@vercel/agent-eval 1.1.0 → 1.2.0

package/dist/lib/agents/claude-code/agent.d.ts

@@ -0,0 +1,30 @@
+/**
+ * Claude Code agent — host-side definition + the thin Agent wrapper.
+ *
+ * The definition is pure data/auth; the actual CLI invocation + transcript capture
+ * live in ./run.mjs (shipped into the sandbox by the orchestrator). The wrapper
+ * keeps the public Agent interface identical so the registry / runner are untouched.
+ */
+import type { Agent } from '../types.js';
+import type { AgentDefinition } from '../plugin/contract.js';
+/**
+ * Build the Claude Code plugin definition.
+ *
+ * Auth has three mutually-exclusive modes, preserved exactly from the old adapter:
+ *   1. Vercel AI Gateway  → ANTHROPIC_BASE_URL + ANTHROPIC_AUTH_TOKEN (+ empty ANTHROPIC_API_KEY)
+ *   2. OAuth (CLAUDE_CODE_OAUTH_TOKEN present in the host env) → CLAUDE_CODE_OAUTH_TOKEN
+ *   3. Direct API (fallback) → ANTHROPIC_API_KEY
+ * getApiKeyEnvVar() and authEnv() must agree on the mode — both read process.env at
+ * call time, so the host's key resolution and the sandbox env stay consistent.
+ */
+export declare function createClaudeCodeDefinition({ useVercelAiGateway }: {
+    useVercelAiGateway: boolean;
+}): AgentDefinition;
+/**
+ * Create the Claude Code Agent. Thin wrapper over the generic orchestrator so the
+ * Agent interface (and thus registry.ts / index.ts / runner.ts) is unchanged.
+ */
+export declare function createClaudeCodeAgent({ useVercelAiGateway }: {
+    useVercelAiGateway: boolean;
+}): Agent;
+//# sourceMappingURL=agent.d.ts.map

package/dist/lib/agents/claude-code/agent.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent.d.ts","sourceRoot":"","sources":["../../../../src/lib/agents/claude-code/agent.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAIH,OAAO,KAAK,EAAE,KAAK,EAAmB,MAAM,aAAa,CAAC;AAG1D,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AAG7D;;;;;;;;;GASG;AACH,wBAAgB,0BAA0B,CAAC,EAAE,kBAAkB,EAAE,EAAE;IAAE,kBAAkB,EAAE,OAAO,CAAA;CAAE,GAAG,eAAe,CA4CnH;AAED;;;GAGG;AACH,wBAAgB,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,EAAE;IAAE,kBAAkB,EAAE,OAAO,CAAA;CAAE,GAAG,KAAK,CAWpG"}

package/dist/lib/agents/claude-code/agent.js

@@ -0,0 +1,80 @@
+/**
+ * Claude Code agent — host-side definition + the thin Agent wrapper.
+ *
+ * The definition is pure data/auth; the actual CLI invocation + transcript capture
+ * live in ./run.mjs (shipped into the sandbox by the orchestrator). The wrapper
+ * keeps the public Agent interface identical so the registry / runner are untouched.
+ */
+import { fileURLToPath } from 'node:url';
+import { AI_GATEWAY, ANTHROPIC_DIRECT } from '../shared.js';
+import { runWithDefinition } from '../plugin/orchestrator.js';
+/**
+ * Build the Claude Code plugin definition.
+ *
+ * Auth has three mutually-exclusive modes, preserved exactly from the old adapter:
+ *   1. Vercel AI Gateway  → ANTHROPIC_BASE_URL + ANTHROPIC_AUTH_TOKEN (+ empty ANTHROPIC_API_KEY)
+ *   2. OAuth (CLAUDE_CODE_OAUTH_TOKEN present in the host env) → CLAUDE_CODE_OAUTH_TOKEN
+ *   3. Direct API (fallback) → ANTHROPIC_API_KEY
+ * getApiKeyEnvVar() and authEnv() must agree on the mode — both read process.env at
+ * call time, so the host's key resolution and the sandbox env stay consistent.
+ */
+export function createClaudeCodeDefinition({ useVercelAiGateway }) {
+    return {
+        name: useVercelAiGateway ? 'vercel-ai-gateway/claude-code' : 'claude-code',
+        displayName: useVercelAiGateway ? 'Claude Code (Vercel AI Gateway)' : 'Claude Code',
+        defaultModel: 'opus',
+        o11yAgentName: 'claude-code',
+        // Resolve run.mjs next to this file (works in src during dev and in dist after
+        // the build copies run.mjs alongside the compiled agent.js).
+        runnerPath: fileURLToPath(new URL('./run.mjs', import.meta.url)),
+        getApiKeyEnvVar() {
+            if (useVercelAiGateway)
+                return AI_GATEWAY.apiKeyEnvVar;
+            if (process.env.CLAUDE_CODE_OAUTH_TOKEN)
+                return 'CLAUDE_CODE_OAUTH_TOKEN';
+            return ANTHROPIC_DIRECT.apiKeyEnvVar;
+        },
+        install(options) {
+            // Project deps (retried once), then the Claude Code CLI globally.
+            const cliPackage = options.agentOptions?.cliPackage || '@anthropic-ai/claude-code';
+            return [
+                { kind: 'command', cmd: 'npm', args: ['install'], retryOnce: true, errorPrefix: 'npm install failed', errorBody: 'last10' },
+                { kind: 'command', cmd: 'npm', args: ['install', '-g', cliPackage], errorPrefix: 'Claude Code install failed', errorBody: 'stderr' },
+            ];
+        },
+        // Claude Code is configured purely via env — no config files.
