@roackb2/heddle 0.0.37 → 0.0.38

package/dist/src/core/eval/trace-analyzer.js

@@ -0,0 +1,106 @@
+import { existsSync, readFileSync } from 'node:fs';
+const MUTATION_TOOLS = new Set(['edit_file', 'delete_file', 'move_file', 'run_shell_mutate']);
+const READ_OR_SEARCH_TOOLS = new Set(['read_file', 'list_files', 'search_files', 'run_shell_inspect']);
+export function analyzeTraceFiles(paths) {
+    return analyzeTrace(paths.flatMap(readTraceFile));
+}
+export function analyzeTrace(trace) {
+    const toolsByName = {};
+    const readOrSearchBeforeMutation = [];
+    let assistantTurns = 0;
+    let toolCalls = 0;
+    let toolResults = 0;
+    let mutations = 0;
+    let approvalsRequested = 0;
+    let approvalsResolved = 0;
+    let toolErrors = 0;
+    let verificationCommandsAfterMutation = 0;
+    const verificationCommandDetails = [];
+    let firstMutationStep;
+    let outcome;
+    let summary;
+    for (const event of trace) {
+        if (event.type === 'assistant.turn') {
+            assistantTurns++;
+            for (const call of event.toolCalls ?? []) {
+                toolCalls++;
+                toolsByName[call.tool] = (toolsByName[call.tool] ?? 0) + 1;
+                if (READ_OR_SEARCH_TOOLS.has(call.tool) && firstMutationStep === undefined) {
+                    readOrSearchBeforeMutation.push(summarizeToolInput(call.tool, call.input));
+                }
+                if (MUTATION_TOOLS.has(call.tool)) {
+                    mutations++;
+                    firstMutationStep ??= event.step;
+                }
+                if (firstMutationStep !== undefined
+                    && (call.tool === 'run_shell_mutate' || call.tool === 'run_shell_inspect')
+                    && isVerificationCommand(call.input)) {
+                    verificationCommandsAfterMutation++;
+                    verificationCommandDetails.push(summarizeToolInput(call.tool, call.input));
+                }
+            }
+            continue;
+        }
+        if (event.type === 'tool.result') {
+            toolResults++;
+            if (!event.result.ok) {
+                toolErrors++;
+            }
+        }
+        if (event.type === 'tool.approval_requested') {
+            approvalsRequested++;
+        }
+        if (event.type === 'tool.approval_resolved') {
+            approvalsResolved++;
+        }
+        if (event.type === 'run.finished') {
+            outcome = event.outcome;
+            summary = event.summary;
+        }
+    }
+    return {
+        assistantTurns,
+        toolCalls,
+        toolResults,
+        mutations,
+        approvalsRequested,
+        approvalsResolved,
+        toolErrors,
+        verificationCommandsAfterMutation,
+        verificationCommandDetails: [...new Set(verificationCommandDetails)],
+        firstMutationStep,
+        outcome,
+        summary,
+        toolsByName,
+        readOrSearchBeforeMutation: [...new Set(readOrSearchBeforeMutation)],
+    };
+}
+export function readTraceFile(path) {
+    if (!existsSync(path)) {
+        return [];
+    }
+    const parsed = JSON.parse(readFileSync(path, 'utf8'));
+    return Array.isArray(parsed) ? parsed.filter(isTraceEvent) : [];
+}
+function isTraceEvent(value) {
+    return Boolean(value && typeof value === 'object' && !Array.isArray(value) && typeof value.type === 'string');
+}
+function summarizeToolInput(tool, input) {
+    const object = input && typeof input === 'object' && !Array.isArray(input) ? input : undefined;
+    const path = object && typeof object.path === 'string' ? object.path : undefined;
+    const query = object && typeof object.query === 'string' ? object.query : undefined;
+    const command = object && typeof object.command === 'string' ? object.command : undefined;
+    return `${tool}:${path ?? query ?? command ?? ''}`.slice(0, 160);
+}
+function isVerificationCommand(input) {
+    if (!input || typeof input !== 'object' || Array.isArray(input)) {
+        return false;
+    }
+    const command = input.command;
+    if (typeof command !== 'string') {
+        return false;
+    }
+    return /\b(yarn|npm|pnpm|vitest|jest|mocha|tsc|eslint|cargo|go|pytest|python|ruff)\b/.test(command)
+        && /\b(test|build|lint|typecheck|check|vitest|tsc|pytest|ruff)\b/.test(command);
+}
+//# sourceMappingURL=trace-analyzer.js.map

