@claude-flow/cli 3.32.3 → 3.32.4

package/dist/src/mcp-tools/agent-execute-core.js

@@ -0,0 +1,587 @@
+/**
+ * Shared agent-execution core.
+ *
+ * Both the agent_execute MCP tool and the workflow runtime (G3) need
+ * to dispatch a prompt to an agent's configured Anthropic model. This
+ * module factors that path out so it's testable and reusable, and
+ * keeps the wire from agent_spawn → ProviderManager (real) in one
+ * place rather than duplicated.
+ */
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';
+import { join } from 'node:path';
+import { getProjectCwd } from './types.js';
+const STORAGE_DIR = '.claude-flow';
+const AGENT_DIR = 'agents';
+const AGENT_FILE = 'store.json';
+function getAgentDir() { return join(getProjectCwd(), STORAGE_DIR, AGENT_DIR); }
+function getAgentPath() { return join(getAgentDir(), AGENT_FILE); }
+function ensureAgentDir() {
+    const dir = getAgentDir();
+    if (!existsSync(dir))
+        mkdirSync(dir, { recursive: true });
+}
+function loadAgentStore() {
+    try {
+        if (existsSync(getAgentPath()))
+            return JSON.parse(readFileSync(getAgentPath(), 'utf-8'));
+    }
+    catch { /* fall through */ }
+    return { agents: {}, version: '3.0.0' };
+}
+function saveAgentStore(store) {
+    ensureAgentDir();
+    writeFileSync(getAgentPath(), JSON.stringify(store, null, 2), 'utf-8');
+}
+// #1906/#2232 — Current model ids (Claude 4.x family):
+//   Opus 4.8    → claude-opus-4-8   (current, the `opus` alias)
+//   Opus 4.7    → claude-opus-4-7   (prior pin, reachable via `opus-4.7`)
+//   Sonnet 5    → claude-sonnet-5   (current, the `sonnet` alias)
+//   Sonnet 4.6  → claude-sonnet-4-6 (prior pin, reachable via `sonnet-4.6`)
+//   Haiku 4.5   → claude-haiku-4-5-20251001
+// `inherit` and the various defaults below all map to Sonnet 5.
+export const DEFAULT_ANTHROPIC_MODEL = 'claude-sonnet-5';
+const MODEL_MAP = {
+    haiku: 'claude-haiku-4-5-20251001',
+    sonnet: 'claude-sonnet-5',
+    'sonnet-4.6': 'claude-sonnet-4-6',
+    opus: 'claude-opus-4-8',
+    'opus-4.7': 'claude-opus-4-7',
+    inherit: DEFAULT_ANTHROPIC_MODEL,
+};
+// #2357 — the adaptive-thinking family (Fable 5, Opus 4.8, Opus 4.7, Sonnet 5)
+// removed the sampling parameters (temperature/top_p/top_k); the Anthropic API
+// returns 400 "Extra inputs are not permitted" when any is present.
+// Prefix-match so dated snapshots (e.g. claude-opus-4-8-YYYYMMDD) are
+// covered. Applies only to the direct Anthropic path — the Ollama/OpenRouter
+// OpenAI-compat paths accept temperature and are unchanged.
+export function modelRejectsSamplingParams(model) {
+    return /^claude-(fable-5|opus-4-8|opus-4-7|sonnet-5)/.test(model);
+}
+/**
+ * Generic Anthropic Messages API call. No agent registry coupling — used
+ * by agent_execute (with the agent's configured model) and by the WASM
+ * agent runtime (G4) when the bundled WASM only echoes input.
+ *
+ * #1725 — falls back to Ollama Cloud (Tier-2, OpenAI-compat) when
+ * ANTHROPIC_API_KEY is unset and OLLAMA_API_KEY is present, or when
+ * RUFLO_PROVIDER=ollama is explicitly set. Response shape is normalized
+ * to the Anthropic-flavored AnthropicCallResult so existing callers
+ * don't need to know which provider answered.
