@kirrosh/zond 0.13.0 → 0.16.0

package/src/cli/commands/compare.ts

@@ -1,129 +0,0 @@
-import { getDb } from "../../db/schema.ts";
-import { getRunById, getResultsByRunId } from "../../db/queries.ts";
-import { printError } from "../output.ts";
-
-const RESET = "\x1b[0m";
-const GREEN = "\x1b[32m";
-const RED = "\x1b[31m";
-const YELLOW = "\x1b[33m";
-const BOLD = "\x1b[1m";
-
-function useColor(): boolean {
-  return process.stdout.isTTY ?? false;
-}
-
-export interface CompareOptions {
-  runA: number;
-  runB: number;
-  dbPath?: string;
-}
-
-export async function compareCommand(options: CompareOptions): Promise<number> {
-  const { runA, runB, dbPath } = options;
-
-  try {
-    getDb(dbPath);
-
-    const runARecord = getRunById(runA);
-    const runBRecord = getRunById(runB);
-
-    if (!runARecord) {
-      printError(`Run #${runA} not found`);
-      return 2;
-    }
-    if (!runBRecord) {
-      printError(`Run #${runB} not found`);
-      return 2;
-    }
-
-    const resultsA = getResultsByRunId(runA);
-    const resultsB = getResultsByRunId(runB);
-
-    // Build lookup maps: "suite_name::test_name" → status
-    const mapA = new Map<string, string>();
-    const mapB = new Map<string, string>();
-
-    for (const r of resultsA) {
-      mapA.set(`${r.suite_name}::${r.test_name}`, r.status);
-    }
-    for (const r of resultsB) {
-      mapB.set(`${r.suite_name}::${r.test_name}`, r.status);
-    }
-
-    const regressions: Array<{ suite: string; test: string; before: string; after: string }> = [];
-    const fixes: Array<{ suite: string; test: string; before: string; after: string }> = [];
-    const unchanged: number[] = [];
-    let newTests = 0;
-    let removedTests = 0;
-
-    // Check all keys from B (current run)
-    for (const [key, statusB] of mapB) {
-      const statusA = mapA.get(key);
-      if (statusA === undefined) {
-        newTests++;
-        continue;
-      }
-      const wasPass = statusA === "pass";
-      const isPass = statusB === "pass";
-      const wasFail = statusA === "fail" || statusA === "error";
-      const isFail = statusB === "fail" || statusB === "error";
-
-      const [suite, test] = key.split("::") as [string, string];
-
-      if (wasPass && isFail) {
-        regressions.push({ suite, test, before: statusA, after: statusB });
-      } else if (wasFail && isPass) {
-        fixes.push({ suite, test, before: statusA, after: statusB });
-      } else {
-        unchanged.push(1);
-      }
-    }
-
-    // Count removed tests
-    for (const key of mapA.keys()) {
-      if (!mapB.has(key)) removedTests++;
-    }
-
-    const color = useColor();
-
-    // Header
-    console.log(`\nComparing run #${runA} (${runARecord.started_at.slice(0, 19)}) → run #${runB} (${runBRecord.started_at.slice(0, 19)})\n`);
-
-    // Summary line
-    const parts = [
-      `${color ? BOLD : ""}${regressions.length} regressions${color ? RESET : ""}`,
-      `${fixes.length} fixes`,
-      `${unchanged.length} unchanged`,
-    ];
-    if (newTests > 0) parts.push(`${newTests} new`);
-    if (removedTests > 0) parts.push(`${removedTests} removed`);
-    console.log(parts.join("  |  ") + "\n");
-
-    // Regressions
-    if (regressions.length > 0) {
-      console.log(`${color ? RED + BOLD : ""}Regressions (pass → fail):${color ? RESET : ""}`);
-      for (const r of regressions) {
-        console.log(`  ${color ? RED : ""}✗${color ? RESET : ""} [${r.suite}] ${r.test}  (${r.before} → ${r.after})`);
-      }
-      console.log("");
-    }
-
-    // Fixes
-    if (fixes.length > 0) {
-      console.log(`${color ? GREEN : ""}Fixes (fail → pass):${color ? RESET : ""}`);
-      for (const f of fixes) {
-        console.log(`  ${color ? GREEN : ""}✓${color ? RESET : ""} [${f.suite}] ${f.test}  (${f.before} → ${f.after})`);
-      }
-      console.log("");
-    }
-
-    if (regressions.length === 0 && fixes.length === 0) {
-      console.log(`${color ? GREEN : ""}No regressions detected.${color ? RESET : ""}`);
-    }
-
-    return regressions.length > 0 ? 1 : 0;
-  } catch (err) {
-    printError(err instanceof Error ? err.message : String(err));
-    return 2;
-  }
-}

