mixdog 0.9.2 → 0.9.4

package/src/runtime/agent/orchestrator/agent-trace-format.mjs

@@ -0,0 +1,362 @@
+import { createHash } from 'crypto';
+import { countJsonNextCalls } from './tools/next-call-utils.mjs';
+import {
+    appendAgentTrace,
+    normalizeSessionId,
+    warnAgentOnce,
+    _resolveToolFailurePath,
+    _appendToolFailureRow,
+} from './agent-trace-io.mjs';
+
+const MIXDOG_SLOW_TOOL_TRACE_MS = (() => {
+    const v = parseInt(process.env.MIXDOG_SLOW_TOOL_TRACE_MS, 10);
+    return Number.isFinite(v) && v > 0 ? v : 3000;
+})();
+const MIXDOG_SLOW_TOOL_TRACE_NAMES_RAW = String(process.env.MIXDOG_SLOW_TOOL_TRACE_NAMES || 'recall,grep,code_graph');
+const MIXDOG_SLOW_TOOL_TRACE_ALL = MIXDOG_SLOW_TOOL_TRACE_NAMES_RAW.trim() === '*';
+const MIXDOG_SLOW_TOOL_TRACE_NAMES = new Set(
+    MIXDOG_SLOW_TOOL_TRACE_NAMES_RAW
+        .split(',')
+        .map((name) => name.trim())
+        .filter(Boolean)
+);
+
+function traceAgentLoop({ sessionId, iteration, sendMs, messageCount, bodyBytesEst }) {
+    if (process.env.MIXDOG_AGENT_TRACE_VERBOSE !== '1') return;
+    appendAgentTrace({
+        sessionId,
+        iteration,
+        kind: 'loop',
+        send_ms: sendMs,
+        message_count: messageCount,
+        body_bytes_est: bodyBytesEst,
+    });
+}
+
+function traceAgentCompact({
+    sessionId,
+    iteration,
+    stage,
+    trigger,
+    compact_type,
+    prune_count,
+    compact_changed,
+    input_prefix_hash,
+    before_count,
+    after_count,
+    before_bytes,
+    after_bytes,
+    context_window,
+    budget_tokens,
+    boundary_tokens,
+    target_budget_tokens,
+    reserve_tokens,
+    pressure_tokens,
+    trigger_tokens,
+    message_tokens_est,
+    duration_ms,
+    provider,
+    model,
+    error,
+    error_code,
+    details,
+}) {
+    appendAgentTrace({
+        sessionId,
+        iteration,
+        kind: 'compact_meta',
+        stage: stage || null,
+        trigger: trigger || null,
+        compact_type: compact_type || null,
+        prune_count: prune_count ?? 0,
+        compact_changed: !!compact_changed,
+        input_prefix_hash: input_prefix_hash || null,
+        before_count: before_count ?? null,
+        after_count: after_count ?? null,
+        before_bytes: before_bytes ?? null,
+        after_bytes: after_bytes ?? null,
+        context_window: context_window ?? null,
+        budget_tokens: budget_tokens ?? null,
+        boundary_tokens: boundary_tokens ?? null,
+        target_budget_tokens: target_budget_tokens ?? null,
+        reserve_tokens: reserve_tokens ?? null,
+        pressure_tokens: pressure_tokens ?? null,
+        trigger_tokens: trigger_tokens ?? null,
+        message_tokens_est: message_tokens_est ?? null,
+        duration_ms: duration_ms ?? null,
+        provider: provider || null,
+        model: model || null,
+        error: error || null,
+        error_code: error_code || null,
+        details: details && typeof details === 'object' ? details : null,
+    });
+}
+
+const TOOL_ARG_KEYS = {
+    read: ['path', 'offset', 'limit', 'line', 'context', 'symbol'],
+    grep: ['pattern', 'path', 'glob', 'output_mode', 'head_limit', 'offset'],
+    glob: ['pattern', 'path', 'head_limit', 'offset', 'sort'],
+    find: ['query', 'path', 'head_limit'],
+    list: ['path', 'head_limit', 'offset'],
+    recall: ['query', 'limit', 'session_id', 'cwd'],
+    search: ['query', 'limit', 'cwd'],
+    explore: ['query', 'queries', 'limit', 'cwd'],
+    code_graph: ['mode', 'file', 'symbol', 'symbols', 'body', 'language', 'limit', 'depth', 'page', 'cwd'],
+    shell: ['command', 'cwd', 'timeout', 'mode', 'run_in_background', 'persistent', 'session_id'],
+    task: ['task_id', 'action', 'timeout_ms', 'poll_ms'],
+    edit: ['path', 'replace_all', 'edits'],
+    edit_many: ['edits'],
+    write: ['path'],
+    apply_patch: ['base_path', 'dry_run'],
+};
+
+const REDACT_KEY_RE = /token|secret|password|passwd|credential|authorization|api[_-]?key/i;
+const BODY_KEY_RE = /content|old_string|new_string|patch|rewrite/i;
+// Redact shell `command` values that look like they carry secrets. Covers
+// assignment forms, Authorization headers, --password / -p flags, and
+function _redactShellCommand(cmd) {
+    if (typeof cmd !== 'string') return cmd;
+    let out = cmd;
+    // Assignment RHS: PASSWORD=, SECRET=, TOKEN=, API_KEY=/APIKEY=.