+ */
+export async function callAnthropicMessages(input) {
+    const explicitProvider = (process.env.RUFLO_PROVIDER || '').toLowerCase();
+    const ollamaKey = process.env.OLLAMA_API_KEY;
+    const anthropicKey = process.env.ANTHROPIC_API_KEY;
+    // #2042 — OpenRouter is an OpenAI-compat endpoint that fronts dozens of
+    // providers. Reporter (@ummcke00) had `providers.openrouter.apiKey` in
+    // their config.yaml but agent_execute hardcoded Anthropic. Detect via
+    // explicit RUFLO_PROVIDER=openrouter OR presence of OPENROUTER_API_KEY
+    // when no Anthropic key is available (same precedence as the Ollama
+    // branch above).
+    const openrouterKey = process.env.OPENROUTER_API_KEY;
+    const useOpenRouter = explicitProvider === 'openrouter' || (!anthropicKey && !!openrouterKey);
+    const useOllama = explicitProvider === 'ollama' || (!anthropicKey && !!ollamaKey && !openrouterKey);
+    if (useOpenRouter && openrouterKey) {
+        return callOpenAICompat({
+            ...input,
+            apiKey: openrouterKey,
+            baseUrl: process.env.OPENROUTER_BASE_URL || 'https://openrouter.ai/api',
+            providerLabel: 'openrouter',
+            // #2357 Finding C: anthropic/claude-3.5-sonnet was retired Oct 2025.
+            // Default to the same canonical family the rest of the resolver uses
+            // (MODEL_MAP). `OPENROUTER_DEFAULT_MODEL` still wins for callers who
+            // want to pin a specific OpenRouter slug.
+            defaultModel: process.env.OPENROUTER_DEFAULT_MODEL || 'anthropic/claude-sonnet-4-6',
+        });
+    }
+    if (useOllama && ollamaKey) {
+        return callOllamaCompat({ ...input, apiKey: ollamaKey });
+    }
+    if (!anthropicKey) {
+        return {
+            success: false,
+            error: 'No LLM provider configured. Set ANTHROPIC_API_KEY (Tier-3), OPENROUTER_API_KEY (#2042), or OLLAMA_API_KEY (Tier-2 — #1725).',
+        };
+    }
+    const model = input.model || DEFAULT_ANTHROPIC_MODEL;
+    const startedAt = Date.now();
+    try {
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), input.timeoutMs || 60000);
+        const res = await fetch('https://api.anthropic.com/v1/messages', {
+            method: 'POST',
+            headers: {
+                'x-api-key': anthropicKey,
+                'anthropic-version': '2023-06-01',
+                'content-type': 'application/json',
+            },
+            body: JSON.stringify({
+                model,
+                max_tokens: input.maxTokens || 1024,
+                // #2357 — omit temperature for models that reject sampling params
+                // (Fable 5 / Opus 4.8 / Opus 4.7 → 400 "Extra inputs are not
+                // permitted"); keep the 0.7 default unchanged for models that still
+                // accept it (sonnet / haiku / opus ≤4.6).
+                ...(modelRejectsSamplingParams(model)
+                    ? {}
+                    : { temperature: typeof input.temperature === 'number' ? input.temperature : 0.7 }),
+                // #8 prompt caching (hermes-agent pattern): mark the (often large,
+                // stable) system prompt as an ephemeral cache breakpoint so repeated
+                // agent_execute calls with the same system prompt hit Anthropic's
+                // prompt cache (~90% discount on cached input tokens, 5-min TTL).
+                ...(input.systemPrompt
+                    ? { system: [{ type: 'text', text: input.systemPrompt, cache_control: { type: 'ephemeral' } }] }
+                    : {}),
+                messages: [{ role: 'user', content: input.prompt }],
+            }),
+            signal: controller.signal,
+        });
+        clearTimeout(timer);
+        if (!res.ok) {
+            const errText = await res.text().catch(() => '<unreadable error body>');
+            return { success: false, model, error: `Anthropic API error ${res.status}: ${errText.slice(0, 400)}` };
+        }
+        const data = await res.json();
+        const textOut = data.content
+            .filter(c => c.type === 'text' && typeof c.text === 'string')
+            .map(c => c.text)
+            .join('');
+        return {
+            success: true,
+            model: data.model,
+            messageId: data.id,
+            stopReason: data.stop_reason,
+            output: textOut,
+            usage: {
+                inputTokens: data.usage.input_tokens,
+                outputTokens: data.usage.output_tokens,
+                totalTokens: data.usage.input_tokens + data.usage.output_tokens,
+            },
+            durationMs: Date.now() - startedAt,
+        };
+    }
+    catch (err) {
+        return {
+            success: false,
+            model,
+            error: err instanceof Error ? err.message : String(err),
+            durationMs: Date.now() - startedAt,
+        };
+    }
+}
+/**
+ * Ollama Cloud / OpenAI-compat provider — Tier-2 routing per ADR-026 + #1725.