package/src/cli/commands/doctor.ts

@@ -1,127 +0,0 @@
-import { existsSync } from "fs";
-import { resolve } from "path";
-import { getDb, closeDb } from "../../db/schema.ts";
-
-export interface DoctorOptions {
-  dbPath?: string;
-}
-
-interface Check {
-  label: string;
-  ok: boolean;
-  detail: string;
-}
-
-export async function doctorCommand(options: DoctorOptions): Promise<number> {
-  const checks: Check[] = [];
-
-  // 1. Database
-  checks.push(checkDatabase(options.dbPath));
-
-  // 2. Test files
-  checks.push(checkTestFiles());
-
-  // 3. OpenAPI spec
-  checks.push(checkOpenApiSpec());
-
-  // 4. Environment files
-  checks.push(checkEnvFiles());
-
-  // 5. Ollama
-  checks.push(await checkOllama());
-
-  // Print results
-  console.log("\nzond doctor\n");
-
-  let hasFailure = false;
-  for (const check of checks) {
-    const icon = check.ok ? "\u2713" : "\u2717";
-    console.log(`  ${icon} ${check.label}: ${check.detail}`);
-    if (!check.ok) hasFailure = true;
-  }
-
-  console.log("");
-  if (hasFailure) {
-    console.log("Some checks failed. See details above.");
-  } else {
-    console.log("All checks passed.");
-  }
-
-  return hasFailure ? 1 : 0;
-}
-
-function checkDatabase(dbPath?: string): Check {
-  const path = dbPath ? resolve(dbPath) : resolve(process.cwd(), "zond.db");
-  try {
-    const db = getDb(path);
-    const runs = (db.query("SELECT COUNT(*) as cnt FROM runs").get() as { cnt: number }).cnt;
-    const envs = (db.query("SELECT COUNT(*) as cnt FROM environments").get() as { cnt: number }).cnt;
-    closeDb();
-    return { label: "Database", ok: true, detail: `${path} (${runs} runs, ${envs} environments)` };
-  } catch (err) {
-    return { label: "Database", ok: false, detail: `Cannot open ${path}: ${(err as Error).message}` };
-  }
-}
-
-function checkTestFiles(): Check {
-  const dirs = [".", "tests", "test"];
-  const found: string[] = [];
-
-  for (const dir of dirs) {
-    const full = resolve(process.cwd(), dir);
-    if (!existsSync(full)) continue;
-    try {
-      const glob = new Bun.Glob("**/*.yaml");
-      for (const file of glob.scanSync({ cwd: full, absolute: false })) {
-        if (!file.startsWith(".env.")) {
-          found.push(`${dir}/${file}`);
-        }
-      }
-    } catch { /* ignore */ }
-  }
-
-  if (found.length > 0) {
-    return { label: "Test files", ok: true, detail: `${found.length} YAML file(s) found` };
-  }
-  return { label: "Test files", ok: false, detail: "No YAML test files found in cwd or tests/" };
-}
-
-function checkOpenApiSpec(): Check {
-  const candidates = ["openapi.yaml", "openapi.json", "openapi.yml", "swagger.yaml", "swagger.json"];
-  for (const name of candidates) {
-    const full = resolve(process.cwd(), name);
-    if (existsSync(full)) {
-      return { label: "OpenAPI spec", ok: true, detail: name };
-    }
-  }
-  return { label: "OpenAPI spec", ok: false, detail: "No openapi.yaml/json found (optional)" };
-}
-
-function checkEnvFiles(): Check {
-  const found: string[] = [];
-  try {
-    const glob = new Bun.Glob(".env.*.yaml");
-    for (const file of glob.scanSync({ cwd: process.cwd(), absolute: false })) {
-      found.push(file);
-    }
-  } catch { /* ignore */ }
-
-  if (found.length > 0) {
-    return { label: "Environment files", ok: true, detail: found.join(", ") };
-  }
-  return { label: "Environment files", ok: false, detail: "No .env.*.yaml files found (optional)" };
-}
-
-async function checkOllama(): Promise<Check> {
-  try {
-    const res = await fetch("http://localhost:11434/api/tags", { signal: AbortSignal.timeout(3000) });
-    if (res.ok) {
-      const data = await res.json() as { models?: { name: string }[] };
-      const count = data.models?.length ?? 0;
-      return { label: "Ollama", ok: true, detail: `Running (${count} model(s) available)` };
-    }
-    return { label: "Ollama", ok: false, detail: `Responded with status ${res.status}` };
-  } catch {
-    return { label: "Ollama", ok: false, detail: "Not reachable at localhost:11434 (optional, needed for chat)" };
-  }
-}