package/dist/src/core/eval/trace-analyzer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"trace-analyzer.js","sourceRoot":"","sources":["../../../../src/core/eval/trace-analyzer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AAInD,MAAM,cAAc,GAAG,IAAI,GAAG,CAAC,CAAC,WAAW,EAAE,aAAa,EAAE,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC;AAC9F,MAAM,oBAAoB,GAAG,IAAI,GAAG,CAAC,CAAC,WAAW,EAAE,YAAY,EAAE,cAAc,EAAE,mBAAmB,CAAC,CAAC,CAAC;AAEvG,MAAM,UAAU,iBAAiB,CAAC,KAAe;IAC/C,OAAO,YAAY,CAAC,KAAK,CAAC,OAAO,CAAC,aAAa,CAAC,CAAC,CAAC;AACpD,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,KAAmB;IAC9C,MAAM,WAAW,GAA2B,EAAE,CAAC;IAC/C,MAAM,0BAA0B,GAAa,EAAE,CAAC;IAChD,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,kBAAkB,GAAG,CAAC,CAAC;IAC3B,IAAI,iBAAiB,GAAG,CAAC,CAAC;IAC1B,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,iCAAiC,GAAG,CAAC,CAAC;IAC1C,MAAM,0BAA0B,GAAa,EAAE,CAAC;IAChD,IAAI,iBAAqC,CAAC;IAC1C,IAAI,OAA2B,CAAC;IAChC,IAAI,OAA2B,CAAC;IAEhC,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,EAAE,CAAC;YACpC,cAAc,EAAE,CAAC;YACjB,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,SAAS,IAAI,EAAE,EAAE,CAAC;gBACzC,SAAS,EAAE,CAAC;gBACZ,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;gBAE3D,IAAI,oBAAoB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,iBAAiB,KAAK,SAAS,EAAE,CAAC;oBAC3E,0BAA0B,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;gBAC7E,CAAC;gBAED,IAAI,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;oBAClC,SAAS,EAAE,CAAC;oBACZ,iBAAiB,KAAK,KAAK,CAAC,IAAI,CAAC;gBACnC,CAAC;gBAED,IACE,iBAAiB,KAAK,SAAS;uBAC5B,CAAC,IAAI,CAAC,IAAI,KAAK,kBAAkB,IAAI,IAAI,CAAC,IAAI,KAAK,mBAAmB,CAAC;uBACvE,qBAAqB,CAAC,IAAI,CAAC,KAAK,CAAC,EACpC,CAAC;oBACD,iCAAiC,EAAE,CAAC;oBACpC,0BAA0B,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;gBAC7E,CAAC;YACH,CAAC;YACD,SAAS;QACX,CAAC;QAED,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa,EAAE,CAAC;YACjC,WAAW,EAAE,CAAC;YACd,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;gBACrB,UAAU,EAAE,CAAC;YACf,CAAC;QACH,CAAC;QAED,IAAI,KAAK,CAAC,IAAI,KAAK,yBAAyB,EAAE,CAAC;YAC7C,kBAAkB,EAAE,CAAC;QACvB,CAAC;QAED,IAAI,KAAK,CAAC,IAAI,KAAK,wBAAwB,EAAE,CAAC;YAC5C,iBAAiB,EAAE,CAAC;QACtB,CAAC;QAED,IAAI,KAAK,CAAC,IAAI,KAAK,cAAc,EAAE,CAAC;YAClC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;YACxB,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;QAC1B,CAAC;IACH,CAAC;IAED,OAAO;QACL,cAAc;QACd,SAAS;QACT,WAAW;QACX,SAAS;QACT,kBAAkB;QAClB,iBAAiB;QACjB,UAAU;QACV,iCAAiC;QACjC,0BAA0B,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,0BAA0B,CAAC,CAAC;QACpE,iBAAiB;QACjB,OAAO;QACP,OAAO;QACP,WAAW;QACX,0BAA0B,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,0BAA0B,CAAC,CAAC;KACrE,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,IAAY;IACxC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;QACtB,OAAO,EAAE,CAAC;IACZ,CAAC;IACD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,MAAM,CAAC,CAAY,CAAC;IACjE,OAAO,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAClE,CAAC;AAED,SAAS,YAAY,CAAC,KAAc;IAClC,OAAO,OAAO,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,OAAQ,KAA4B,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC;AACxI,CAAC;AAED,SAAS,kBAAkB,CAAC,IAAY,EAAE,KAAc;IACtD,MAAM,MAAM,GAAG,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAgC,CAAC,CAAC,CAAC,SAAS,CAAC;IAC1H,MAAM,IAAI,GAAG,MAAM,IAAI,OAAO,MAAM,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;IACjF,MAAM,KAAK,GAAG,MAAM,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;IACpF,MAAM,OAAO,GAAG,MAAM,IAAI,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC;IAC1F,OAAO,GAAG,IAAI,IAAI,IAAI,IAAI,KAAK,IAAI,OAAO,IAAI,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;AACnE,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAc;IAC3C,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QAChE,OAAO,KAAK,CAAC;IACf,CAAC;IACD,MAAM,OAAO,GAAI,KAA+B,CAAC,OAAO,CAAC;IACzD,IAAI,OAAO,OAAO,KAAK,QAAQ,EAAE,CAAC;QAChC,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,8EAA8E,CAAC,IAAI,CAAC,OAAO,CAAC;WAC9F,8DAA8D,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;AACpF,CAAC"}

package/dist/src/core/eval/workspace-fixture.d.ts

@@ -0,0 +1,14 @@
+import type { EvalProgressReporter } from './progress.js';
+import type { AgentEvalCase, EvalCheckResult, EvalRunResult } from './schema.js';
+export type PreparedEvalWorkspace = {
+    workspaceRoot: string;
+    setupResults: EvalCheckResult[];
+    fixture: EvalRunResult['fixture'];
+};
+export declare function prepareEvalWorkspace(args: {
+    testCase: AgentEvalCase;
+    repoRoot: string;
+    workRoot?: string;
+    progress?: EvalProgressReporter;
+}): Promise<PreparedEvalWorkspace>;
+//# sourceMappingURL=workspace-fixture.d.ts.map

package/dist/src/core/eval/workspace-fixture.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"workspace-fixture.d.ts","sourceRoot":"","sources":["../../../../src/core/eval/workspace-fixture.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,eAAe,CAAC;AAC1D,OAAO,KAAK,EAAE,aAAa,EAAE,eAAe,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAEjF,MAAM,MAAM,qBAAqB,GAAG;IAClC,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,eAAe,EAAE,CAAC;IAChC,OAAO,EAAE,aAAa,CAAC,SAAS,CAAC,CAAC;CACnC,CAAC;AAEF,wBAAsB,oBAAoB,CAAC,IAAI,EAAE;IAC/C,QAAQ,EAAE,aAAa,CAAC;IACxB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,oBAAoB,CAAC;CACjC,GAAG,OAAO,CAAC,qBAAqB,CAAC,CA8BjC"}