+ *
+ * Endpoint: https://ollama.com/v1/chat/completions
+ * Auth: Authorization: Bearer <OLLAMA_API_KEY>
+ *
+ * Translates the Anthropic-flavored input shape onto OpenAI chat-completions
+ * and translates the response back so callers never see provider-specific
+ * fields. Logical model names are mapped to Ollama Cloud defaults:
+ *   - 'haiku'  / 'sonnet'  → 'gpt-oss:120b-cloud' (sensible single default)
+ *   - 'opus'              → 'gpt-oss:120b-cloud' (no opus tier on Ollama)
+ *   - explicit 'ollama:<model>' or bare provider-native name → passed through
+ */
+async function callOllamaCompat(input) {
+    const model = resolveOllamaModel(input.model);
+    const startedAt = Date.now();
+    // OLLAMA_BASE_URL lets users point at local/self-hosted endpoints
+    // (e.g. http://ruvultra:11434, http://localhost:11434) instead of
+    // Ollama Cloud. Default is the public cloud endpoint.
+    const base = (process.env.OLLAMA_BASE_URL || 'https://ollama.com').replace(/\/+$/, '');
+    const url = `${base}/v1/chat/completions`;
+    // Self-hosted endpoints typically don't need an Authorization header
+    // (the daemon binds to 11434 with no auth by default), but Ollama Cloud
+    // does. Send the bearer when the key is non-empty AND looks cloud-shaped.
+    const sendAuth = input.apiKey && input.apiKey !== 'local';
+    try {
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), input.timeoutMs || 60000);
+        const res = await fetch(url, {
+            method: 'POST',
+            headers: {
+                ...(sendAuth ? { Authorization: `Bearer ${input.apiKey}` } : {}),
+                'content-type': 'application/json',
+            },
+            body: JSON.stringify({
+                model,
+                max_tokens: input.maxTokens || 1024,
+                temperature: typeof input.temperature === 'number' ? input.temperature : 0.7,
+                messages: [
+                    ...(input.systemPrompt
+                        ? [{ role: 'system', content: input.systemPrompt }]
+                        : []),
+                    { role: 'user', content: input.prompt },
+                ],
+            }),
+            signal: controller.signal,
+        });
+        clearTimeout(timer);
+        if (!res.ok) {
+            const errText = await res.text().catch(() => '<unreadable error body>');
+            return { success: false, model, error: `Ollama API error ${res.status} at ${url}: ${errText.slice(0, 400)}` };
+        }
+        const data = (await res.json());
+        const textOut = data.choices?.[0]?.message?.content ?? '';
+        const usage = data.usage ?? {};
+        return {
+            success: true,
+            model: data.model ?? model,
+            messageId: data.id ?? `ollama-${Date.now()}`,
+            stopReason: data.choices?.[0]?.finish_reason ?? 'end_turn',
+            output: textOut,
+            usage: {
+                inputTokens: usage.prompt_tokens ?? 0,
+                outputTokens: usage.completion_tokens ?? 0,
+                totalTokens: usage.total_tokens ?? 0,
+            },
+            durationMs: Date.now() - startedAt,
+        };
+    }
+    catch (err) {
+        return {
+            success: false,
+            model,
+            error: err instanceof Error ? err.message : String(err),
+            durationMs: Date.now() - startedAt,
+        };
+    }
+}
+/**
+ * Generic OpenAI-compat caller for OpenRouter and other OpenAI-shaped
+ * endpoints. #2042 — reporter (@ummcke00) configured OpenRouter via
+ * config.yaml but agent_execute hardcoded the Anthropic fetch. This is
+ * the same shape as `callOllamaCompat` but routes to a configurable
+ * baseUrl + sends an OpenRouter-friendly default model when none is
+ * specified. Logical model names (haiku/sonnet/opus) pass through —
+ * OpenRouter accepts vendor-prefixed names like `anthropic/claude-3.5-sonnet`.