package/src/cli/commands/runs.ts

@@ -1,108 +0,0 @@
-import { getDb } from "../../db/schema.ts";
-import { listRuns, getRunById, getResultsByRunId } from "../../db/queries.ts";
-import { printError } from "../output.ts";
-
-export interface RunsOptions {
-  runId?: number;
-  limit?: number;
-  dbPath?: string;
-}
-
-const RESET = "\x1b[0m";
-const GREEN = "\x1b[32m";
-const RED = "\x1b[31m";
-const YELLOW = "\x1b[33m";
-
-function useColor(): boolean {
-  return process.stdout.isTTY ?? false;
-}
-
-function statusIcon(passed: number, failed: number): string {
-  const color = useColor();
-  if (failed === 0) return color ? `${GREEN}PASS${RESET}` : "PASS";
-  return color ? `${RED}FAIL${RESET}` : "FAIL";
-}
-
-export function runsCommand(options: RunsOptions): number {
-  const { runId, limit = 20, dbPath } = options;
-
-  try {
-    getDb(dbPath);
-  } catch (err) {
-    printError(`Failed to open database: ${(err as Error).message}`);
-    return 2;
-  }
-
-  if (runId !== undefined) {
-    return showRunDetail(runId);
-  }
-  return showRunList(limit);
-}
-
-function showRunList(limit: number): number {
-  const runs = listRuns(limit);
-
-  if (runs.length === 0) {
-    console.log("No runs found.");
-    return 0;
-  }
-
-  // Print table
-  const header = "ID     STATUS  TOTAL  PASS  FAIL  ENV         DURATION  STARTED";
-  console.log(header);
-  console.log("-".repeat(header.length));
-
-  for (const run of runs) {
-    const status = statusIcon(run.passed, run.failed);
-    const env = (run.environment ?? "-").slice(0, 10).padEnd(10);
-    const duration = run.duration_ms != null ? `${run.duration_ms}ms` : "-";
-    const started = run.started_at.slice(0, 19).replace("T", " ");
-    console.log(
-      `${String(run.id).padEnd(6)} ${status.padEnd(useColor() ? 14 : 6)}  ${String(run.total).padEnd(5)}  ${String(run.passed).padEnd(4)}  ${String(run.failed).padEnd(4)}  ${env}  ${duration.padEnd(8)}  ${started}`,
-    );
-  }
-
-  return 0;
-}
-
-function showRunDetail(runId: number): number {
-  const run = getRunById(runId);
-  if (!run) {
-    printError(`Run #${runId} not found`);
-    return 1;
-  }
-
-  const color = useColor();
-
-  console.log(`Run #${run.id}`);
-  console.log(`  Started:     ${run.started_at}`);
-  if (run.finished_at) console.log(`  Finished:    ${run.finished_at}`);
-  if (run.environment) console.log(`  Environment: ${run.environment}`);
-  if (run.duration_ms != null) console.log(`  Duration:    ${run.duration_ms}ms`);
-  console.log(`  Total: ${run.total}  Passed: ${run.passed}  Failed: ${run.failed}  Skipped: ${run.skipped}`);
-
-  const results = getResultsByRunId(runId);
-  if (results.length === 0) {
-    console.log("\nNo step results recorded.");
-    return 0;
-  }
-
-  console.log("\nSteps:");
-  for (const r of results) {
-    let statusStr: string;
-    if (r.status === "pass") {
-      statusStr = color ? `${GREEN}PASS${RESET}` : "PASS";
-    } else if (r.status === "fail" || r.status === "error") {
-      statusStr = color ? `${RED}${r.status.toUpperCase()}${RESET}` : r.status.toUpperCase();
-    } else {
-      statusStr = color ? `${YELLOW}SKIP${RESET}` : "SKIP";
-    }
-
-    console.log(`  ${statusStr} ${r.test_name} (${r.duration_ms}ms)`);
-    if (r.error_message) {
-      console.log(`       ${color ? RED : ""}${r.error_message}${color ? RESET : ""}`);
-    }
-  }
-
-  return 0;
-}

