@interf/compiler 0.5.1 → 0.6.1

package/dist/commands/source-config-wizard.js

@@ -8,10 +8,11 @@ import { draftTruthChecks } from "./check-draft.js";
 export const DEFAULT_COMPILED_NAME = "dataset1";
 export const DEFAULT_COMPILED_CHECK_QUESTION_PLACEHOLDER = "A question you can already verify from this dataset";
 export const DEFAULT_COMPILED_CHECK_ANSWER_PLACEHOLDER = "The expected answer in plain English";
-export const DEFAULT_COMPILED_ABOUT_PLACEHOLDER = "Example: forward-demand metrics, board-prep questions, or chart reads from this dataset.";
+export const DEFAULT_COMPILED_ABOUT_PLACEHOLDER = "Example: board-prep questions from these files, chart reads from this report, or tax review from these exports.";
 export const DEFAULT_DATASET_PATH_PLACEHOLDER = "./dataset1";
-export const DEFAULT_COMPILE_RETRY_ATTEMPTS = 1;
+export const DEFAULT_COMPILE_RETRY_ATTEMPTS = 3;
 export const DEFAULT_SELF_IMPROVING_LOOPS = 3;
+const DEFAULT_SELF_IMPROVING_ATTEMPTS = 1;
 function normalizeOptionalPromptText(value) {
     if (typeof value !== "string")
         return undefined;
@@ -75,9 +76,9 @@ export function defaultDatasetPathForPrompt(name, introStyle, cwd = process.cwd(
 }
 export function datasetAboutPromptMessage(introStyle) {
     if (introStyle === "edit") {
-        return "What should this dataset stay accurate about?";
+        return "What task should this dataset stay focused on?";
     }
-    return "What should this dataset be accurate about?";
+    return "What task should this dataset help with?";
 }
 function formatAttemptLabel(count, noun) {
     return `${count} ${noun}${count === 1 ? "" : "s"}`;
@@ -95,10 +96,29 @@ export function describeCompileLoopSelection(options) {
         return "Compile once.";
     }
     if (mode === "retry") {
-        return "Compile retries are enabled for this dataset.";
+        const attempts = options.maxAttempts ?? DEFAULT_COMPILE_RETRY_ATTEMPTS;
+        return `Compile retries are enabled for this dataset (${formatAttemptLabel(attempts, "total attempt")}).`;
     }
     const loops = options.maxLoops ?? DEFAULT_SELF_IMPROVING_LOOPS;
-    return `Self-improving loops are enabled for this dataset (${formatAttemptLabel(loops, "workflow revision")}).`;
+    const attempts = options.maxAttempts ?? DEFAULT_SELF_IMPROVING_ATTEMPTS;
+    return `Self-improving loops are enabled for this dataset (${formatAttemptLabel(attempts, "attempt")} per variation, ${formatAttemptLabel(loops, "workflow revision")}).`;
+}
+async function promptCompileRetryAttempts(options) {
+    const maxAttempts = await p.text({
+        message: options.message,
+        placeholder: String(DEFAULT_COMPILE_RETRY_ATTEMPTS),
+        initialValue: String(options.initialMaxAttempts ?? DEFAULT_COMPILE_RETRY_ATTEMPTS),
+        validate: (value) => {
+            const parsed = Number.parseInt(value.trim(), 10);
+            if (!Number.isInteger(parsed) || parsed < 2 || parsed > 5) {
+                return "Enter a whole number from 2 to 5";
+            }
+            return undefined;
+        },
+    });
+    if (p.isCancel(maxAttempts))
+        return null;
+    return compiledMaxAttempts(Number.parseInt(String(maxAttempts).trim(), 10)) ?? DEFAULT_COMPILE_RETRY_ATTEMPTS;
 }
 export async function promptCheckCases(options) {
     p.log.info(options.heading);
@@ -263,7 +283,7 @@ async function promptCompiledChecks(options) {
         if (options.initialAbout) {
             p.log.info(`About: ${options.initialAbout}`);
         }
-        p.log.info("Truth checks are the questions this dataset should already answer correctly.");
+        p.log.info("Truth checks are the questions an agent should be able to answer from the data behind this task.");
         const mode = await p.select({
             message: "How do you want to create them?",
             options: [