+        configFiles() {
+            return [];
+        },
+        authEnv(options) {
+            if (useVercelAiGateway) {
+                return {
+                    ANTHROPIC_BASE_URL: AI_GATEWAY.baseUrl,
+                    ANTHROPIC_AUTH_TOKEN: options.apiKey,
+                    ANTHROPIC_API_KEY: '',
+                };
+            }
+            if (process.env.CLAUDE_CODE_OAUTH_TOKEN) {
+                return { CLAUDE_CODE_OAUTH_TOKEN: options.apiKey };
+            }
+            return { ANTHROPIC_API_KEY: options.apiKey };
+        },
+    };
+}
+/**
+ * Create the Claude Code Agent. Thin wrapper over the generic orchestrator so the
+ * Agent interface (and thus registry.ts / index.ts / runner.ts) is unchanged.
+ */
+export function createClaudeCodeAgent({ useVercelAiGateway }) {
+    const definition = createClaudeCodeDefinition({ useVercelAiGateway });
+    return {
+        name: definition.name,
+        displayName: definition.displayName,
+        getApiKeyEnvVar: definition.getApiKeyEnvVar,
+        getDefaultModel() {
+            return definition.defaultModel;
+        },
+        run: (fixturePath, options) => runWithDefinition(definition, fixturePath, options),
+    };
+}
+//# sourceMappingURL=agent.js.map

package/dist/lib/agents/claude-code/agent.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent.js","sourceRoot":"","sources":["../../../../src/lib/agents/claude-code/agent.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAIzC,OAAO,EAAE,UAAU,EAAE,gBAAgB,EAAE,MAAM,cAAc,CAAC;AAE5D,OAAO,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAE9D;;;;;;;;;GASG;AACH,MAAM,UAAU,0BAA0B,CAAC,EAAE,kBAAkB,EAAmC;IAChG,OAAO;QACL,IAAI,EAAE,kBAAkB,CAAC,CAAC,CAAC,+BAA+B,CAAC,CAAC,CAAC,aAAa;QAC1E,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC,iCAAiC,CAAC,CAAC,CAAC,aAAa;QACnF,YAAY,EAAE,MAAM;QACpB,aAAa,EAAE,aAAa;QAC5B,+EAA+E;QAC/E,6DAA6D;QAC7D,UAAU,EAAE,aAAa,CAAC,IAAI,GAAG,CAAC,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAEhE,eAAe;YACb,IAAI,kBAAkB;gBAAE,OAAO,UAAU,CAAC,YAAY,CAAC;YACvD,IAAI,OAAO,CAAC,GAAG,CAAC,uBAAuB;gBAAE,OAAO,yBAAyB,CAAC;YAC1E,OAAO,gBAAgB,CAAC,YAAY,CAAC;QACvC,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,kEAAkE;YAClE,MAAM,UAAU,GAAI,OAAO,CAAC,YAAY,EAAE,UAAqB,IAAI,2BAA2B,CAAC;YAC/F,OAAO;gBACL,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,SAAS,CAAC,EAAE,SAAS,EAAE,IAAI,EAAE,WAAW,EAAE,oBAAoB,EAAE,SAAS,EAAE,QAAQ,EAAE;gBAC3H,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,SAAS,EAAE,IAAI,EAAE,UAAU,CAAC,EAAE,WAAW,EAAE,4BAA4B,EAAE,SAAS,EAAE,QAAQ,EAAE;aACrI,CAAC;QACJ,CAAC;QAED,8DAA8D;QAC9D,WAAW;YACT,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,IAAI,kBAAkB,EAAE,CAAC;gBACvB,OAAO;oBACL,kBAAkB,EAAE,UAAU,CAAC,OAAO;oBACtC,oBAAoB,EAAE,OAAO,CAAC,MAAM;oBACpC,iBAAiB,EAAE,EAAE;iBACtB,CAAC;YACJ,CAAC;YACD,IAAI,OAAO,CAAC,GAAG,CAAC,uBAAuB,EAAE,CAAC;gBACxC,OAAO,EAAE,uBAAuB,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;YACrD,CAAC;YACD,OAAO,EAAE,iBAAiB,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;QAC/C,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,qBAAqB,CAAC,EAAE,kBAAkB,EAAmC;IAC3F,MAAM,UAAU,GAAG,0BAA0B,CAAC,EAAE,kBAAkB,EAAE,CAAC,CAAC;IACtE,OAAO;QACL,IAAI,EAAE,UAAU,CAAC,IAAI;QACrB,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,eAAe,EAAE,UAAU,CAAC,eAAe;QAC3C,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QACD,GAAG,EAAE,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC,iBAAiB,CAAC,UAAU,EAAE,WAAW,EAAE,OAAO,CAAC;KACnF,CAAC;AACJ,CAAC"}

package/dist/lib/agents/claude-code/run.mjs

@@ -0,0 +1,217 @@
+/**
+ * Claude Code in-sandbox runner.
+ *
+ * This file is shipped INTO the sandbox by the orchestrator and executed there as
+ * `node __agent_eval__/run.mjs '<AgentRunInput JSON>'`. It is intentionally
+ * ZERO-DEPENDENCY (only `node:*` builtins) because the sandbox only has the
+ * fixture's own deps + the installed `claude` CLI — it cannot import anything from
+ * the @vercel/agent-eval package.