package/dist/src/core/eval/workspace-fixture.js

@@ -0,0 +1,235 @@
+import { appendFileSync, cpSync, existsSync, mkdirSync, mkdtempSync, rmSync, writeFileSync } from 'node:fs';
+import { dirname, join, resolve } from 'node:path';
+import { tmpdir } from 'node:os';
+import { runCommand, runShellCommand } from './process.js';
+export async function prepareEvalWorkspace(args) {
+    const parent = args.workRoot ? resolve(args.workRoot) : mkdtempSync(join(tmpdir(), 'heddle-eval-'));
+    mkdirSync(parent, { recursive: true });
+    const workspaceRoot = join(parent, args.testCase.id);
+    const fixture = args.testCase.fixture;
+    if (fixture.type === 'inline') {
+        rmSync(workspaceRoot, { recursive: true, force: true });
+    }
+    const prepared = fixture.type === 'git-worktree' ?
+        await prepareGitWorktreeFixture({
+            repoRoot: args.repoRoot,
+            workspaceRoot,
+            testCase: args.testCase,
+            progress: args.progress,
+        })
+        : await prepareInlineFixture({
+            repoRoot: args.repoRoot,
+            workspaceRoot,
+            testCase: args.testCase,
+            progress: args.progress,
+        });
+    await copyAuthState({
+        repoRoot: args.repoRoot,
+        workspaceRoot,
+    });
+    return prepared;
+}
+async function prepareInlineFixture(args) {
+    mkdirSync(args.workspaceRoot, { recursive: true });
+    const setupResults = await applyEvalSetup({
+        workspaceRoot: args.workspaceRoot,
+        repoRoot: args.repoRoot,
+        testCase: args.testCase,
+        progress: args.progress,
+    });
+    const baselineCommit = await trackProgress(args.progress, {
+        phase: 'workspace.baseline',
+        message: 'commit inline fixture baseline',
+        run: () => initializeGitBaseline(args.workspaceRoot, args.testCase.setup.commitMessage ?? `Initial eval fixture for ${args.testCase.id}`),
+    });
+    return {
+        workspaceRoot: args.workspaceRoot,
+        setupResults,
+        fixture: {
+            type: 'inline',
+            baselineCommit,
+        },
+    };
+}
+async function prepareGitWorktreeFixture(args) {
+    const fixture = args.testCase.fixture;
+    if (fixture.type !== 'git-worktree') {
+        throw new Error(`Unsupported fixture type for git worktree setup: ${fixture.type}`);
+    }
+    const sourceRepo = resolve(args.repoRoot, fixture.repo);
+    const resolvedRef = await trackProgress(args.progress, {
+        phase: 'workspace.resolve-ref',
+        message: `resolve target ref ${fixture.ref}`,
+        run: () => resolveGitCommit(sourceRepo, fixture.ref),
+    });
+    await trackProgress(args.progress, {
+        phase: 'workspace.cleanup',
+        message: 'remove existing dogfood worktree if needed',
+        run: () => removeGitWorktree(sourceRepo, args.workspaceRoot),
+    });
+    await trackProgress(args.progress, {
+        phase: 'workspace.worktree',
+        message: `create pinned worktree at ${fixture.ref}`,
+        run: () => runRequiredGit(sourceRepo, ['worktree', 'add', '--detach', args.workspaceRoot, resolvedRef]),
+    });
+    const setupResults = await applyEvalSetup({
+        workspaceRoot: args.workspaceRoot,
+        repoRoot: args.repoRoot,
+        testCase: args.testCase,
+        progress: args.progress,
+    });
+    const baselineCommit = await hasGitChanges(args.workspaceRoot) ?
+        await trackProgress(args.progress, {
+            phase: 'workspace.baseline',
+            message: 'commit seeded eval baseline',
+            run: () => commitAll(args.workspaceRoot, args.testCase.setup.commitMessage ?? `Eval setup for ${args.testCase.id}`),
+        })
+        : resolvedRef;
+    return {
+        workspaceRoot: args.workspaceRoot,
+        setupResults,
+        fixture: {
+            type: 'git-worktree',
+            repo: sourceRepo,
+            ref: fixture.ref,
+            resolvedRef,
+            baselineCommit,
+        },
+    };
+}
+async function applyEvalSetup(args) {
+    for (const [sourceRelativePath, targetRelativePath] of Object.entries(args.testCase.setup.copyFiles ?? {})) {
+        const sourcePath = resolve(args.repoRoot, sourceRelativePath);
+        const targetPath = join(args.workspaceRoot, targetRelativePath);
+        mkdirSync(dirname(targetPath), { recursive: true });
+        cpSync(sourcePath, targetPath);
+    }
+    for (const [relativePath, content] of Object.entries(args.testCase.setup.files ?? {})) {
+        const path = join(args.workspaceRoot, relativePath);
+        mkdirSync(dirname(path), { recursive: true });
+        writeFileSync(path, content, 'utf8');
+    }
+    const setupResults = [];
+    for (const command of args.testCase.setup.commands ?? []) {
+        const name = command.name ?? command.command;
+        const result = await trackProgress(args.progress, {
+            phase: 'workspace.setup-command',
+            message: `run setup command: ${name}`,
+            heartbeatMessage: `still running setup command: ${name}`,
+            run: () => runShellCommand({
+                command: command.command,
+                cwd: args.workspaceRoot,
+                timeoutMs: command.timeoutMs,
+            }),
+        });
+        setupResults.push({
+            name,
+            command: command.command,
+            exitCode: result.exitCode,
+            stdout: result.stdout,
+            stderr: result.stderr,
+            durationMs: result.durationMs,
+            passed: result.exitCode === 0 && !result.timedOut,
+            timedOut: result.timedOut,
+        });
+        if (result.exitCode !== 0 || result.timedOut) {
+            throw new Error(`Eval setup command failed for ${args.testCase.id}: ${command.command}`);
+        }
+    }
+    return setupResults;
+}
+async function copyAuthState(args) {
+    const memorySource = join(args.repoRoot, '.heddle', 'auth.json');
+    const memoryTarget = join(args.workspaceRoot, '.heddle', 'auth.json');
+    if (existsSync(memorySource) && !existsSync(memoryTarget)) {
+        await excludeAuthStateFromGit(args.workspaceRoot);
+        mkdirSync(dirname(memoryTarget), { recursive: true });
+        cpSync(memorySource, memoryTarget);
+    }
+}
+async function excludeAuthStateFromGit(workspaceRoot) {
+    const result = await runCommand({
+        command: 'git',
+        args: ['rev-parse', '--git-path', 'info/exclude'],
+        cwd: workspaceRoot,
+        timeoutMs: 10_000,
+    });
+    if (result.exitCode !== 0 || result.timedOut) {
+        throw new Error(`git rev-parse --git-path info/exclude failed: ${result.stderr || result.stdout}`);
+    }
+    const excludePath = resolve(workspaceRoot, result.stdout.trim());
+    if (!excludePath) {
+        return;
+    }
+    mkdirSync(dirname(excludePath), { recursive: true });
+    appendFileSync(excludePath, '\n/.heddle/auth.json\n', 'utf8');
+}
+async function trackProgress(progress, args) {
+    return progress ? await progress.track(args) : await args.run();
+}
+async function initializeGitBaseline(workspaceRoot, message) {
+    await runRequiredGit(workspaceRoot, ['init']);
+    return await commitAll(workspaceRoot, message, true);
+}
+async function commitAll(workspaceRoot, message, allowEmpty = false) {
+    await runRequiredGit(workspaceRoot, ['add', '.']);
+    await runRequiredGit(workspaceRoot, [
+        '-c',
+        'user.name=Heddle Eval',
+        '-c',
+        'user.email=heddle-eval@example.com',
+        '-c',
+        'commit.gpgsign=false',
+        'commit',
+        ...(allowEmpty ? ['--allow-empty'] : []),
+        '-m',
+        message,
+    ]);
+    return await resolveGitCommit(workspaceRoot, 'HEAD');
+}
+async function resolveGitCommit(cwd, ref) {
+    const result = await runCommand({
+        command: 'git',
+        args: ['rev-parse', '--verify', `${ref}^{commit}`],
+        cwd,
+        timeoutMs: 20_000,
+    });
+    if (result.exitCode !== 0 || result.timedOut) {
+        throw new Error(`git rev-parse failed for ${ref}: ${result.stderr || result.stdout}`);
+    }
+    return result.stdout.trim();
+}
+async function hasGitChanges(workspaceRoot) {
+    const result = await runCommand({
+        command: 'git',
+        args: ['status', '--porcelain'],
+        cwd: workspaceRoot,
+        timeoutMs: 20_000,
+    });
+    if (result.exitCode !== 0 || result.timedOut) {
+        throw new Error(`git status --porcelain failed: ${result.stderr || result.stdout}`);
+    }
+    return result.stdout.trim().length > 0;
+}
+async function removeGitWorktree(repoRoot, workspaceRoot) {
+    if (!existsSync(workspaceRoot)) {
+        return;
+    }
+    const result = await runCommand({
+        command: 'git',
+        args: ['worktree', 'remove', '--force', workspaceRoot],
+        cwd: repoRoot,
+        timeoutMs: 20_000,
+    });
+    if (result.exitCode !== 0 || result.timedOut) {
+        rmSync(workspaceRoot, { recursive: true, force: true });
+        await runRequiredGit(repoRoot, ['worktree', 'prune']);
+    }
+}
+async function runRequiredGit(cwd, args) {
+    const result = await runCommand({ command: 'git', args, cwd, timeoutMs: 20_000 });
+    if (result.exitCode !== 0 || result.timedOut) {
+        throw new Error(`git ${args.join(' ')} failed: ${result.stderr || result.stdout}`);
+    }
+}
+//# sourceMappingURL=workspace-fixture.js.map

