision, included for context continuity. */
-export interface DecisionHistoryEntry {
-    event: string;
-    promptText: string;
-    action: string;
-    response?: string;
-    reasoning: string;
-}
-/** Summary of a sibling task in the same swarm — for cross-task context. */
-export interface SiblingTaskSummary {
-    label: string;
-    agentType: string;
-    originalTask: string;
-    status: string;
-    /** Last significant decision or action taken by this sibling. */
-    lastKeyDecision?: string;
-    /** Summary of what the sibling accomplished (populated on completion). */
-    completionSummary?: string;
-}
-/** A significant creative or architectural decision made by an agent in the swarm. */
-export interface SharedDecision {
-    /** Which agent made this decision. */
-    agentLabel: string;
-    /** Brief description of the decision. */
-    summary: string;
-    /** When it was recorded. */
-    timestamp: number;
-}
-/** Parsed LLM response for a coordination decision. */
-export interface CoordinationLLMResponse {
-    action: "respond" | "escalate" | "ignore" | "complete";
-    /** Text to send (for action=respond with plain text input). */
-    response?: string;
-    /** Whether to use sendKeysToSession instead of sendToSession. */
-    useKeys?: boolean;
-    /** Key sequence to send (for TUI interactions). e.g. ["enter"] or ["down","enter"]. */
-    keys?: string[];
-    /** LLM's reasoning for the decision. */
-    reasoning: string;
-    /** Brief summary of a significant creative/architectural decision the agent made, if any. */
-    keyDecision?: string;
-}
-/**
- * Build the LLM prompt for making a coordination decision about a blocked agent.
- */
-export declare function buildCoordinationPrompt(taskCtx: TaskContextSummary, promptText: string, recentOutput: string, decisionHistory: DecisionHistoryEntry[], siblingTasks?: SiblingTaskSummary[], sharedDecisions?: SharedDecision[], swarmContext?: string): string;
-/**
- * Build the LLM prompt for checking on an idle session that hasn't
- * produced any events for a while.
- */
-export declare function buildIdleCheckPrompt(taskCtx: TaskContextSummary, recentOutput: string, idleMinutes: number, idleCheckNumber: number, maxIdleChecks: number, decisionHistory: DecisionHistoryEntry[], siblingTasks?: SiblingTaskSummary[], sharedDecisions?: SharedDecision[], swarmContext?: string): string;
-/**
- * Build the LLM prompt for assessing whether a completed turn means the
- * overall task is done, or if the agent needs more turns.
- *
- * Called when the adapter detects "task_complete" (agent finished a turn and
- * returned to the idle prompt). The LLM decides whether to stop the session
- * or send a follow-up instruction.
- */
-export declare function buildTurnCompletePrompt(taskCtx: TaskContextSummary, turnOutput: string, decisionHistory: DecisionHistoryEntry[], siblingTasks?: SiblingTaskSummary[], sharedDecisions?: SharedDecision[], swarmContext?: string): string;
-/**
- * Build a natural language event message describing a blocked agent, intended
- * to be processed by Milaidy's full ElizaOS pipeline (with conversation memory,
- * personality, and actions). Unlike buildCoordinationPrompt(), this omits the
- * "You are Eliza" preamble (she already IS Eliza in the pipeline) and asks
- * for a fenced JSON action block at the end of her response.
- */
-export declare function buildBlockedEventMessage(taskCtx: TaskContextSummary, promptText: string, recentOutput: string, decisionHistory: DecisionHistoryEntry[], siblingTasks?: SiblingTaskSummary[], sharedDecisions?: SharedDecision[], swarmContext?: string): string;
-/**
- * Build a natural language event message describing a turn completion, intended
- * to be processed by Milaidy's full ElizaOS pipeline.
- */
-export declare function buildTurnCompleteEventMessage(taskCtx: TaskContextSummary, turnOutput: string, decisionHistory: DecisionHistoryEntry[], siblingTasks?: SiblingTaskSummary[], sharedDecisions?: SharedDecision[], swarmContext?: string): string;
-/**
- * Parse the LLM's coordination response from raw text output.
- * Returns null if the response is invalid or unparseable.