+ *
+ * Dual mode:
+ *   - runnable: invoked directly → reads argv, runs the agent, writes the result
+ *     file + prints a status line, exits 0.
+ *   - importable: `import { runAgent } from './run.mjs'` → returns a RunnerResult
+ *     (no file write, no exit). This is what a future in-sandbox judge reuses.
+ *
+ * `buildClaudeCodeCliArgs` and `extractObservedModelFromClaudeTranscript` are
+ * exported so the host unit tests can verify them directly (the test imports them
+ * from this very file, so there is no risk of the tested logic drifting from the
+ * logic the sandbox actually runs).
+ */
+
+import { spawnSync } from 'node:child_process';
+import { mkdirSync, readdirSync, readFileSync, statSync, writeFileSync } from 'node:fs';
+import { homedir } from 'node:os';
+import { dirname, join } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+/**
+ * Build the Claude Code CLI argument list.
+ *
+ * Order is regression-sensitive (preserved verbatim from the old adapter):
+ * `--allowedTools` is variadic — it keeps consuming positional tokens until the
+ * next flag. So (1) the tools must be a single comma-separated value, and (2)
+ * `--allowedTools` must be followed by another flag, never directly by the prompt.
+ * `--dangerously-skip-permissions` is always emitted and terminates the variadic
+ * capture before the trailing positional prompt (the #141 regression).
+ *
+ * @param {{prompt:string, model?:string, webResearch?:boolean, agentOptions?:Record<string,unknown>}} input
+ * @returns {string[]}
+ */
+export function buildClaudeCodeCliArgs(input) {
+  const cliArgs = ['--print'];
+  if (input.webResearch) {
+    cliArgs.push('--allowedTools', 'WebSearch,WebFetch');
+  }
+  if (input.model) {
+    cliArgs.push('--model', input.model);
+  }
+  cliArgs.push('--dangerously-skip-permissions');
+  const effort = input.agentOptions?.effort;
+  if (effort) {
+    cliArgs.push('--effort', String(effort));
+  }
+  cliArgs.push(input.prompt);
+  return cliArgs;
+}
+
+/**
+ * Scan a Claude Code transcript (JSONL) for the last `message.model` seen.
+ * Preserved verbatim from the old adapter. Non-JSON lines are ignored.
+ *
+ * @param {string|undefined|null} transcript
+ * @returns {string|undefined}
+ */
+export function extractObservedModelFromClaudeTranscript(transcript) {
+  if (!transcript) return undefined;
+
+  let observedModel;
+  for (const line of transcript.split('\n')) {
+    if (!line.trim()) continue;
+    try {
+      const event = JSON.parse(line);
+      if (typeof event.message?.model === 'string') {
+        observedModel = event.message.model;
+      }
+    } catch {
+      // Ignore non-JSON transcript lines.
+    }
+  }
+  return observedModel;
+}
+
+/**
+ * Locate + read the Claude Code transcript file.
+ *
+ * Claude Code writes JSONL transcripts to
+ * `~/.claude/projects/{cwd-with-slashes-as-dashes}/{session}.jsonl`. We pick the
+ * newest `.jsonl` by mtime — the equivalent of the old adapter's
+ * `ls -t ~/.claude/projects/<path>/*.jsonl | head -1`.
+ *
+ * Best-effort: any failure → null (never throws).
+ *
+ * @param {string} cwd post-relocation sandbox cwd (e.g. /workspace)
+ * @returns {string|null}
+ */
+function captureClaudeTranscript(cwd) {
+  try {
+    const projectPath = cwd.replace(/\//g, '-');
+    const dir = join(homedir(), '.claude', 'projects', projectPath);
+    const newest = readdirSync(dir)
+      .filter((f) => f.endsWith('.jsonl'))
+      .map((f) => {
+        const full = join(dir, f);
+        return { full, mtime: statSync(full).mtimeMs };
+      })
+      .sort((a, b) => b.mtime - a.mtime)[0];
+    if (!newest) return null;
+    return readFileSync(newest.full, 'utf8') || null;
+  } catch {
+    return null;
+  }
+}
+
+/**
+ * Run Claude Code over the workspace at `input.cwd` and return a RunnerResult.
+ *
+ * Auth (ANTHROPIC_BASE_URL/ANTHROPIC_AUTH_TOKEN/...) arrives via process.env — the
+ * orchestrator sets it on the `node run.mjs` invocation, and we pass process.env
+ * straight through to the CLI. The runner never handles secrets itself.
+ *
+ * @param {import('../plugin/contract.js').AgentRunInput} input
+ * @returns {Promise<import('../plugin/contract.js').RunnerResult>}
+ */
+export async function runAgent(input) {
+  const args = buildClaudeCodeCliArgs(input);
+
+  // spawnSync (not a shell string): the prompt is a plain argv element, so there is
+  // no shell quoting/escaping to get wrong. Blocking is fine — the runner has
+  // nothing else to do while the agent works. The sandbox-level timeout bounds it.
+  const res = spawnSync('claude', args, {
+    cwd: input.cwd,
+    env: process.env,
+    encoding: 'utf8',
+    maxBuffer: 64 * 1024 * 1024,
+  });
+
+  const stdout = res.stdout || '';
+  const stderr = res.stderr || '';
+  // Preserve the old concatenation order: stdout THEN stderr.
+  const output = stdout + stderr;
+  // spawnSync sets status=null + error when the binary can't be spawned at all.
+  const agentExitCode = res.status == null ? -1 : res.status;
+
+  // Capture transcript + observed model regardless of success (the old adapter did
+  // this even on the non-zero-exit path).