package/dist/src/core/eval/workspace-fixture.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"workspace-fixture.js","sourceRoot":"","sources":["../../../../src/core/eval/workspace-fixture.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAC5G,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACnD,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,MAAM,cAAc,CAAC;AAU3D,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,IAK1C;IACC,MAAM,MAAM,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,cAAc,CAAC,CAAC,CAAC;IACpG,SAAS,CAAC,MAAM,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACvC,MAAM,aAAa,GAAG,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;IAErD,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC;IACtC,IAAI,OAAO,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;QAC9B,MAAM,CAAC,aAAa,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;IAC1D,CAAC;IACD,MAAM,QAAQ,GACZ,OAAO,CAAC,IAAI,KAAK,cAAc,CAAC,CAAC;QAC/B,MAAM,yBAAyB,CAAC;YAC9B,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,aAAa;YACb,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ;SACxB,CAAC;QACJ,CAAC,CAAC,MAAM,oBAAoB,CAAC;YACzB,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,aAAa;YACb,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ;SACxB,CAAC,CAAC;IAEP,MAAM,aAAa,CAAC;QAClB,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,aAAa;KACd,CAAC,CAAC;IAEH,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,KAAK,UAAU,oBAAoB,CAAC,IAKnC;IACC,SAAS,CAAC,IAAI,CAAC,aAAa,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACnD,MAAM,YAAY,GAAG,MAAM,cAAc,CAAC;QACxC,aAAa,EAAE,IAAI,CAAC,aAAa;QACjC,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ;KACxB,CAAC,CAAC;IACH,MAAM,cAAc,GAAG,MAAM,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE;QACxD,KAAK,EAAE,oBAAoB;QAC3B,OAAO,EAAE,gCAAgC;QACzC,GAAG,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAC9B,IAAI,CAAC,aAAa,EAClB,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,aAAa,IAAI,4BAA4B,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CACpF;KACF,CAAC,CAAC;IAEH,OAAO;QACL,aAAa,EAAE,IAAI,CAAC,aAAa;QACjC,YAAY;QACZ,OAAO,EAAE;YACP,IAAI,EAAE,QAAQ;YACd,cAAc;SACf;KACF,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,yBAAyB,CAAC,IAKxC;IACC,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC;IACtC,IAAI,OAAO,CAAC,IAAI,KAAK,cAAc,EAAE,CAAC;QACpC,MAAM,IAAI,KAAK,CAAC,oDAAoD,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IACtF,CAAC;IAED,MAAM,UAAU,GAAG,OAAO,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,IAAI,CAAC,CAAC;IACxD,MAAM,WAAW,GAAG,MAAM,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE;QACrD,KAAK,EAAE,uBAAuB;QAC9B,OAAO,EAAE,sBAAsB,OAAO,CAAC,GAAG,EAAE;QAC5C,GAAG,EAAE,GAAG,EAAE,CAAC,gBAAgB,CAAC,UAAU,EAAE,OAAO,CAAC,GAAG,CAAC;KACrD,CAAC,CAAC;IACH,MAAM,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE;QACjC,KAAK,EAAE,mBAAmB;QAC1B,OAAO,EAAE,4CAA4C;QACrD,GAAG,EAAE,GAAG,EAAE,CAAC,iBAAiB,CAAC,UAAU,EAAE,IAAI,CAAC,aAAa,CAAC;KAC7D,CAAC,CAAC;IACH,MAAM,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE;QACjC,KAAK,EAAE,oBAAoB;QAC3B,OAAO,EAAE,6BAA6B,OAAO,CAAC,GAAG,EAAE;QACnD,GAAG,EAAE,GAAG,EAAE,CAAC,cAAc,CAAC,UAAU,EAAE,CAAC,UAAU,EAAE,KAAK,EAAE,UAAU,EAAE,IAAI,CAAC,aAAa,EAAE,WAAW,CAAC,CAAC;KACxG,CAAC,CAAC;IAEH,MAAM,YAAY,GAAG,MAAM,cAAc,CAAC;QACxC,aAAa,EAAE,IAAI,CAAC,aAAa;QACjC,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ;KACxB,CAAC,CAAC;IACH,MAAM,cAAc,GAClB,MAAM,aAAa,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC;QACvC,MAAM,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE;YACjC,KAAK,EAAE,oBAAoB;YAC3B,OAAO,EAAE,6BAA6B;YACtC,GAAG,EAAE,GAAG,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,aAAa,EAAE,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,aAAa,IAAI,kBAAkB,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;SACpH,CAAC;QACJ,CAAC,CAAC,WAAW,CAAC;IAEhB,OAAO;QACL,aAAa,EAAE,IAAI,CAAC,aAAa;QACjC,YAAY;QACZ,OAAO,EAAE;YACP,IAAI,EAAE,cAAc;YACpB,IAAI,EAAE,UAAU;YAChB,GAAG,EAAE,OAAO,CAAC,GAAG;YAChB,WAAW;YACX,cAAc;SACf;KACF,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,cAAc,CAAC,IAK7B;IACC,KAAK,MAAM,CAAC,kBAAkB,EAAE,kBAAkB,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,SAAS,IAAI,EAAE,CAAC,EAAE,CAAC;QAC3G,MAAM,UAAU,GAAG,OAAO,CAAC,IAAI,CAAC,QAAQ,EAAE,kBAAkB,CAAC,CAAC;QAC9D,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,kBAAkB,CAAC,CAAC;QAChE,SAAS,CAAC,OAAO,CAAC,UAAU,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACpD,MAAM,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IACjC,CAAC;IAED,KAAK,MAAM,CAAC,YAAY,EAAE,OAAO,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,KAAK,IAAI,EAAE,CAAC,EAAE,CAAC;QACtF,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,YAAY,CAAC,CAAC;QACpD,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC9C,aAAa,CAAC,IAAI,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACvC,CAAC;IAED,MAAM,YAAY,