- */
-export declare function parseCoordinationResponse(llmOutput: string): CoordinationLLMResponse | null;
-//# sourceMappingURL=swarm-coordinator-prompts.d.ts.map

package/packages/typescript/src/features/orchestrator/services/swarm-coordinator-prompts.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"swarm-coordinator-prompts.d.ts","sourceRoot":"","sources":["../../../../../../../../typescript/src/features/orchestrator/services/swarm-coordinator-prompts.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,wEAAwE;AACxE,MAAM,WAAW,kBAAkB;IAClC,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAED,yEAAyE;AACzE,MAAM,WAAW,oBAAoB;IACpC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,4EAA4E;AAC5E,MAAM,WAAW,kBAAkB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,MAAM,CAAC;IACf,iEAAiE;IACjE,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,0EAA0E;IAC1E,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC3B;AAED,sFAAsF;AACtF,MAAM,WAAW,cAAc;IAC9B,sCAAsC;IACtC,UAAU,EAAE,MAAM,CAAC;IACnB,yCAAyC;IACzC,OAAO,EAAE,MAAM,CAAC;IAChB,4BAA4B;IAC5B,SAAS,EAAE,MAAM,CAAC;CAClB;AA+CD,uDAAuD;AACvD,MAAM,WAAW,uBAAuB;IACvC,MAAM,EAAE,SAAS,GAAG,UAAU,GAAG,QAAQ,GAAG,UAAU,CAAC;IACvD,+DAA+D;IAC/D,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,iEAAiE;IACjE,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,uFAAuF;IACvF,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,wCAAwC;IACxC,SAAS,EAAE,MAAM,CAAC;IAClB,6FAA6F;IAC7F,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB;AAED;;GAEG;AACH,wBAAgB,uBAAuB,CACtC,OAAO,EAAE,kBAAkB,EAC3B,UAAU,EAAE,MAAM,EAClB,YAAY,EAAE,MAAM,EACpB,eAAe,EAAE,oBAAoB,EAAE,EACvC,YAAY,CAAC,EAAE,kBAAkB,EAAE,EACnC,eAAe,CAAC,EAAE,cAAc,EAAE,EAClC,YAAY,CAAC,EAAE,MAAM,GACnB,MAAM,CAiER;AAED;;;GAGG;AACH,wBAAgB,oBAAoB,CACnC,OAAO,EAAE,kBAAkB,EAC3B,YAAY,EAAE,MAAM,EACpB,WAAW,EAAE,MAAM,EACnB,eAAe,EAAE,MAAM,EACvB,aAAa,EAAE,MAAM,EACrB,eAAe,EAAE,oBAAoB,EAAE,EACvC,YAAY,CAAC,EAAE,kBAAkB,EAAE,EACnC,eAAe,CAAC,EAAE,cAAc,EAAE,EAClC,YAAY,CAAC,EAAE,MAAM,GACnB,MAAM,CAgER;AAED;;;;;;;GAOG;AACH,wBAAgB,uBAAuB,CACtC,OAAO,EAAE,kBAAkB,EAC3B,UAAU,EAAE,MAAM,EAClB,eAAe,EAAE,oBAAoB,EAAE,EACvC,YAAY,CAAC,EAAE,kBAAkB,EAAE,EACnC,eAAe,CAAC,EAAE,cAAc,EAAE,EAClC,YAAY,CAAC,EAAE,MAAM,GACnB,MAAM,CA6CR;AAID;;;;;;GAMG;AACH,wBAAgB,wBAAwB,CACvC,OAAO,EAAE,kBAAkB,EAC3B,UAAU,EAAE,MAAM,EAClB,YAAY,EAAE,MAAM,EACpB,eAAe,EAAE,oBAAoB,EAAE,EACvC,YAAY,CAAC,EAAE,kBAAkB,EAAE,EACnC,eAAe,CAAC,EAAE,cAAc,EAAE,EAClC,YAAY,CAAC,EAAE,MAAM,GACnB,MAAM,CAyCR;AAED;;;GAGG;AACH,wBAAgB,6BAA6B,CAC5C,OAAO,EAAE,kBAAkB,EAC3B,UAAU,EAAE,MAAM,EAClB,eAAe,EAAE,oBAAoB,EAAE,EACvC,YAAY,CAAC,EAAE,kBAAkB,EAAE,EACnC,eAAe,CAAC,EAAE,cAAc,EAAE,EAClC,YAAY,CAAC,EAAE,MAAM,GACnB,MAAM,CAwCR;AAED;;;GAGG;AACH,wBAAgB,yBAAyB,CACxC,SAAS,EAAE,MAAM,GACf,uBAAuB,GAAG,IAAI,CAmChC"}