+  const transcript = captureClaudeTranscript(input.cwd);
+  const observedModel = extractObservedModelFromClaudeTranscript(transcript) ?? null;
+
+  if (res.error || agentExitCode !== 0) {
+    // Mirror the old error string: last 5 lines of output, else a coded fallback.
+    const errorLines = output.trim().split('\n').slice(-5).join('\n');
+    const fallback = res.error
+      ? `Failed to run claude: ${res.error.message}`
+      : `Claude Code exited with code ${agentExitCode}`;
+    return {
+      ok: false,
+      output,
+      transcript,
+      observedModel,
+      error: errorLines || fallback,
+      agentExitCode,
+    };
+  }
+
+  return { ok: true, output, transcript, observedModel, error: null, agentExitCode };
+}
+
+/* ─────────────────────────── runnable (CLI) entry ─────────────────────────── */
+
+// True when this file is executed directly (`node run.mjs ...`), false when imported.
+const isMain = process.argv[1] && fileURLToPath(import.meta.url) === process.argv[1];
+
+if (isMain) {
+  // argv[2] is the AgentRunInput JSON (never contains secrets).
+  const input = JSON.parse(process.argv[2]);
+
+  // Always produce a RunnerResult, even if the runner itself throws, so the host
+  // always has a result file to read (node exit code stays 0 except on a truly
+  // unrecoverable crash before we can write).
+  let result;
+  try {
+    result = await runAgent(input);
+  } catch (e) {
+    result = {
+      ok: false,
+      output: '',
+      transcript: null,
+      observedModel: null,
+      error: e && e.message ? e.message : String(e),
+      agentExitCode: -1,
+    };
+  }
+
+  // Source of truth: the result file the host reads back via sandbox.readFile.
+  try {
+    mkdirSync(dirname(input.resultPath), { recursive: true });
+    writeFileSync(input.resultPath, JSON.stringify(result));
+  } catch {
+    // If the file can't be written, the host falls back to the marker line below.
+  }
+
+  // Fallback channel: a compact status line (no transcript — it can be huge).
+  process.stdout.write(
+    '__AGENT_RESULT__ ' +
+      JSON.stringify({
+        ok: result.ok,
+        observedModel: result.observedModel,
+        error: result.error,
+        agentExitCode: result.agentExitCode,
+      }) +
+      '\n'
+  );
+
+  // Exit 0: "the runner ran". Agent success/failure is conveyed via result.ok, not
+  // the node exit code (the host distinguishes the two).
+  process.exit(0);
+}

package/dist/lib/agents/codex/agent.d.ts

@@ -0,0 +1,61 @@
+/**
+ * OpenAI Codex agent — host-side definition + the thin Agent wrapper.
+ *
+ * The definition is pure data/auth/config; the actual CLI invocation + transcript
+ * capture live in ./run.mjs (shipped into the sandbox by the orchestrator). The
+ * wrapper keeps the public Agent interface identical so the registry / runner are
+ * untouched.
+ *
+ * Codex is the one agent that needs BOTH a config file (a TOML profile written to
+ * the absolute path `~/.codex/default.config.toml`) AND a host-computed value that
+ * the runner must reproduce verbatim on the CLI (`-c model_reasoning_effort=...`,
+ * `-c model_verbosity=...`, `--model ...`). Those CLI values MUST match the values
+ * baked into the TOML, so they are derived ONCE on the host (from parseModelString)
+ * and handed to the runner via `runnerExtra()` → `input.extra`. parseModelString
+ * stays host-side (and exported) so the unit tests keep importing it.
+ */
+import type { Agent } from '../types.js';
+import type { AgentDefinition } from '../plugin/contract.js';
+/**
+ * Parse model string with optional query parameters.
+ * e.g. "gpt-5.2-codex?reasoningEffort=high" → { model: "gpt-5.2-codex", reasoningEffort: "high" }
+ *
+ * Stays host-side (and exported) — the unit tests import it, and the resulting
+ * values must be baked into BOTH the TOML config (configFiles) and the runner CLI
+ * args (runnerExtra), so it must be computed exactly once on the host.
+ */
+export declare function parseModelString(model: string): {
+    model: string;
+    reasoningEffort?: string;
+};
+/**
+ * Generate Codex profile config content.
+ *
+ * `reasoningEffort` and `model_verbosity` are written into the profile so a
+ * fresh `codex exec` doesn't pick up the CLI defaults of `"low"`, which are
+ * rejected by `gpt-5.2-codex`. Defaults to `"medium"` when `reasoningEffort`
+ * is omitted.
+ *
+ * Stays host-side (and exported) — the unit tests import it directly.
+ */
+export declare function generateCodexConfig(model: string | undefined, useVercelAiGateway: boolean, reasoningEffort?: string, webResearch?: boolean): string;
+/**
+ * Build the Codex plugin definition.
+ *
+ * Auth is two mutually-exclusive modes, preserved exactly from the old adapter:
+ *   1. Vercel AI Gateway → AI_GATEWAY_API_KEY (the gateway "vercel" provider's env_key)
+ *   2. Direct OpenAI API → OPENAI_API_KEY (the built-in "openai" provider's env_key)
+ * The runner additionally pipes the same key into `codex login --with-api-key` over
+ * stdin (reading it from process.env, never from the argv JSON).
+ */
+export declare function createCodexDefinition({ useVercelAiGateway }: {
+    useVercelAiGateway: boolean;
+}): AgentDefinition;
+/**
+ * Create the Codex Agent. Thin wrapper over the generic orchestrator so the Agent
+ * interface (and thus registry.ts / index.ts / runner.ts) is unchanged.