GAAsB,EAAE,CAAC;IAC3C,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,QAAQ,IAAI,EAAE,EAAE,CAAC;QACzD,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,IAAI,OAAO,CAAC,OAAO,CAAC;QAC7C,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE;YAChD,KAAK,EAAE,yBAAyB;YAChC,OAAO,EAAE,sBAAsB,IAAI,EAAE;YACrC,gBAAgB,EAAE,gCAAgC,IAAI,EAAE;YACxD,GAAG,EAAE,GAAG,EAAE,CAAC,eAAe,CAAC;gBACzB,OAAO,EAAE,OAAO,CAAC,OAAO;gBACxB,GAAG,EAAE,IAAI,CAAC,aAAa;gBACvB,SAAS,EAAE,OAAO,CAAC,SAAS;aAC7B,CAAC;SACH,CAAC,CAAC;QACH,YAAY,CAAC,IAAI,CAAC;YAChB,IAAI;YACJ,OAAO,EAAE,OAAO,CAAC,OAAO;YACxB,QAAQ,EAAE,MAAM,CAAC,QAAQ;YACzB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,UAAU,EAAE,MAAM,CAAC,UAAU;YAC7B,MAAM,EAAE,MAAM,CAAC,QAAQ,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ;YACjD,QAAQ,EAAE,MAAM,CAAC,QAAQ;SAC1B,CAAC,CAAC;QACH,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YAC7C,MAAM,IAAI,KAAK,CAAC,iCAAiC,IAAI,CAAC,QAAQ,CAAC,EAAE,KAAK,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3F,CAAC;IACH,CAAC;IACD,OAAO,YAAY,CAAC;AACtB,CAAC;AAED,KAAK,UAAU,aAAa,CAAC,IAG5B;IACC,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,SAAS,EAAE,WAAW,CAAC,CAAC;IACjE,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,SAAS,EAAE,WAAW,CAAC,CAAC;IACtE,IAAI,UAAU,CAAC,YAAY,CAAC,IAAI,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,CAAC;QAC1D,MAAM,uBAAuB,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QAClD,SAAS,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACtD,MAAM,CAAC,YAAY,EAAE,YAAY,CAAC,CAAC;IACrC,CAAC;AACH,CAAC;AAED,KAAK,UAAU,uBAAuB,CAAC,aAAqB;IAC1D,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC;QAC9B,OAAO,EAAE,KAAK;QACd,IAAI,EAAE,CAAC,WAAW,EAAE,YAAY,EAAE,cAAc,CAAC;QACjD,GAAG,EAAE,aAAa;QAClB,SAAS,EAAE,MAAM;KAClB,CAAC,CAAC;IACH,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;QAC7C,MAAM,IAAI,KAAK,CAAC,iDAAiD,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IACrG,CAAC;IAED,MAAM,WAAW,GAAG,OAAO,CAAC,aAAa,EAAE,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;IACjE,IAAI,CAAC,WAAW,EAAE,CAAC;QACjB,OAAO;IACT,CAAC;IACD,SAAS,CAAC,OAAO,CAAC,WAAW,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACrD,cAAc,CAAC,WAAW,EAAE,wBAAwB,EAAE,MAAM,CAAC,CAAC;AAChE,CAAC;AAED,KAAK,UAAU,aAAa,CAC1B,QAA0C,EAC1C,IAMC;IAED,OAAO,QAAQ,CAAC,CAAC,CAAC,MAAM,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,GAAG,EAAE,CAAC;AAClE,CAAC;AAED,KAAK,UAAU,qBAAqB,CAAC,aAAqB,EAAE,OAAe;IACzE,MAAM,cAAc,CAAC,aAAa,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;IAC9C,OAAO,MAAM,SAAS,CAAC,aAAa,EAAE,OAAO,EAAE,IAAI,CAAC,CAAC;AACvD,CAAC;AAED,KAAK,UAAU,SAAS,CAAC,aAAqB,EAAE,OAAe,EAAE,UAAU,GAAG,KAAK;IACjF,MAAM,cAAc,CAAC,aAAa,EAAE,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,CAAC;IAClD,MAAM,cAAc,CAAC,aAAa,EAAE;QAClC,IAAI;QACJ,uBAAuB;QACvB,IAAI;QACJ,oCAAoC;QACpC,IAAI;QACJ,sBAAsB;QACtB,QAAQ;QACR,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QACxC,IAAI;QACJ,OAAO;KACR,CAAC,CAAC;IACH,OAAO,MAAM,gBAAgB,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC;AACvD,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,GAAW,EAAE,GAAW;IACtD,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC;QAC9B,OAAO,EAAE,KAAK;QACd,IAAI,EAAE,CAAC,WAAW,EAAE,UAAU,EAAE,GAAG,GAAG,WAAW,CAAC;QAClD,GAAG;QACH,SAAS,EAAE,MAAM;KAClB,CAAC,CAAC;IACH,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;QAC7C,MAAM,IAAI,KAAK,CAAC,4BAA4B,GAAG,KAAK,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IACxF,CAAC;IACD,OAAO,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;AAC9B,CAAC;AAED,KAAK,UAAU,aAAa,CAAC,aAAqB;IAChD,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC;QAC9B,OAAO,EAAE,KAAK;QACd,IAAI,EAAE,CAAC,QAAQ,EAAE,aAAa,CAAC;QAC/B,GAAG,EAAE,aAAa;QAClB,SAAS,EAAE,MAAM;KAClB,CAAC,CAAC;IACH,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;QAC7C,MAAM,IAAI,KAAK,CAAC,kCAAkC,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IACtF,CAAC;IACD,OAAO,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;AACzC,CAAC;AAED,KAAK,UAAU,iBAAiB,CAAC,QAAgB,EAAE,aAAqB;IACtE,IAAI,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,CAAC;QAC/B,OAAO;IACT,CAAC;IACD,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC;QAC9B,OAAO,EAAE,KAAK;QACd,IAAI,EAAE,CAAC,UAAU,EAAE,QAAQ,EAAE,SAAS,EAAE,aAAa,CAAC;QACtD,GAAG,EAAE,QAAQ;QACb,SAAS,EAAE,MAAM;KAClB,CAAC,CAAC;IACH,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;QAC7C,MAAM,CAAC,aAAa,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACxD,MAAM,cAAc,CAAC,QAAQ,EAAE,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC;IACxD,CAAC;AACH,CAAC;AAED,KAAK,UAAU,cAAc,CAAC,GAAW,EAAE,IAAc;IACvD,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC,CAAC;IAClF,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;QAC7C,MAAM,IAAI,KAAK,CAAC,OAAO,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,YAAY,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IACrF,CAAC;AACH,CAAC"}