+ */
+async function callOpenAICompat(input) {
+    const model = resolveOpenAICompatModel(input.model, input.defaultModel);
+    const startedAt = Date.now();
+    const base = input.baseUrl.replace(/\/+$/, '');
+    const url = `${base}/v1/chat/completions`;
+    try {
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), input.timeoutMs || 60000);
+        const messages = [];
+        if (input.systemPrompt)
+            messages.push({ role: 'system', content: input.systemPrompt });
+        messages.push({ role: 'user', content: input.prompt });
+        const res = await fetch(url, {
+            method: 'POST',
+            headers: {
+                Authorization: `Bearer ${input.apiKey}`,
+                'content-type': 'application/json',
+                // OpenRouter convention: identify the integrating app for analytics
+                // and rate-limit tiering. Harmless on other OpenAI-compat backends.
+                'HTTP-Referer': 'https://github.com/ruvnet/ruflo',
+                'X-Title': 'Ruflo',
+            },
+            body: JSON.stringify({
+                model,
+                max_tokens: input.maxTokens || 1024,
+                temperature: typeof input.temperature === 'number' ? input.temperature : 0.7,
+                messages,
+            }),
+            signal: controller.signal,
+        });
+        clearTimeout(timer);
+        if (!res.ok) {
+            const errText = await res.text().catch(() => '<unreadable error body>');
+            return { success: false, model, error: `${input.providerLabel} API error ${res.status}: ${errText.slice(0, 400)}` };
+        }
+        const data = await res.json();
+        const textOut = data.choices?.[0]?.message?.content ?? '';
+        const usage = data.usage ?? {};
+        return {
+            success: true,
+            model: data.model || model,
+            messageId: data.id,
+            stopReason: data.choices?.[0]?.finish_reason ?? 'end_turn',
+            output: textOut,
+            usage: {
+                inputTokens: usage.prompt_tokens ?? 0,
+                outputTokens: usage.completion_tokens ?? 0,
+                totalTokens: usage.total_tokens ?? 0,
+            },
+            durationMs: Date.now() - startedAt,
+        };
+    }
+    catch (err) {
+        return {
+            success: false,
+            model,
+            error: err instanceof Error ? err.message : String(err),
+            durationMs: Date.now() - startedAt,
+        };
+    }
+}
+function resolveOpenAICompatModel(input, fallback) {
+    if (!input)
+        return fallback;
+    // Logical Claude names → OpenRouter Anthropic-vendored names.
+    // #2357 Finding C: the 3.5 / 3-opus slugs were retired Oct 2025; align with
+    // MODEL_MAP (claude-haiku-4-5 / claude-sonnet-4-6 / claude-opus-4-8).
+    if (input === 'haiku')
+        return 'anthropic/claude-haiku-4-5';
+    if (input === 'sonnet' || input === 'inherit')
+        return 'anthropic/claude-sonnet-4-6';
+    if (input === 'opus')
+        return 'anthropic/claude-opus-4-8';
+    return input;
+}
+function resolveOllamaModel(input) {
+    const DEFAULT = 'gpt-oss:120b-cloud';
+    if (!input)
+        return DEFAULT;
+    // Logical → cloud default
+    if (input === 'haiku' || input === 'sonnet' || input === 'opus' || input === 'inherit') {
+        return DEFAULT;
+    }
+    // Explicit provider prefix
+    if (input.startsWith('ollama:'))
+        return input.slice('ollama:'.length);
+    // Bare name with cloud suffix (e.g. 'llama3:70b-cloud') passes through
+    return input;
+}
+/**
+ * Resolve a model identifier to an Anthropic model ID. Accepts:
+ * - logical names: 'haiku', 'sonnet', 'opus', 'inherit'
+ * - prefixed: 'anthropic:claude-sonnet-4-6'
+ * - direct: 'claude-sonnet-4-6'
+ */
+export function resolveAnthropicModel(input) {
+    if (!input)
+        return DEFAULT_ANTHROPIC_MODEL;
+    if (input in MODEL_MAP)
+        return MODEL_MAP[input];
+    if (input.startsWith('anthropic:'))
+        return input.slice('anthropic:'.length);
+    return input;
+}
+export async function executeAgentTask(input) {
+    const store = loadAgentStore();
+    const agent = store.agents[input.agentId];
+    if (!agent)
+        return { success: false, agentId: input.agentId, error: 'Agent not found' };
+    if (agent.status === 'terminated')
+        return { success: false, agentId: input.agentId, error: 'Agent has been terminated' };
+    // ADR-149 iter 13 — first-call dispatch prefers `agent.modelId` (the
+    // cost-optimal pick from the neural backend) over `MODEL_MAP[agent.model]`
+    // when present. Before this fix the first attempt always used the tier
+    // mapping; only the iter-7 fallback chain used modelId on retry, which
+    // meant the cost-optimal pick was wasted unless the first call failed
+    // with a 429/5xx.