+ */
+export declare function createCodexAgent({ useVercelAiGateway }: {
+    useVercelAiGateway: boolean;
+}): Agent;
+//# sourceMappingURL=agent.d.ts.map

package/dist/lib/agents/codex/agent.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent.d.ts","sourceRoot":"","sources":["../../../../src/lib/agents/codex/agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAIH,OAAO,KAAK,EAAE,KAAK,EAAmB,MAAM,aAAa,CAAC;AAG1D,OAAO,KAAK,EAAE,eAAe,EAA2B,MAAM,uBAAuB,CAAC;AAGtF;;;;;;;GAOG;AACH,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,MAAM,GAAG;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,eAAe,CAAC,EAAE,MAAM,CAAA;CAAE,CAmB3F;AA+BD;;;;;;;;;GASG;AACH,wBAAgB,mBAAmB,CACjC,KAAK,EAAE,MAAM,GAAG,SAAS,EACzB,kBAAkB,EAAE,OAAO,EAC3B,eAAe,CAAC,EAAE,MAAM,EACxB,WAAW,CAAC,EAAE,OAAO,GACpB,MAAM,CA4BR;AAED;;;;;;;;GAQG;AACH,wBAAgB,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,EAAE;IAAE,kBAAkB,EAAE,OAAO,CAAA;CAAE,GAAG,eAAe,CAsF9G;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAAC,EAAE,kBAAkB,EAAE,EAAE;IAAE,kBAAkB,EAAE,OAAO,CAAA;CAAE,GAAG,KAAK,CAW/F"}

package/dist/lib/agents/codex/agent.js

@@ -0,0 +1,221 @@
+/**
+ * OpenAI Codex agent — host-side definition + the thin Agent wrapper.
+ *
+ * The definition is pure data/auth/config; the actual CLI invocation + transcript
+ * capture live in ./run.mjs (shipped into the sandbox by the orchestrator). The
+ * wrapper keeps the public Agent interface identical so the registry / runner are
+ * untouched.
+ *
+ * Codex is the one agent that needs BOTH a config file (a TOML profile written to
+ * the absolute path `~/.codex/default.config.toml`) AND a host-computed value that
+ * the runner must reproduce verbatim on the CLI (`-c model_reasoning_effort=...`,
+ * `-c model_verbosity=...`, `--model ...`). Those CLI values MUST match the values
+ * baked into the TOML, so they are derived ONCE on the host (from parseModelString)
+ * and handed to the runner via `runnerExtra()` → `input.extra`. parseModelString
+ * stays host-side (and exported) so the unit tests keep importing it.
+ */
+import { fileURLToPath } from 'node:url';
+import { AI_GATEWAY, OPENAI_DIRECT } from '../shared.js';
+import { runWithDefinition } from '../plugin/orchestrator.js';
+/**
+ * Parse model string with optional query parameters.
+ * e.g. "gpt-5.2-codex?reasoningEffort=high" → { model: "gpt-5.2-codex", reasoningEffort: "high" }
+ *
+ * Stays host-side (and exported) — the unit tests import it, and the resulting
+ * values must be baked into BOTH the TOML config (configFiles) and the runner CLI
+ * args (runnerExtra), so it must be computed exactly once on the host.
+ */
+export function parseModelString(model) {
+    const qIndex = model.indexOf('?');
+    if (qIndex === -1)
+        return { model };
+    const base = model.slice(0, qIndex);
+    const query = model.slice(qIndex + 1);
+    let reasoningEffort;
+    for (const pair of query.split('&')) {
+        const eqIndex = pair.indexOf('=');
+        if (eqIndex === -1)
+            continue;
+        const key = pair.slice(0, eqIndex);
+        const value = decodeURIComponent(pair.slice(eqIndex + 1));
+        if (key === 'reasoningEffort') {
+            reasoningEffort = value;
+        }
+    }
+    return { model: base, reasoningEffort };
+}
+/**
+ * Parse the run's model option into { model, reasoningEffort }, or all-undefined
+ * for native-default runs. Used by BOTH configFiles() (the TOML profile) and
+ * runnerExtra() (the CLI args), which MUST agree — so the parse happens once.
+ */
+function parseOptionsModel(options) {
+    return options.model ? parseModelString(options.model) : { model: undefined, reasoningEffort: undefined };
+}
+/**
+ * Default reasoning effort and verbosity baked into the generated Codex
+ * profile.
+ *
+ * The Codex CLI itself defaults both `model_reasoning_effort` and
+ * `model_verbosity` to `"low"`, but the default Codex model
+ * (`gpt-5.2-codex`) only accepts `"medium"` for both — so an out-of-the-box
+ * `codex exec` against the AI Gateway fails with:
+ *   "Unsupported value: 'low' is not supported with the 'gpt-5.2-codex'
+ *    model. Supported values are: 'medium'."
+ * (the error covers both the `reasoning.effort` and `text.verbosity` request
+ * parameters, depending on which the model rejected first).
+ *
+ * `"medium"` is also a valid value for the non-codex GPT-5.x models, so it's
+ * a safe default. Callers can override reasoning effort per-run via
+ * `model: "gpt-5.2-codex?reasoningEffort=high"`.
+ */
+const DEFAULT_REASONING_EFFORT = 'medium';
+const DEFAULT_MODEL_VERBOSITY = 'medium';
+/**
+ * Generate Codex profile config content.