package/dist/src/core/llm/model-policy.d.ts

@@ -2,9 +2,16 @@ import type { ProviderCredentialSource } from '../runtime/api-keys.js';
 import type { LlmProvider } from './types.js';
 export type SystemModelPurpose = 'chat-compaction' | 'session-title';
 export type ModelCredentialMode = 'api-key' | 'oauth' | 'missing';
+export type CredentialAwareModelOption = {
+    id: string;
+    label?: string;
+    disabled: boolean;
+    disabledReason?: string;
+};
 export declare const OPENAI_API_KEY_COMPACTION_MODEL = "gpt-5.1-codex-mini";
 export declare const OPENAI_OAUTH_SYSTEM_MODEL = "gpt-5.4";
 export declare const ANTHROPIC_COMPACTION_MODEL = "claude-haiku-4-5";
+export declare const OPENAI_OAUTH_MODE_DESCRIPTION = "OAuth mode supports a smaller OpenAI allowlist.";
 export declare function credentialModeFromSource(source: ProviderCredentialSource | undefined): ModelCredentialMode;
 export declare function resolveSystemSelectedModel(args: {
     purpose: SystemModelPurpose;
@@ -30,5 +37,24 @@ export declare function assertModelCredentialCompatibility(args: {
     usageLabel?: string;
 }): void;
 export declare function resolveOpenAiOAuthImageCandidateModels(activeModel: string): string[];
+export declare function buildCredentialAwareModelOption(args: {
+    model: string;
+    provider: LlmProvider;
+    credentialMode?: ModelCredentialMode;
+    label?: string;
+}): CredentialAwareModelOption;
+export declare function summarizeCredentialCompatibilityError(args: {
+    model: string;
+    provider: LlmProvider;
+    credentialMode?: ModelCredentialMode;
+}): string | undefined;
+export declare function resolveCompatibleActiveModel(args: {
+    activeModel: string;
+    provider: LlmProvider;
+    credentialMode?: ModelCredentialMode;
+}): {
+    model: string;
+    warning?: string;
+};
 export declare function formatOpenAiAccountSignInModels(): string;
 //# sourceMappingURL=model-policy.d.ts.map

package/dist/src/core/llm/model-policy.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"model-policy.d.ts","sourceRoot":"","sources":["../../../../src/core/llm/model-policy.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,wBAAwB,EAAE,MAAM,wBAAwB,CAAC;AAEvE,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAE9C,MAAM,MAAM,kBAAkB,GAAG,iBAAiB,GAAG,eAAe,CAAC;AACrE,MAAM,MAAM,mBAAmB,GAAG,SAAS,GAAG,OAAO,GAAG,SAAS,CAAC;AAElE,eAAO,MAAM,+BAA+B,uBAAuB,CAAC;AACpE,eAAO,MAAM,yBAAyB,YAAY,CAAC;AACnD,eAAO,MAAM,0BAA0B,qBAAqB,CAAC;AAI7D,wBAAgB,wBAAwB,CAAC,MAAM,EAAE,wBAAwB,GAAG,SAAS,GAAG,mBAAmB,CAU1G;AAED,wBAAgB,0BAA0B,CAAC,IAAI,EAAE;IAC/C,OAAO,EAAE,kBAAkB,CAAC;IAC5B,QAAQ,EAAE,WAAW,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,cAAc,CAAC,EAAE,mBAAmB,CAAC;CACtC,GAAG,MAAM,CAcT;AAED,wBAAgB,oCAAoC,CAAC,IAAI,EAAE;IACzD,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,WAAW,CAAC;IACtB,cAAc,CAAC,EAAE,mBAAmB,CAAC;IACrC,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG;IAAE,EAAE,EAAE,IAAI,CAAA;CAAE,GAAG;IAAE,EAAE,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAS9C;AAED,wBAAgB,kCAAkC,CAAC,IAAI,EAAE;IACvD,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,WAAW,CAAC;IACtB,cAAc,CAAC,EAAE,mBAAmB,CAAC;IACrC,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,QAKA;AAED,wBAAgB,sCAAsC,CAAC,WAAW,EAAE,MAAM,GAAG,MAAM,EAAE,CAEpF;AAED,wBAAgB,+BAA+B,IAAI,MAAM,CAExD"}
+{"version":3,"file":"model-policy.d.ts","sourceRoot":"","sources":["../../../../src/core/llm/model-policy.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,wBAAwB,EAAE,MAAM,wBAAwB,CAAC;AAEvE,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAE9C,MAAM,MAAM,kBAAkB,GAAG,iBAAiB,GAAG,eAAe,CAAC;AACrE,MAAM,MAAM,mBAAmB,GAAG,SAAS,GAAG,OAAO,GAAG,SAAS,CAAC;AAElE,MAAM,MAAM,0BAA0B,GAAG;IACvC,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,OAAO,CAAC;IAClB,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB,CAAC;AAEF,eAAO,MAAM,+BAA+B,uBAAuB,CAAC;AACpE,eAAO,MAAM,yBAAyB,YAAY,CAAC;AACnD,eAAO,MAAM,0BAA0B,qBAAqB,CAAC;AAE7D,eAAO,MAAM,6BAA6B,oDAAoD,CAAC;AAI/F,wBAAgB,wBAAwB,CAAC,MAAM,EAAE,wBAAwB,GAAG,SAAS,GAAG,mBAAmB,CAU1G;AAED,wBAAgB,0BAA0B,CAAC,IAAI,EAAE;IAC/C,OAAO,EAAE,kBAAkB,CAAC;IAC5B,QAAQ,EAAE,WAAW,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,cAAc,CAAC,EAAE,mBAAmB,CAAC;CACtC,GAAG,MAAM,CAcT;AAED,wBAAgB,oCAAoC,CAAC,IAAI,EAAE;IACzD,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,WAAW,CAAC;IACtB,cAAc,CAAC,EAAE,mBAAmB,CAAC;IACrC,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG;IAAE,EAAE,EAAE,IAAI,CAAA;CAAE,GAAG;IAAE,EAAE,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAS9C;AAED,wBAAgB,kCAAkC,CAAC,IAAI,EAAE;IACvD,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,WAAW,CAAC;IACtB,cAAc,CAAC,EAAE,mBAAmB,CAAC;IACrC,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,QAKA;AAED,wBAAgB,sCAAsC,CAAC,WAAW,EAAE,MAAM,GAAG,MAAM,EAAE,CAEpF;AAED,wBAAgB,+BAA+B,CAAC,IAAI,EAAE;IACpD,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,WAAW,CAAC;IACtB,cAAc,CAAC,EAAE,mBAAmB,CAAC;IACrC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB,GAAG,0BAA0B,CAa7B;AAED,wBAAgB,qCAAqC,CAAC,IAAI,EAAE;IAC1D,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,WAAW,CAAC;IACtB,cAAc,CAAC,EAAE,mBAAmB,CAAC;CACtC,GAAG,MAAM,GAAG,SAAS,CAMrB;AAED,wBAAgB,4BAA4B,CAAC,IAAI,EAAE;IACjD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,WAAW,CAAC;IACtB,cAAc,CAAC,EAAE,mBAAmB,CAAC;CACtC,GAAG;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,MAAM,CAAA;CAAE,CA4BtC;AAED,wBAAgB,+BAA+B,IAAI,MAAM,CAExD"}