+    out = out.replace(/((?:PASSWORD|SECRET|TOKEN|API_KEY|APIKEY)\s*=\s*)\S+/gi, '$1[redacted]');
+    // Authorization: Bearer <token>.
+    out = out.replace(/(Authorization:\s*Bearer\s+)\S+/gi, '$1[redacted]');
+    // Long flags: --password <v> / --password=<v> (also --token, --secret, --api-key).
+    out = out.replace(/(--(?:password|token|secret|api[-_]?key)(?:\s+|=))\S+/gi, '$1[redacted]');
+    // Short -p <v> flag (mysql/psql/curl style).
+    out = out.replace(/((?:^|\s)-p(?:\s+|=))\S+/g, '$1[redacted]');
+    // URL userinfo: scheme://user:secret@host -> scheme://user:[redacted]@host.
+    out = out.replace(/(:\/\/[^:\/\s@]+:)[^@\s]+(@)/g, '$1[redacted]$2');
+    // URL query params carrying tokens/keys.
+    out = out.replace(/([?&](?:token|api[-_]?key|access[-_]?token|auth|password|secret)=)[^&\s#]+/gi, '$1[redacted]');
+    return out;
+}
+
+function compactTraceArgValue(value, key = '', depth = 0) {
+    if (REDACT_KEY_RE.test(key)) return '[redacted]';
+    if (value === null || value === undefined) return value;
+    if (typeof value === 'string') {
+        // Redact shell commands that embed secrets before length-truncating.
+        if (key === 'command') {
+            value = _redactShellCommand(value);
+        }
+        const limit = BODY_KEY_RE.test(key) ? 60 : 180;
+        return value.length > limit ? `${value.slice(0, limit)}...` : value;
+    }
+    if (typeof value === 'number' || typeof value === 'boolean') return value;
+    if (Array.isArray(value)) {
+        if (depth >= 2) return `[${value.length} items]`;
+        return value.slice(0, 6).map((v) => compactTraceArgValue(v, key, depth + 1));
+    }
+    if (typeof value === 'object') {
+        if (depth >= 2) return '{...}';
+        const out = {};
+        for (const [k, v] of Object.entries(value).slice(0, 12)) {
+            out[k] = compactTraceArgValue(v, k, depth + 1);
+        }
+        return out;
+    }
+    return String(value);
+}
+
+function summarizeToolArgs(toolName, args) {
+    if (!args || typeof args !== 'object') return null;
+    const keys = TOOL_ARG_KEYS[String(toolName || '')] || Object.keys(args).slice(0, 8);
+    const out = {};
+    for (const key of keys) {
+        if (Object.prototype.hasOwnProperty.call(args, key)) out[key] = compactTraceArgValue(args[key], key);
+    }
+    for (const countKey of ['edits', 'writes']) {
+        if (Array.isArray(args[countKey])) out[`${countKey}_count`] = args[countKey].length;
+    }
+    if (toolName === 'read' && Array.isArray(args.path)) {
+        out.path_count = args.path.length;
+    }
+    return Object.keys(out).length ? out : null;
+}
+
+function stableTraceStringify(value) {
+    if (value === null || value === undefined) return JSON.stringify(value);
+    if (typeof value !== 'object') return JSON.stringify(value);
+    if (Array.isArray(value)) return '[' + value.map(stableTraceStringify).join(',') + ']';
+    const keys = Object.keys(value).sort();
+    return '{' + keys.map((k) => `${JSON.stringify(k)}:${stableTraceStringify(value[k])}`).join(',') + '}';
+}
+
+function hashTraceValue(value) {
+    try {
+        return createHash('sha256').update(stableTraceStringify(value)).digest('hex').slice(0, 16);
+    } catch {
+        return null;
+    }
+}
+
+function _firstNonEmptyLine(text) {
+    return String(text ?? '').split(/\r?\n/).find((line) => line.trim())?.trim() || '';
+}
+
+function _redactLogText(text) {
+    if (typeof text !== 'string') return '';