+ *
+ * `reasoningEffort` and `model_verbosity` are written into the profile so a
+ * fresh `codex exec` doesn't pick up the CLI defaults of `"low"`, which are
+ * rejected by `gpt-5.2-codex`. Defaults to `"medium"` when `reasoningEffort`
+ * is omitted.
+ *
+ * Stays host-side (and exported) — the unit tests import it directly.
+ */
+export function generateCodexConfig(model, useVercelAiGateway, reasoningEffort, webResearch) {
+    // Opt-in web research. Note: through the AI Gateway (`wire_api =
+    // "responses"`), no `web_search` items have been observed — Codex instead
+    // researches via shell commands when it decides to. The setting is still
+    // written so direct-OpenAI runs and future gateway support pick it up; it
+    // is harmless when the server-side tool is unavailable.
+    const toolsSection = webResearch ? `\n[tools]\nweb_search = true\n` : '';
+    if (useVercelAiGateway) {
+        // AI Gateway uses prefixed model names like "openai/gpt-5.2-codex".
+        // Native-default runs intentionally omit model and reasoning overrides.
+        const fullModel = model ? (model.includes('/') ? model : `openai/${model}`) : undefined;
+        return `# Codex configuration for Vercel AI Gateway
+model_provider = "vercel"
+${fullModel ? `model = "${fullModel}"\n` : ''}${model ? `model_reasoning_effort = "${reasoningEffort ?? DEFAULT_REASONING_EFFORT}"\nmodel_verbosity = "${DEFAULT_MODEL_VERBOSITY}"\n` : ''}
+[model_providers.vercel]
+name = "Vercel AI Gateway"
+base_url = "${AI_GATEWAY.openAiBaseUrl}"
+env_key = "${AI_GATEWAY.apiKeyEnvVar}"
+wire_api = "responses"
+${toolsSection}`;
+    }
+    else {
+        // Direct OpenAI API — use the built-in "openai" provider (no custom provider needed).
+        // Native-default runs intentionally omit model and reasoning overrides.
+        const directModel = model ? (model.includes('/') ? model.split('/').pop() : model) : undefined;
+        return `# Direct OpenAI API configuration
+model_provider = "openai"
+${directModel ? `model = "${directModel}"\n` : ''}${model ? `model_reasoning_effort = "${reasoningEffort ?? DEFAULT_REASONING_EFFORT}"\nmodel_verbosity = "${DEFAULT_MODEL_VERBOSITY}"\n` : ''}${toolsSection}`;
+    }
+}
+/**
+ * Build the Codex plugin definition.
+ *
+ * Auth is two mutually-exclusive modes, preserved exactly from the old adapter:
+ *   1. Vercel AI Gateway → AI_GATEWAY_API_KEY (the gateway "vercel" provider's env_key)
+ *   2. Direct OpenAI API → OPENAI_API_KEY (the built-in "openai" provider's env_key)
+ * The runner additionally pipes the same key into `codex login --with-api-key` over
+ * stdin (reading it from process.env, never from the argv JSON).
+ */
+export function createCodexDefinition({ useVercelAiGateway }) {
+    return {
+        name: useVercelAiGateway ? 'vercel-ai-gateway/codex' : 'codex',
+        displayName: useVercelAiGateway ? 'OpenAI Codex (Vercel AI Gateway)' : 'OpenAI Codex',
+        defaultModel: 'openai/gpt-5.2-codex',
+        o11yAgentName: 'codex',
+        // Resolve run.mjs next to this file (works in src during dev and in dist after
+        // the build copies run.mjs alongside the compiled agent.js).
+        runnerPath: fileURLToPath(new URL('./run.mjs', import.meta.url)),
+        getApiKeyEnvVar() {
+            return useVercelAiGateway ? AI_GATEWAY.apiKeyEnvVar : OPENAI_DIRECT.apiKeyEnvVar;
+        },
+        install(_options) {
+            // Project deps (retried once), then the Codex CLI globally. Error wording is
+            // preserved verbatim: 'npm install failed (exit code N):\n<last 10>' and
+            // 'Codex CLI install failed: <stderr>'.
+            return [
+                { kind: 'command', cmd: 'npm', args: ['install'], retryOnce: true, errorPrefix: 'npm install failed', errorBody: 'last10' },
+                { kind: 'command', cmd: 'npm', args: ['install', '-g', '@openai/codex'], errorPrefix: 'Codex CLI install failed', errorBody: 'stderr' },
+            ];
+        },
+        configFiles(options) {
+            // Recent Codex CLI versions reject the old top-level `profile = "default"`
+            // key in config.toml and instead load `$CODEX_HOME/<profile>.config.toml`
+            // when `--profile <profile>` is set. For explicit models, reasoning_effort
+            // is baked into the profile (rather than passed via -c at runtime) so the
+            // value is visible in saved configs and so the CLI's own default of "low"
+            // can't sneak through. For native-default runs we omit these settings.
+            const parsed = parseOptionsModel(options);
+            const configContent = generateCodexConfig(parsed.model, useVercelAiGateway, parsed.reasoningEffort, options.webResearch);
+            // Absolute `~` path writeFiles can't target → heredoc via shell. Combines the
+            // old `mkdir -p ~/.codex` + `cat > ~/.codex/default.config.toml << 'EOF'`.