package/src/cli/commands/update.ts

@@ -1,142 +0,0 @@
-import { tmpdir } from "os";
-import { join } from "path";
-import { existsSync, unlinkSync, renameSync, copyFileSync, chmodSync } from "fs";
-import { VERSION } from "../index.ts";
-import { isCompiledBinary } from "../runtime.ts";
-
-export interface UpdateCommandOptions {
-  force?: boolean;
-}
-
-export function detectTarget(): { target: string; archive: "tar.gz" | "zip" } {
-  const platform = process.platform;
-  const arch = process.arch;
-
-  const os = platform === "win32" ? "win" : platform;
-  const archSuffix = arch === "arm64" ? "arm64" : "x64";
-  const target = `${os}-${archSuffix}`;
-  const archive = platform === "win32" ? "zip" : ("tar.gz" as const);
-
-  return { target, archive };
-}
-
-export function parseVersion(tag: string): string {
-  return tag.replace(/^v/, "");
-}
-
-export function compareVersions(a: string, b: string): number {
-  const pa = a.split(".").map(Number);
-  const pb = b.split(".").map(Number);
-  for (let i = 0; i < Math.max(pa.length, pb.length); i++) {
-    const na = pa[i] ?? 0;
-    const nb = pb[i] ?? 0;
-    if (na !== nb) return na - nb;
-  }
-  return 0;
-}
-
-export async function updateCommand(options: UpdateCommandOptions): Promise<number> {
-  if (!isCompiledBinary()) {
-    console.log("Running from source — use git pull to update.");
-    return 0;
-  }
-
-  console.log("Checking for updates...");
-
-  // Fetch latest release
-  const res = await fetch("https://api.github.com/repos/kirrosh/zond/releases/latest", {
-    headers: { "User-Agent": "zond-updater" },
-  });
-
-  if (!res.ok) {
-    console.error(`Failed to check for updates: HTTP ${res.status}`);
-    return 1;
-  }
-
-  const release = (await res.json()) as { tag_name: string };
-  const latestVersion = parseVersion(release.tag_name);
-  const currentVersion = VERSION;
-
-  if (!options.force && compareVersions(currentVersion, latestVersion) >= 0) {
-    console.log(`Already up to date (v${currentVersion}).`);
-    return 0;
-  }
-
-  console.log(`Updating v${currentVersion} → v${latestVersion}...`);
-
-  const { target, archive } = detectTarget();
-  const assetName = `zond-${target}.${archive}`;
-  const downloadUrl = `https://github.com/kirrosh/zond/releases/download/${release.tag_name}/${assetName}`;
-
-  // Download artifact
-  const dlRes = await fetch(downloadUrl);
-  if (!dlRes.ok) {
-    console.error(`Failed to download ${assetName}: HTTP ${dlRes.status}`);
-    return 1;
-  }
-
-  const tempDir = tmpdir();
-  const archivePath = join(tempDir, assetName);
-  const archiveBytes = new Uint8Array(await dlRes.arrayBuffer());
-  await Bun.write(archivePath, archiveBytes);
-
-  // Extract
-  const extractDir = join(tempDir, `zond-update-${Date.now()}`);
-  const mkdirResult = Bun.spawnSync(["mkdir", "-p", extractDir]);
-  if (mkdirResult.exitCode !== 0) {
-    // Fallback for Windows
-    const { mkdirSync } = await import("fs");
-    mkdirSync(extractDir, { recursive: true });
-  }
-
-  if (archive === "tar.gz") {
-    const tar = Bun.spawnSync(["tar", "-xzf", archivePath, "-C", extractDir]);
-    if (tar.exitCode !== 0) {
-      console.error("Failed to extract archive.");
-      return 1;
-    }
-  } else {
-    // Windows zip — use tar (Windows 10+ includes bsdtar)
-    const tar = Bun.spawnSync(["tar", "-xf", archivePath, "-C", extractDir]);
-    if (tar.exitCode !== 0) {
-      console.error("Failed to extract archive.");
-      return 1;
-    }
-  }
-
-  // Find extracted binary
-  const binaryName = process.platform === "win32" ? "zond.exe" : "zond";
-  const newBinary = join(extractDir, binaryName);
-
-  if (!existsSync(newBinary)) {
-    console.error(`Expected binary not found: ${newBinary}`);
-    return 1;
-  }
-
-  // Replace current binary
-  const currentBinary = process.execPath;
-
-  if (process.platform === "win32") {
-    // Windows: can't overwrite running exe — rename current to .old, copy new
-    const oldPath = currentBinary + ".old";
-    try {
-      if (existsSync(oldPath)) unlinkSync(oldPath);
-    } catch { /* ignore */ }
-    renameSync(currentBinary, oldPath);
-    copyFileSync(newBinary, currentBinary);
-  } else {
-    // Unix: rename new over current (atomic on same filesystem, but we copy across)
-    unlinkSync(currentBinary);
-    copyFileSync(newBinary, currentBinary);
-    chmodSync(currentBinary, 0o755);
-  }
-
-  // Cleanup
-  try {
-    unlinkSync(archivePath);
-    unlinkSync(newBinary);
-  } catch { /* best effort */ }
-
-  console.log(`Updated to v${latestVersion}.`);
-  return 0;
-}