+    let out = text;
+    out = out.replace(/(Authorization:\s*Bearer\s+)\S+/gi, '$1[redacted]');
+    out = out.replace(/([?&](?:token|api[-_]?key|access[-_]?token|auth|password|secret)=)[^&\s#]+/gi, '$1[redacted]');
+    out = out.replace(/((?:PASSWORD|SECRET|TOKEN|API_KEY|APIKEY)\s*=\s*)\S+/gi, '$1[redacted]');
+    return out;
+}
+
+function classifyToolFailure(resultText, toolName) {
+    const text = String(resultText ?? '').toLowerCase();
+    if (/requires either|invalid arguments|unknown parameter|must be|schema|expected|required|old_string is .*>=/.test(text)) return 'schema/args';
+    if (/enoent|cannot find|not found at this path|path does not exist|no such file|file not found in graph|unreadable/.test(text)) return 'path/cwd';
+    if (/timed out|timeout|interrupted|aborted/.test(text)) return 'timeout/abort';
+    if (/hunk rejected|patch failed|context mismatch|expected first old\/context|context not found/.test(text)) return 'patch/context';
+    if (/not in allow-list|permission|denied|forbidden|not allowed/.test(text)) return 'permission';
+    if (/unknown tool|tool.*not.*available|missing.*tool/.test(text)) return 'tool-surface';
+    if (String(toolName || '') === 'shell' || /^\s*\[exit code:\s*\d+\]/i.test(String(resultText ?? ''))) return 'command-exit';
+    return 'runtime/failure';
+}
+
+function traceAgentToolFailure({ sessionId, iteration, toolName, toolKind, toolMs, toolArgs, agent, model, cwd, resultText, resultKind = 'error' }) {
+    if (process.env.MIXDOG_AGENT_TRACE_DISABLE === '1') return;
+    if (!_resolveToolFailurePath()) return;
+    try {
+        const cleanText = _redactLogText(String(resultText ?? ''));
+        const row = {
+            ts: Date.now(),
+            session_id: normalizeSessionId(sessionId),
+            iteration: iteration ?? null,
+            tool_name: toolName || null,
+            tool_kind: toolKind || null,
+            result_kind: resultKind || 'error',
+            category: classifyToolFailure(cleanText, toolName),
+            agent: agent || null,
+            model: model || null,
+            cwd: cwd || null,
+            tool_ms: Number.isFinite(Number(toolMs)) ? Number(toolMs) : null,
+            tool_args: summarizeToolArgs(toolName, toolArgs),
+            error_first_line: _firstNonEmptyLine(cleanText).slice(0, 300),
+            error_preview: cleanText.slice(0, 1200),
+            result_bytes_est: Buffer.byteLength(cleanText, 'utf8'),
+            result_lines_est: cleanText.length > 0 ? cleanText.split('\n').length : 0,
+        };
+        _appendToolFailureRow(row);
+    } catch (err) {
+        warnAgentOnce('tool-failure-log:append', `[tool-failure-log] append failed (${err?.message})`);
+    }
+}
+
+function traceAgentTool({ sessionId, iteration, toolName, toolKind, toolMs, toolArgs, agent, resultKind, model, resultText, cwd }) {
+    const nextCallCount = countJsonNextCalls(resultText);
+    const resultBytesEst = typeof resultText === 'string' ? Buffer.byteLength(resultText, 'utf8') : 0;
+    const resultLinesEst = typeof resultText === 'string' && resultText.length > 0 ? resultText.split('\n').length : 0;
+    const numericToolMs = Number(toolMs);
+    const summarizedArgs = summarizeToolArgs(toolName, toolArgs);
+    // Hash the FULL args, not the summary: summaries drop payload fields
+    // (e.g. apply_patch keeps only base_path), which made every patch in a
+    // session collide to one hash and broke duplicate/retry detection.