package/dist/src/core/llm/model-policy.js

@@ -3,6 +3,8 @@ import { isOpenAiAccountSignInModel, OPENAI_ACCOUNT_SIGN_IN_MODELS } from './ope
 export const OPENAI_API_KEY_COMPACTION_MODEL = 'gpt-5.1-codex-mini';
 export const OPENAI_OAUTH_SYSTEM_MODEL = 'gpt-5.4';
 export const ANTHROPIC_COMPACTION_MODEL = 'claude-haiku-4-5';
+const OPENAI_OAUTH_DISABLED_REASON = 'Not supported';
+export const OPENAI_OAUTH_MODE_DESCRIPTION = 'OAuth mode supports a smaller OpenAI allowlist.';
 const OPENAI_OAUTH_IMAGE_MODEL_PREFERENCES = ['gpt-5.4', 'gpt-5.4-mini'];
 export function credentialModeFromSource(source) {
     if (source?.type === 'oauth') {
@@ -43,6 +45,51 @@ export function assertModelCredentialCompatibility(args) {
 export function resolveOpenAiOAuthImageCandidateModels(activeModel) {
     return uniqueModels([activeModel, ...OPENAI_OAUTH_IMAGE_MODEL_PREFERENCES, ...OPENAI_ACCOUNT_SIGN_IN_MODELS]);
 }
+export function buildCredentialAwareModelOption(args) {
+    const compatibility = validateModelCredentialCompatibility({
+        model: args.model,
+        provider: args.provider,
+        credentialMode: args.credentialMode,
+    });
+    return {
+        id: args.model,
+        label: args.label,
+        disabled: !compatibility.ok,
+        disabledReason: compatibility.ok ? undefined : summarizeCredentialCompatibilityError(args),
+    };
+}
+export function summarizeCredentialCompatibilityError(args) {
+    if (args.provider === 'openai' && args.credentialMode === 'oauth' && !isOpenAiAccountSignInModel(args.model)) {
+        return OPENAI_OAUTH_DISABLED_REASON;
+    }
+    return undefined;
+}
+export function resolveCompatibleActiveModel(args) {
+    const compatibility = validateModelCredentialCompatibility({
+        model: args.activeModel,
+        provider: args.provider,
+        credentialMode: args.credentialMode,
+    });
+    if (compatibility.ok) {
+        return { model: args.activeModel };
+    }
+    const fallback = resolveSystemSelectedModel({
+        purpose: 'session-title',
+        provider: args.provider,
+        activeModel: args.activeModel,
+        credentialMode: args.credentialMode,
+    });
+    if (fallback === args.activeModel) {
+        return {
+            model: args.activeModel,
+            warning: `Model ${args.activeModel} is not supported with OpenAI account sign-in. Pick a supported OAuth model with /model set or use API-key mode.`,
+        };
+    }
+    return {
+        model: fallback,
+        warning: `Model ${args.activeModel} is not supported with OpenAI account sign-in. Switched to ${fallback} for this session.`,
+    };
+}
 export function formatOpenAiAccountSignInModels() {
     return OPENAI_ACCOUNT_SIGN_IN_MODELS.join(', ');
 }