+    //
+    // Rules:
+    //   - agent.modelId is set AND it's a non-Anthropic slug (e.g.
+    //     'inclusionai/ling-2.6-flash', 'openai/gpt-4.1') → dispatch
+    //     directly via that id. callAnthropicMessages forwards non-Anthropic
+    //     ids through OpenRouter (#2042), so this gets us the cost-optimal
+    //     model on attempt 1.
+    //   - agent.modelId starts with 'anthropic/' → strip the prefix, use the
+    //     bare Anthropic id (e.g. 'anthropic/claude-haiku-4.5' → 'claude-haiku-4.5').
+    //   - agent.modelId is unset (no neural backend fired) → fall back to the
+    //     legacy tier-mapped MODEL_MAP path. Existing behaviour preserved.
+    let firstCallModel;
+    if (agent.modelId) {
+        firstCallModel = agent.modelId.startsWith('anthropic/')
+            ? agent.modelId.slice('anthropic/'.length)
+            : agent.modelId;
+    }
+    else {
+        firstCallModel = resolveAnthropicModel(agent.model || 'sonnet');
+    }
+    // Kept for legacy error-path remediation message + final-result `model` field
+    // (returned when the request fully fails with no successful retry).
+    const anthropicModel = firstCallModel;
+    const systemPrompt = input.systemPrompt ||
+        `You are a ${agent.agentType} agent operating as part of a Ruflo swarm. ` +
+            `Agent ID: ${input.agentId}. Domain: ${agent.domain ?? 'general'}. ` +
+            `Respond directly and stay focused on the task. If you need information you don't have, state that explicitly.`;
+    agent.status = 'busy';
+    agent.taskCount = (agent.taskCount || 0) + 1;
+    saveAgentStore(store);
+    const startedAt = Date.now();
+    // #2042 — delegate to callAnthropicMessages so the v3 provider router
+    // (Anthropic / Ollama / OpenRouter) governs which backend is hit.
+    let result = await callAnthropicMessages({
+        model: anthropicModel,
+        prompt: input.prompt,
+        systemPrompt,
+        maxTokens: input.maxTokens,
+        temperature: input.temperature,
+        timeoutMs: input.timeoutMs,
+    });
+    // ADR-149 iter 7 — fallback chain on retryable failures (429, 5xx,
+    // timeout). When the cost-optimal neural backend picked a specific
+    // model id and that model fails for a transient reason, fall back to
+    // the next-cheapest candidate that clears the quality bar. Budget is
+    // bounded by CLAUDE_FLOW_ROUTER_FALLBACK_MAX_RETRIES (default 1) so
+    // upstream outages don't cause retry storms.
+    const fallbackBudget = Math.max(0, parseInt(process.env.CLAUDE_FLOW_ROUTER_FALLBACK_MAX_RETRIES ?? '1', 10) || 1);
+    const fallbackHistory = [];
+    if (!result.success && agent.modelId && fallbackBudget > 0) {
+        const isRetryable = /\b(429|500|502|503|504|timeout|ECONNRESET|ETIMEDOUT)\b/i.test(result.error ?? '');
+        if (isRetryable) {
+            try {
+                const { nextCostOptimalAlternative } = await import('../ruvector/neural-router.js');
+                // ADR-149 iter 9 — delegate to the shared task-embedder LRU. The
+                // pipeline + cache are shared with agent-tools.ts, so the embedding
+                // for this prompt is almost always already cached from the initial
+                // routing decision (no extra inference cost in steady state).