package/src/core/agent/agent-loop.ts

@@ -1,116 +0,0 @@
-// Suppress AI SDK v2 spec compatibility warnings for Ollama (cosmetic, tool calling works fine)
-(globalThis as any).AI_SDK_LOG_WARNINGS = false;
-
-import { generateText, stepCountIs } from "ai";
-import { createOpenAI } from "@ai-sdk/openai";
-import { createAnthropic } from "@ai-sdk/anthropic";
-import { AGENT_SYSTEM_PROMPT } from "./system-prompt.ts";
-import { buildAgentTools } from "./tools/index.ts";
-import type { AgentConfig, AgentTurnResult, ToolEvent } from "./types.ts";
-import type { ModelMessage } from "ai";
-
-export function buildProvider(config: AgentConfig) {
-  const { provider } = config.provider;
-
-  if (provider === "anthropic") {
-    return createAnthropic({
-      apiKey: config.provider.apiKey,
-      baseURL: config.provider.baseUrl || undefined,
-    });
-  }
-
-  // openai, ollama, custom all use OpenAI-compatible API
-  return createOpenAI({
-    apiKey: config.provider.apiKey ?? "ollama",
-    baseURL: config.provider.baseUrl,
-  });
-}
-
-function buildModel(config: AgentConfig) {
-  const provider = buildProvider(config);
-  const { provider: providerType } = config.provider;
-
-  // For ollama/custom, use .chat() to avoid the responses API which they don't support.
-  if (providerType === "ollama" || providerType === "custom") {
-    return (provider as ReturnType<typeof createOpenAI>).chat(config.provider.model);
-  }
-
-  return provider(config.provider.model);
-}
-
-/**
- * Prepare messages with system prompt.
- * Some small/local models (e.g. qwen3 thinking mode via Ollama) break tool calling
- * when a separate `system` message is present. For ollama/custom providers, we inject
- * the system prompt into the first user message instead.
- */
-function prepareMessages(
-  messages: ModelMessage[],
-  config: AgentConfig,
-): { system?: string; messages: ModelMessage[] } {
-  const { provider } = config.provider;
-
-  if (provider === "ollama" || provider === "custom") {
-    // Inject system prompt into first user message to avoid breaking tool calling
-    const prepared = [...messages];
-    const firstUserIdx = prepared.findIndex(
-      (m) => m.role === "user" && typeof m.content === "string",
-    );
-
-    if (firstUserIdx >= 0) {
-      const msg = prepared[firstUserIdx] as { role: "user"; content: string };
-      prepared[firstUserIdx] = {
-        ...msg,
-        content: `[System instructions]\n${AGENT_SYSTEM_PROMPT}\n[End instructions]\n\n${msg.content}`,
-      };
-    }
-
-    return { messages: prepared };
-  }
-
-  // For OpenAI/Anthropic, use the standard system parameter
-  return { system: AGENT_SYSTEM_PROMPT, messages };
-}
-
-export async function runAgentTurn(
-  messages: ModelMessage[],
-  config: AgentConfig,
-  onToolEvent?: (event: ToolEvent) => void,
-): Promise<AgentTurnResult> {
-  const model = buildModel(config);
-  const tools = buildAgentTools(config);
-  const { system, messages: prepared } = prepareMessages(messages, config);
-  const toolEvents: ToolEvent[] = [];
-
-  const result = await generateText({
-    model,
-    system,
-    messages: prepared,
-    tools,
-    stopWhen: stepCountIs(config.maxSteps ?? 10),
-    maxOutputTokens: config.provider.maxTokens ?? 4096,
-    onStepFinish: ({ toolCalls, toolResults }) => {
-      if (toolCalls) {
-        for (let i = 0; i < toolCalls.length; i++) {
-          const call = toolCalls[i]!;
-          const toolResult = toolResults?.[i];
-          const event: ToolEvent = {
-            toolName: call.toolName,
-            args: ("input" in call ? call.input : {}) as Record<string, unknown>,
-            result: toolResult ?? null,
-            timestamp: new Date().toISOString(),
-          };
-          toolEvents.push(event);
-          onToolEvent?.(event);
-        }
-      }
-    },
-  });
-
-  return {
-    text: result.text,
-    toolEvents,
-    inputTokens: result.usage?.inputTokens ?? 0,
-    outputTokens: result.usage?.outputTokens ?? 0,
-  };
-}