+    const toolArgsHash = toolArgs && typeof toolArgs === 'object'
+        ? hashTraceValue(toolArgs)
+        : (summarizedArgs ? hashTraceValue(summarizedArgs) : null);
+    // Keep a short redacted error preview on the tool row itself so trace
+    // analysis can see WHY a call failed without joining the failure log.
+    const errorFirstLine = resultKind === 'error'
+        ? _firstNonEmptyLine(_redactLogText(String(resultText ?? ''))).slice(0, 200) || null
+        : null;
+    // Flat shape — fields named exactly as the agent_calls PG columns so
+    // insertAgentCalls can pick them up by direct property access without
+    // a payload-unwrap step. result_kind has no column and rides as plain
+    // sibling metadata for downstream consumers.
+    appendAgentTrace({
+        sessionId,
+        iteration,
+        kind: 'tool',
+        agent: agent || null,
+        model: model || null,
+        tool_name: toolName,
+        tool_kind: toolKind,
+        tool_ms: toolMs,
+        tool_args: summarizedArgs,
+        tool_args_hash: toolArgsHash,
+        tool_args_summary: summarizedArgs,
+        result_kind: resultKind || null,
+        result_error_first_line: errorFirstLine,
+        result_has_next_call: nextCallCount > 0,
+        result_next_call_count: nextCallCount,
+        result_bytes_est: resultBytesEst,
+        result_lines_est: resultLinesEst,
+    });
+    if (
+        Number.isFinite(numericToolMs)
+        && numericToolMs >= MIXDOG_SLOW_TOOL_TRACE_MS
+        && (MIXDOG_SLOW_TOOL_TRACE_ALL || MIXDOG_SLOW_TOOL_TRACE_NAMES.size === 0 || MIXDOG_SLOW_TOOL_TRACE_NAMES.has(String(toolName || '')))
+    ) {
+        appendAgentTrace({
+            sessionId,
+            iteration,
+            kind: 'tool_slow',
+            agent: agent || null,
+            model: model || null,
+            tool_name: toolName,
+            tool_kind: toolKind,
+            tool_ms: numericToolMs,
+            payload: {
+                threshold_ms: MIXDOG_SLOW_TOOL_TRACE_MS,
+                result_kind: resultKind || null,
+                tool_args: summarizedArgs,
+                tool_args_hash: toolArgsHash,
+                tool_args_summary: summarizedArgs,
+                result_has_next_call: nextCallCount > 0,
+                result_next_call_count: nextCallCount,
+                result_bytes_est: resultBytesEst,
+                result_lines_est: resultLinesEst,
+                cwd: cwd || null,
+            },
+        });
+    }
+    if (resultKind === 'error') {
+        traceAgentToolFailure({ sessionId, iteration, toolName, toolKind, toolMs, toolArgs, agent, model, cwd, resultText, resultKind });
+    }
+}
+
+// Compression layer trace (result-compression.mjs). One row per tool call
+// where compression actually changed the byte count, so `gain` analytics
+// can sum savings_pct over a window (mirrors RTK's `rtk gain` model
+// without an external binary). No-op rows are dropped at the call site.
+export function traceAgentCompress({ sessionId, toolName, before, after }) {
+    // bytes_before/after/savings_pct moved into payload because the
+    // trace_events table only carries known top-level columns (id, ts,
+    // session_id, kind, tool_name, payload, ...) — fields outside that
+    // set are silently dropped at insert time. payload is jsonb so any
+    // shape survives. Aggregation: SELECT (payload->>'bytes_before')::int.
+    appendAgentTrace({
+        sessionId,
+        kind: 'compress',
+        tool_name: toolName,
+        payload: {
+            bytes_before: before,
+            bytes_after: after,
+            savings_pct: before > 0 ? Math.round((1 - after / before) * 100) : 0,
+        },
+    });
+}
+
+// Per-turn batch shape — one row per assistant turn with the number of
+// tool calls observed. Lets a consumer compute Lead-side multi-tool
+// adoption ratio (calls > 1 / total turns) directly from trace rows
+// instead of re-parsing every assistant message body.
+export function traceAgentBatch({ sessionId, toolCallCount }) {
+    appendAgentTrace({
+        sessionId,
+        kind: 'batch',
+        tool_call_count: toolCallCount,
+    });
+}
+
+export {
+    traceAgentLoop,
+    traceAgentCompact,
+    traceAgentTool,
+    traceAgentToolFailure,
+    summarizeToolArgs,
+    classifyToolFailure,
+};