+                const { embedTaskWithCache } = await import('../ruvector/task-embedder.js');
+                const embedding = await embedTaskWithCache(input.prompt);
+                if (embedding) {
+                    const excludeIds = [agent.modelId];
+                    for (let attempt = 0; attempt < fallbackBudget && !result.success; attempt++) {
+                        fallbackHistory.push({ modelId: excludeIds[excludeIds.length - 1], error: result.error ?? '' });
+                        const alt = await nextCostOptimalAlternative(embedding, excludeIds);
+                        if (!alt || !alt.modelId)
+                            break;
+                        excludeIds.push(alt.modelId);
+                        const altResult = await callAnthropicMessages({
+                            model: alt.modelId,
+                            prompt: input.prompt,
+                            systemPrompt,
+                            maxTokens: input.maxTokens,
+                            temperature: input.temperature,
+                            timeoutMs: input.timeoutMs,
+                        });
+                        // Record the model that ACTUALLY answered (or errored). On success,
+                        // update agent.modelId so downstream observers see the retry winner.
+                        agent.modelId = alt.modelId;
+                        result = altResult;
+                    }
+                }
+            }
+            catch {
+                // Fallback chain is best-effort — preserve the original error result.
+            }
+        }
+    }
+    agent.status = 'idle';
+    // ADR-149 — close the bandit feedback loop. `recordModelOutcome` updates
+    // the Beta(α,β) prior for the agent's tier so the Thompson sampler learns
+    // from production traffic instead of staying frozen at install-day priors.
+    // This is best-effort: any error here must NOT break the agent execution.
+    // For now, "success" = the model returned a response without an API error.
+    // A finer-grained signal (user-accepted output / regression-detected) is a
+    // follow-up; this commit closes the bandit's basic learning loop.
+    try {
+        const { recordModelOutcome, recordModelOutcomeByModelId } = await import('../ruvector/model-router.js');
+        // Bandit priors are keyed on the 3 canonical tiers (haiku/sonnet/opus/inherit);
+        // collapse opus-4.7 → opus before recording so the bandit's per-tier Beta
+        // updates correctly.
+        const tier = agent.model === 'opus-4.7' ? 'opus' :
+            agent.model ?? 'sonnet';
+        const outcome = result.success ? 'success' : 'failure';
+        recordModelOutcome(input.prompt, tier, outcome);
+        // ADR-149 — also write to the shadow per-modelId priors when the cost-
+        // optimal neural backend picked a concrete model id. Selection logic
+        // still uses tier priors, but the per-modelId data accumulates so a
+        // future refactor can switch the selector over.
+        if (agent.modelId) {
+            recordModelOutcomeByModelId(input.prompt, agent.modelId, outcome);
+        }
+        // ADR-149 iter 17 — close the production-data side of the feedback loop.
+        // The trajectory recorder (CLAUDE_FLOW_ROUTER_TRAJECTORY=1) writes
+        // per-decision rows to .swarm/model-router-trajectories.jsonl from the
+        // model-router itself. Now we ALSO write outcome rows here so future
+        // training (scripts/train-from-trajectories.mjs, follow-up) can pair
+        // decision+outcome by task_hash and produce DRACO-shaped retraining
+        // rows from real production traffic. quality = 1.0 on success, 0.0 on
+        // failure (coarse signal; finer-grained quality from user ratings or
+        // regression-detection is a separate hook).
+        if (process.env.CLAUDE_FLOW_ROUTER_TRAJECTORY === '1') {
+            try {
+                const { recordTrajectoryOutcome } = await import('../ruvector/router-trajectory.js');
+                const scores = agent.modelId
+                    ? { [agent.modelId]: outcome === 'success' ? 1.0 : 0.0 }
+                    : undefined;
+                // iter 31 — pass token usage + modelId so the recorder can compute
+                // USD spend at write time. Consumers (`router decisions`, cost-
+                // savings reports) get cost without their own price table.