+            return [
+                {
+                    viaShell: `mkdir -p ~/.codex && cat > ~/.codex/default.config.toml << 'EOF'\n${configContent}\nEOF`,
+                },
+            ];
+        },
+        authEnv(options) {
+            // env_key the generated TOML expects: AI_GATEWAY_API_KEY for the "vercel"
+            // provider, OPENAI_API_KEY for the built-in "openai" provider. The runner
+            // also reads this same env to pipe the key into `codex login`.
+            const envVar = useVercelAiGateway ? AI_GATEWAY.apiKeyEnvVar : OPENAI_DIRECT.apiKeyEnvVar;
+            return { [envVar]: options.apiKey };
+        },
+        /**
+         * Host-computed values the runner must reproduce verbatim on the CLI. These
+         * derive from parseModelString and MUST match the TOML written in
+         * configFiles(), so they are computed here (host-side) and threaded into the
+         * runner via input.extra rather than re-deriving parseModelString in run.mjs.
+         *
+         *  - cliModel:                value for `--model` (gateway keeps the prefix,
+         *                             direct OpenAI strips it), or null to omit.
+         *  - reasoningEffort:         value for `-c model_reasoning_effort=...`, or null.
+         *  - verbosity:               value for `-c model_verbosity=...` (only when a
+         *                             model is set), or null to omit.
+         *
+         * (The login env var is chosen inside run.mjs from process.env — AI_GATEWAY_API_KEY
+         * else OPENAI_API_KEY — since authEnv() sets exactly one, so no flag is needed here.)
+         */
+        runnerExtra(options) {
+            const parsed = parseOptionsModel(options);
+            const baseModel = parsed.model;
+            // Direct OpenAI API needs unprefixed model names (e.g. "gpt-5.2-codex" not
+            // "openai/gpt-5.2-codex"); the gateway keeps the prefix as-is.
+            const cliModel = baseModel
+                ? (useVercelAiGateway ? baseModel : (baseModel.includes('/') ? baseModel.split('/').pop() : baseModel))
+                : undefined;
+            // Pass reasoning effort and verbosity via -c too; CLI flags have the highest
+            // precedence and we've observed Codex CLI silently falling back to its "low"
+            // defaults for both fields even when the profile sets them. Both default to
+            // "medium" for compatibility with gpt-5.2-codex. Omitted on native-default.
+            const reasoningEffort = baseModel ? (parsed.reasoningEffort ?? DEFAULT_REASONING_EFFORT) : undefined;
+            const verbosity = baseModel ? DEFAULT_MODEL_VERBOSITY : undefined;
+            return {
+                cliModel: cliModel ?? null,
+                reasoningEffort: reasoningEffort ?? null,
+                verbosity: verbosity ?? null,
+            };
+        },
+    };
+}
+/**
+ * Create the Codex Agent. Thin wrapper over the generic orchestrator so the Agent
+ * interface (and thus registry.ts / index.ts / runner.ts) is unchanged.
+ */
+export function createCodexAgent({ useVercelAiGateway }) {
+    const definition = createCodexDefinition({ useVercelAiGateway });
+    return {
+        name: definition.name,
+        displayName: definition.displayName,
+        getApiKeyEnvVar: definition.getApiKeyEnvVar,
+        getDefaultModel() {
+            return definition.defaultModel;
+        },
+        run: (fixturePath, options) => runWithDefinition(definition, fixturePath, options),
+    };
+}
+//# sourceMappingURL=agent.js.map

package/dist/lib/agents/codex/agent.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent.js","sourceRoot":"","sources":["../../../../src/lib/agents/codex/agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAIzC,OAAO,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAEzD,OAAO,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAE9D;;;;;;;GAOG;AACH,MAAM,UAAU,gBAAgB,CAAC,KAAa;IAC5C,MAAM,MAAM,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IAClC,IAAI,MAAM,KAAK,CAAC,CAAC;QAAE,OAAO,EAAE,KAAK,EAAE,CAAC;IAEpC,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC;IACpC,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACtC,IAAI,eAAmC,CAAC;IAExC,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;QACpC,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;QAClC,IAAI,OAAO,KAAK,CAAC,CAAC