package/dist/src/core/llm/model-policy.js.map

@@ -1 +1 @@
-{"version":3,"file":"model-policy.js","sourceRoot":"","sources":["../../../../src/core/llm/model-policy.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,uBAAuB,EAAE,oBAAoB,EAAE,MAAM,cAAc,CAAC;AAE7E,OAAO,EAAE,0BAA0B,EAAE,6BAA6B,EAAE,MAAM,oBAAoB,CAAC;AAM/F,MAAM,CAAC,MAAM,+BAA+B,GAAG,oBAAoB,CAAC;AACpE,MAAM,CAAC,MAAM,yBAAyB,GAAG,SAAS,CAAC;AACnD,MAAM,CAAC,MAAM,0BAA0B,GAAG,kBAAkB,CAAC;AAE7D,MAAM,oCAAoC,GAAG,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC;AAEzE,MAAM,UAAU,wBAAwB,CAAC,MAA4C;IACnF,IAAI,MAAM,EAAE,IAAI,KAAK,OAAO,EAAE,CAAC;QAC7B,OAAO,OAAO,CAAC;IACjB,CAAC;IAED,IAAI,MAAM,EAAE,IAAI,KAAK,SAAS,EAAE,CAAC;QAC/B,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,0BAA0B,CAAC,IAK1C;IACC,IAAI,IAAI,CAAC,QAAQ,KAAK,WAAW,EAAE,CAAC;QAClC,OAAO,IAAI,CAAC,OAAO,KAAK,iBAAiB,CAAC,CAAC,CAAC,0BAA0B,CAAC,CAAC,CAAC,uBAAuB,CAAC;IACnG,CAAC;IAED,IAAI,IAAI,CAAC,QAAQ,KAAK,QAAQ,EAAE,CAAC;QAC/B,IAAI,IAAI,CAAC,cAAc,KAAK,OAAO,EAAE,CAAC;YACpC,OAAO,IAAI,CAAC,WAAW,IAAI,0BAA0B,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,yBAAyB,CAAC;QACzH,CAAC;QAED,OAAO,IAAI,CAAC,OAAO,KAAK,iBAAiB,CAAC,CAAC,CAAC,+BAA+B,CAAC,CAAC,CAAC,oBAAoB,CAAC;IACrG,CAAC;IAED,OAAO,oBAAoB,CAAC;AAC9B,CAAC;AAED,MAAM,UAAU,oCAAoC,CAAC,IAKpD;IACC,IAAI,IAAI,CAAC,QAAQ,KAAK,QAAQ,IAAI,IAAI,CAAC,cAAc,KAAK,OAAO,IAAI,CAAC,0BAA0B,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QAC7G,OAAO;YACL,EAAE,EAAE,KAAK;YACT,KAAK,EAAE,mDAAmD,IAAI,CAAC,KAAK,gBAAgB,+BAA+B,EAAE,uDAAuD,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,GAAG;SAChO,CAAC;IACJ,CAAC;IAED,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;AACtB,CAAC;AAED,MAAM,UAAU,kCAAkC,CAAC,IAKlD;IACC,MAAM,MAAM,GAAG,oCAAoC,CAAC,IAAI,CAAC,CAAC;IAC1D,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAChC,CAAC;AACH,CAAC;AAED,MAAM,UAAU,sCAAsC,CAAC,WAAmB;IACxE,OAAO,YAAY,CAAC,CAAC,WAAW,EAAE,GAAG,oCAAoC,EAAE,GAAG,6BAA6B,CAAC,CAAC,CAAC;AAChH,CAAC;AAED,MAAM,UAAU,+BAA+B;IAC7C,OAAO,6BAA6B,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAClD,CAAC;AAED,SAAS,YAAY,CAAC,MAAgB;IACpC,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC;AAC9B,CAAC"}
+{"version":3,"file":"model-policy.js","sourceRoot":"","sources":["../../../../src/core/llm/model-policy.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,uBAAuB,EAAE,oBAAoB,EAAE,MAAM,cAAc,CAAC;AAE7E,OAAO,EAAE,0BAA0B,EAAE,6BAA6B,EAAE,MAAM,oBAAoB,CAAC;AAa/F,MAAM,CAAC,MAAM,+BAA+B,GAAG,oBAAoB,CAAC;AACpE,MAAM,CAAC,MAAM,yBAAyB,GAAG,SAAS,CAAC;AACnD,MAAM,CAAC,MAAM,0BAA0B,GAAG,kBAAkB,CAAC;AAC7D,MAAM,4BAA4B,GAAG,eAAe,CAAC;AACrD,MAAM,CAAC,MAAM,6BAA6B,GAAG,iDAAiD,CAAC;AAE/F,MAAM,oCAAoC,GAAG,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC;AAEzE,MAAM,UAAU,wBAAwB,CAAC,MAA4C;IACnF,IAAI,MAAM,EAAE,IAAI,KAAK,OAAO,EAAE,CAAC;QAC7B,OAAO,OAAO,CAAC;IACjB,CAAC;IAED,IAAI,MAAM,EAAE,IAAI,KAAK,SAAS,EAAE,CAAC;QAC/B,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,0BAA0B,CAAC,IAK1C;IACC,IAAI,IAAI,CAAC,QAAQ,KAAK,WAAW,EAAE,CAAC;QAClC,OAAO,IAAI,CAAC,OAAO,KAAK,iBAAiB,CAAC,CAAC,CAAC,0BAA0B,CAAC,CAAC,CAAC,uBAAuB,CAAC;IACnG,CAAC;IAED,IAAI,IAAI,CAAC,QAAQ,KAAK,QAAQ,EAAE,CAAC;QAC/B,IAAI,IAAI,CAAC,cAAc,KAAK,OAAO,EAAE,CAAC;YACpC,OAAO,IAAI,CAAC,WAAW,IAAI,0BAA0B,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,yBAAyB,CAAC;QACzH,CAAC;QAED,OAAO,IAAI,CAAC,OAAO,KAAK,iBAAiB,CAAC,CAAC,CAAC,+BAA+B,CAAC,CAAC,CAAC,oBAAoB,CAAC;IACrG,CAAC;IAED,OAAO,oBAAoB,CAAC;AAC9B,CAAC;AAED,MAAM,UAAU,oCAAoC,CAAC,IAKpD;IACC,IAAI,IAAI,CAAC,QAAQ,KAAK,QAAQ,IAAI,IAAI,CAAC,cAAc,KAAK,OAAO,IAAI,CAAC,0BAA0B,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QAC7G,OAAO;YACL,EAAE,EAAE,KAAK;YACT,KAAK,EAAE,mDAAmD,IAAI,CAAC,KAAK,gBAAgB,+BAA+B,EAAE,uDAAuD,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,GAAG;SAChO,CAAC;IACJ,CAAC;IAED,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;AACtB,CAAC;AAED,MAAM,UAAU,kCAAkC,CAAC,IAKlD;IACC,MAAM,MAAM,GAAG,oCAAoC,CAAC,IAAI,CAAC,CAAC;IAC1D,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAChC,CAAC;AACH,CAAC;AAED,MAAM,UAAU,sCAAsC,CAAC,WAAmB;IACxE,OAAO,YAAY,CAAC,CAAC,WAAW,EAAE,GAAG,oCAAoC,EAAE,GAAG,6BAA6B,CAAC,CAAC,CAAC;AAChH,CAAC;AAED,MAAM,UAAU,+BAA+B,CAAC,IAK/C;IACC,MAAM,aAAa,GAAG,oCAAoC,CAAC;QACzD,KAAK,EAAE,IAAI,CAAC,KAAK;QACjB,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,cAAc,EAAE,IAAI,CAAC,cAAc;KACpC,CAAC,CAAC;IAEH,OAAO;QACL,EAAE,EAAE,IAAI,CAAC,KAAK;QACd,KAAK,EAAE,IAAI,CAAC,KAAK;QACjB,QAAQ,EAAE,CAAC,aAAa,CAAC,EAAE;QAC3B,cAAc,EAAE,aAAa,CAAC,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,qCAAqC,CAAC,IAAI,CAAC;KAC3F,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qCAAqC,CAAC,IAIrD;IACC,IAAI,IAAI,CAAC,QAAQ,KAAK,QAAQ,IAAI,IAAI,CAAC,cAAc,KAAK,OAAO,IAAI,CAAC,0BAA0B,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QAC7G,OAAO,4BAA4B,CAAC;IACtC,CAAC;IAED,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,4BAA4B,CAAC,IAI5C;IACC,MAAM,aAAa,GAAG,oCAAoC,CAAC;QACzD,KAAK,EAAE,IAAI,CAAC,WAAW;QACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,cAAc,EAAE,IAAI,CAAC,cAAc;KACpC,CAAC,CAAC;IACH,IAAI,aAAa,CAAC,EAAE,EAAE,CAAC;QACrB,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,WAAW,EAAE,CAAC;IACrC,CAAC;IAED,MAAM,QAAQ,GAAG,0BAA0B,CAAC;QAC1C,OAAO,EAAE,eAAe;QACxB,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,WAAW,EAAE,IAAI,CAAC,WAAW;QAC7B,cAAc,EAAE,IAAI,CAAC,cAAc;KACpC,CAAC,CAAC;IAEH,IAAI,QAAQ,KAAK,IAAI,CAAC,WAAW,EAAE,CAAC;QAClC,OAAO;YACL,KAAK,EAAE,IAAI,CAAC,WAAW;YACvB,OAAO,EAAE,SAAS,IAAI,CAAC,WAAW,kHAAkH;SACrJ,CAAC;IACJ,CAAC;IAED,OAAO;QACL,KAAK,EAAE,QAAQ;QACf,OAAO,EAAE,SAAS,IAAI,CAAC,WAAW,8DAA8D,QAAQ,oBAAoB;KAC7H,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,+BAA+B;IAC7C,OAAO,6BAA6B,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAClD,CAAC;AAED,SAAS,YAAY,CAAC,MAAgB;IACpC,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC;AAC9B,CAAC"}

package/dist/src/core/prompts/system-prompt.d.ts

@@ -1,6 +1,6 @@
 /**
  * Build the system prompt for the agent.
- * Encourages purposeful tool use and a clear execution workflow.
+ * Keep it high-level and concise unless stronger steering is clearly needed.
  */
 export declare function buildSystemPrompt(toolNames: string[], projectContext?: string): string;
 //# sourceMappingURL=system-prompt.d.ts.map

package/dist/src/core/prompts/system-prompt.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"system-prompt.d.ts","sourceRoot":"","sources":["../../../../src/core/prompts/system-prompt.ts"],"names":[],"mappings":"AAKA;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,EAAE,cAAc,CAAC,EAAE,MAAM,GAAG,MAAM,CAuHtF"}
+{"version":3,"file":"system-prompt.d.ts","sourceRoot":"","sources":["../../../../src/core/prompts/system-prompt.ts"],"names":[],"mappings":"AAKA;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,EAAE,cAAc,CAAC,EAAE,MAAM,GAAG,MAAM,CAsCtF"}