+                recordTrajectoryOutcome({
+                    task: input.prompt,
+                    quality: outcome === 'success' ? 1.0 : 0.0,
+                    scores,
+                    source: 'agent-execute',
+                    tokens: result.usage ? { input: result.usage.inputTokens, output: result.usage.outputTokens } : undefined,
+                    modelId: agent.modelId,
+                });
+            }
+            catch { /* never break execution */ }
+        }
+        // ADR-150 weight-eft capture seam — the ONLY place a run's full outcome is
+        // known here: prompt (issue), assistant transcript, model, tier, and a
+        // resolved PROXY. Gated behind CLAUDE_FLOW_RUN_TRANSCRIPTS=1 (off by
+        // default; PII/retention surface, mirrors the router trajectory recorder).
+        // HONESTY: `resolved` here is the WEAKEST proxy — 'api-success' means only
+        // that the model returned without an API error, NOT that the output is
+        // correct (ruflo has no SWE-bench gold oracle). model_patch is '' because
+        // this single-shot execute path produces no unified diff. Both facts are
+        // stamped so no downstream weight-eft export mistakes this for gold data.
+        if (process.env.CLAUDE_FLOW_RUN_TRANSCRIPTS === '1') {
+            try {
+                const { recordRunTranscript, tierForModel } = await import('../ruvector/run-transcript-recorder.js');
+                const modelId = agent.modelId ?? String(agent.model ?? 'unknown');
+                const messages = [];
+                if (systemPrompt)
+                    messages.push({ role: 'system', content: systemPrompt });
+                messages.push({ role: 'user', content: input.prompt });
+                if (result.success && typeof result.output === 'string') {
+                    messages.push({ role: 'assistant', content: result.output });
+                }
+                recordRunTranscript({
+                    task: input.prompt,
+                    model: modelId,
+                    tier: tierForModel(modelId),
+                    resolved: outcome === 'success',
+                    resolvedSource: 'api-success',
+                    messages,
+                    source: 'agent-execute',
+                    tokens: result.usage ? { input: result.usage.inputTokens, output: result.usage.outputTokens } : undefined,
+                });
+            }
+            catch { /* never break execution */ }
+        }
+    }
+    catch {
+        // Silent — bandit feedback must never block routing.
+    }
+    if (result.success) {
+        const out = {
+            success: true,
+            agentId: input.agentId,
+            messageId: result.messageId,
+            model: result.model,
+            stopReason: result.stopReason,
+            output: result.output,
+            usage: result.usage,
+            durationMs: result.durationMs ?? Date.now() - startedAt,
+            ...(fallbackHistory.length > 0 ? { fallbackHistory } : {}),
+        };
+        agent.lastResult = out;
+        saveAgentStore(store);
+        return out;
+    }
+    saveAgentStore(store);
+    // No-provider-configured error → surface the same actionable message
+    // the router built, with a #2042-aware remediation pointer.
+    const noProvider = (result.error || '').includes('No LLM provider configured');
+    return {
+        success: false,
+        agentId: input.agentId,
+        model: anthropicModel,
+        error: result.error || 'agent_execute failed',
+        durationMs: result.durationMs ?? Date.now() - startedAt,
+        ...(fallbackHistory.length > 0 ? { fallbackHistory } : {}),
+        ...(noProvider && {
+            remediation: 'Set one of ANTHROPIC_API_KEY, OPENROUTER_API_KEY (+ optional OPENROUTER_BASE_URL), or OLLAMA_API_KEY. ' +
+                'Or set RUFLO_PROVIDER=openrouter|ollama to force a specific provider.',
+        }),
+    };
+}
+//# sourceMappingURL=agent-execute-core.js.map

package/dist/src/mcp-tools/agent-tools.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Agent MCP Tools for CLI
+ *
+ * Tool definitions for agent lifecycle management with file persistence.
+ * Includes model routing integration for intelligent model selection.
+ */
+import { type MCPTool } from './types.js';
+export declare const agentTools: MCPTool[];
+//# sourceMappingURL=agent-tools.d.ts.map