;YAAE,SAAS;QAC7B,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;QACnC,MAAM,KAAK,GAAG,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC;QAC1D,IAAI,GAAG,KAAK,iBAAiB,EAAE,CAAC;YAC9B,eAAe,GAAG,KAAK,CAAC;QAC1B,CAAC;IACH,CAAC;IAED,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,eAAe,EAAE,CAAC;AAC1C,CAAC;AAED;;;;GAIG;AACH,SAAS,iBAAiB,CAAC,OAAwB;IACjD,OAAO,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,gBAAgB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,eAAe,EAAE,SAAS,EAAE,CAAC;AAC5G,CAAC;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,wBAAwB,GAAG,QAAQ,CAAC;AAC1C,MAAM,uBAAuB,GAAG,QAAQ,CAAC;AAEzC;;;;;;;;;GASG;AACH,MAAM,UAAU,mBAAmB,CACjC,KAAyB,EACzB,kBAA2B,EAC3B,eAAwB,EACxB,WAAqB;IAErB,iEAAiE;IACjE,0EAA0E;IAC1E,yEAAyE;IACzE,0EAA0E;IAC1E,wDAAwD;IACxD,MAAM,YAAY,GAAG,WAAW,CAAC,CAAC,CAAC,gCAAgC,CAAC,CAAC,CAAC,EAAE,CAAC;IACzE,IAAI,kBAAkB,EAAE,CAAC;QACvB,oEAAoE;QACpE,wEAAwE;QACxE,MAAM,SAAS,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,UAAU,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QACxF,OAAO;;EAET,SAAS,CAAC,CAAC,CAAC,YAAY,SAAS,KAAK,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,CAAC,CAAC,CAAC,6BAA6B,eAAe,IAAI,wBAAwB,yBAAyB,uBAAuB,KAAK,CAAC,CAAC,CAAC,EAAE;;;cAG5K,UAAU,CAAC,aAAa;aACzB,UAAU,CAAC,YAAY;;EAElC,YAAY,EAAE,CAAC;IACf,CAAC;SAAM,CAAC;QACN,sFAAsF;QACtF,wEAAwE;QACxE,MAAM,WAAW,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,EAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QAChG,OAAO;;EAET,WAAW,CAAC,CAAC,CAAC,YAAY,WAAW,KAAK,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,CAAC,CAAC,CAAC,6BAA6B,eAAe,IAAI,wBAAwB,yBAAyB,uBAAuB,KAAK,CAAC,CAAC,CAAC,EAAE,GAAG,YAAY,EAAE,CAAC;IAC9M,CAAC;AACH,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,qBAAqB,CAAC,EAAE,kBAAkB,EAAmC;IAC3F,OAAO;QACL,IAAI,EAAE,kBAAkB,CAAC,CAAC,CAAC,yBAAyB,CAAC,CAAC,CAAC,OAAO;QAC9D,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC,kCAAkC,CAAC,CAAC,CAAC,cAAc;QACrF,YAAY,EAAE,sBAAsB;QACpC,aAAa,EAAE,OAAO;QACtB,+EAA+E;QAC/E,6DAA6D;QAC7D,UAAU,EAAE,aAAa,CAAC,IAAI,GAAG,CAAC,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAEhE,eAAe;YACb,OAAO,kBAAkB,CAAC,CAAC,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,aAAa,CAAC,YAAY,CAAC;QACnF,CAAC;QAED,OAAO,CAAC,QAAyB;YAC/B,6EAA6E;YAC7E,yEAAyE;YACzE,wCAAwC;YACxC,OAAO;gBACL,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,SAAS,CAAC,EAAE,SAAS,EAAE,IAAI,EAAE,WAAW,EAAE,oBAAoB,EAAE,SAAS,EAAE,QAAQ,EAAE;gBAC3H,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,SAAS,EAAE,IAAI,EAAE,eAAe,CAAC,EAAE,WAAW,EAAE,0BAA0B,EAAE,SAAS,EAAE,QAAQ,EAAE;aACxI,CAAC;QACJ,CAAC;QAED,WAAW,CAAC,OAAwB;YAClC,2EAA2E;YAC3E,0EAA0E;YAC1E,2EAA2E;YAC3E,0EAA0E;YAC1E,0EAA0E;YAC1E,uEAAuE;YACvE,MAAM,MAAM,GAAG,iBAAiB,CAAC,OAAO,CAAC,CAAC;YAC1C,MAAM,aAAa,GAAG,mBAAmB,CAAC,MAAM,CAAC,KAAK,EAAE,kBAAkB,EAAE,MAAM,CAAC,eAAe,EAAE,OAAO,CAAC,WAAW,CAAC,CAAC;YACzH,8EAA8E;YAC9E,2EAA2E;YAC3E,OAAO;gBACL;oBACE,QAAQ,EAAE,qEAAqE,aAAa,OAAO;iBACpG;aACF,CAAC;QACJ,CAAC;QAED,OAAO,CAAC,OAAwB;YAC9B,0EAA0E;YAC1E,0EAA0E;YAC1E,+DAA+D;YAC/D,MAAM,MAAM,GAAG,kBAAkB,CAAC,CAAC,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,aAAa,CAAC,YAAY,CAAC;YACzF,OAAO,EAAE,CAAC,MAAM,CAAC,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;QACtC,CAAC;QAED;;;;;;;;;;;;;;WAcG;QACH,WAAW,CAAC,OAAwB;YAClC,MAAM,MAAM,GAAG,iBAAiB,CAAC,OAAO,CAAC,CAAC;YAC1C,MAAM,SAAS,GAAG,MAAM,CAAC,KAAK,CAAC;YAC/B,2EAA2E;YAC3E,+DAA+D;YAC/D,MAAM,QAAQ,GAAG,SAAS;gBACxB,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,EAAG,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;gBACxG,CAAC,CAAC,SAAS,CAAC;YACd,6EAA6E;YAC7E,6EAA6E;YAC7E,4EAA4E;YAC5E,4EAA4E;YAC5E,MAAM,eAAe,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,eAAe,IAAI,wBAAwB,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;YACrG,MAAM,SAAS,GAAG,SAAS,CAAC,CAAC,CAAC,uBAAuB,CAAC,CAAC,CAAC,SAAS,CAAC;YAClE,OAAO;gBACL,QAAQ,EAAE,QAAQ,IAAI,IAAI;gBAC1B,eAAe,EAAE,eAAe,IAAI,IAAI;gBACxC,SAAS,EAAE,SAAS,IAAI,IAAI;aAC7B,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAE,kBAAkB,EAAmC;IACtF,MAAM,UAAU,GAAG,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,CAAC,CAAC;IACjE,OAAO;QACL,IAAI,EAAE,UAAU,CAAC,IAAI;QACrB,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,eAAe,EAAE,UAAU,CAAC,eAAe;QAC3C,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QACD,GAAG,EAAE,CAAC,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC,iBAAiB,CAAC,UAAU,EAAE,WAAW,EAAE,OAAO,CAAC;KACnF,CAAC;AACJ,CAAC"}