package/src/core/agent/context-manager.ts

@@ -1,41 +0,0 @@
-import type { CoreMessageFormat } from "../../db/queries.ts";
-
-const MAX_MESSAGES = 20;
-const KEEP_RECENT_TURNS = 6; // 6 turns = 12 messages (user + assistant pairs)
-const KEEP_RECENT_MESSAGES = KEEP_RECENT_TURNS * 2;
-
-export function trimContext(messages: CoreMessageFormat[]): CoreMessageFormat[] {
-  if (messages.length <= MAX_MESSAGES) {
-    return messages;
-  }
-
-  const oldMessages = messages.slice(0, messages.length - KEEP_RECENT_MESSAGES);
-  const recentMessages = messages.slice(messages.length - KEEP_RECENT_MESSAGES);
-
-  const summary = buildSummary(oldMessages);
-
-  // Use role "user" for the summary so that the conversation always starts with a user message.
-  // Some providers require conversations to begin with a user turn.
-  return [
-    { role: "user" as const, content: summary },
-    ...recentMessages,
-  ];
-}
-
-function buildSummary(messages: CoreMessageFormat[]): string {
-  const userMessages = messages.filter((m) => m.role === "user");
-  const topics = userMessages
-    .map((m) => m.content.slice(0, 80))
-    .slice(0, 5);
-
-  const topicList = topics.length > 0
-    ? topics.map((t) => `- ${t}`).join("\n")
-    : "- General conversation";
-
-  return `[Conversation summary — ${messages.length} earlier messages condensed]
-
-Topics discussed:
-${topicList}
-
-The conversation continues below with the most recent messages.`;
-}

package/src/core/agent/system-prompt.ts

@@ -1,27 +0,0 @@
-export const AGENT_SYSTEM_PROMPT = `You are an API testing assistant powered by zond. You help users run, create, and diagnose API tests.
-
-You have access to the following tools:
-
-- **run_tests**: Execute API test suites from YAML files or directories. Returns pass/fail summary with run ID.
-- **query_results**: Query historical test run results and collections from the database.
-- **diagnose_failure**: Analyze a failed test run to identify root causes and suggest fixes.
-- **send_request**: Send an ad-hoc HTTP request for quick testing.
-
-Tool usage examples:
-- run_tests: { testPath: "tests/api.yaml" } or { testPath: "tests/", envName: "staging", safe: true }
-- query_results: action must be "list_runs", "get_run" (requires runId), or "list_collections"
-  - List runs: { action: "list_runs", limit: 10 }
-  - Get run details: { action: "get_run", runId: 1 }
-  - List collections: { action: "list_collections" }
-- diagnose_failure: { runId: 1 }
-
-Guidelines:
-- When asked to run tests, use the run_tests tool and report results clearly.
-- When a test run has failures, proactively use diagnose_failure to analyze the issues.
-- When asked about past results, use query_results to look up run history.
-- Always provide actionable suggestions when tests fail.
-- Be concise but thorough in your explanations.
-- If a tool call fails with a validation error, re-read the tool schema and retry with corrected arguments.
-- When in safe mode, only GET (read-only) tests will be executed.
-- When using thinking/reasoning, keep your internal reasoning focused and share conclusions with the user.
-`;