@@ -420,8 +440,8 @@ export async function promptCompileLoopSelection(options) {
         return {};
     }
     const currentMode = resolveCompileLoopMode(options.initialMaxAttempts, options.initialMaxLoops);
-    const currentPromptMode = currentMode === "self-improving" ? "self-improving" : "once";
-    const recommendedMode = options.recommendedMode ?? "self-improving";
+    const currentPromptMode = currentMode;
+    const recommendedMode = options.recommendedMode ?? "once";
     const includeCompileOnce = options.includeCompileOnce !== false;
     const hasSavedPolicy = typeof options.initialMaxAttempts === "number" ||
         typeof options.initialMaxLoops === "number";
@@ -447,18 +467,20 @@ export async function promptCompileLoopSelection(options) {
         },
         retry: {
             value: "retry",
-            label: "Compile once",
-            hint: "Build the compiled dataset once with the selected workflow",
+            label: "Retry same workflow",
+            hint: `Retry compile and compiled-side truth checks up to ${options.initialMaxAttempts ?? DEFAULT_COMPILE_RETRY_ATTEMPTS} total time${(options.initialMaxAttempts ?? DEFAULT_COMPILE_RETRY_ATTEMPTS) === 1 ? "" : "s"} without editing the workflow`,
         },
     };
     const orderedModes = (hasSavedPolicy
         ? [
             currentPromptMode,
             ...(recommendedMode !== currentPromptMode ? [recommendedMode] : []),
+            "retry",
             "once",
         ]
         : [
             recommendedMode,
+            "retry",
             "once",
         ]).filter((mode, index, list) => list.indexOf(mode) === index)
         .filter((mode) => includeCompileOnce || mode !== "once");
@@ -471,6 +493,17 @@ export async function promptCompileLoopSelection(options) {
     if (selectedMode === "once") {
         return {};
     }
+    if (selectedMode === "retry") {
+        const maxAttempts = await promptCompileRetryAttempts({
+            initialMaxAttempts: options.initialMaxAttempts ?? DEFAULT_COMPILE_RETRY_ATTEMPTS,
+            message: "How many total attempts should Interf try before stopping?",
+        });
+        if (maxAttempts === null)
+            return null;
+        return {
+            max_attempts: maxAttempts,
+        };
+    }
     const maxLoops = await promptSelfImprovingLoops({
         initialMaxLoops: options.initialMaxLoops ?? DEFAULT_SELF_IMPROVING_LOOPS,
         message: "How many workflow revisions should Interf try?",
@@ -478,7 +511,7 @@ export async function promptCompileLoopSelection(options) {
     if (maxLoops === null)
         return null;
     return {
-        max_attempts: compiledMaxAttempts(DEFAULT_COMPILE_RETRY_ATTEMPTS) ?? DEFAULT_COMPILE_RETRY_ATTEMPTS,
+        max_attempts: DEFAULT_SELF_IMPROVING_ATTEMPTS,
         max_loops: maxLoops,
     };
 }
@@ -487,7 +520,7 @@ export async function promptSingleCompiledConfig(options = {}) {
     const introStyle = options.introStyle ?? "first";
     if (introStyle === "first") {
         p.log.info("Pick the dataset folder you want to measure.");
-        p.log.info("Then say what this dataset should be accurate about and save a few truth checks you can verify.");
+        p.log.info("Then say what task this dataset should help with and save a few truth checks you can verify.");
     }
     else if (introStyle === "additional") {
         p.log.info("Add another dataset only if this project needs a separate folder or set of truth checks.");
@@ -544,6 +577,9 @@ export async function promptSingleCompiledConfig(options = {}) {
     if (p.isCancel(datasetPath))
         return null;
     const normalizedDatasetPath = String(datasetPath).trim();
+    if (options.selectedWorkflowLabel) {
+        p.log.info(`Workflow: ${options.selectedWorkflowLabel}`);
+    }
     let compileLoopSelection = {
         ...(typeof initial?.max_attempts === "number" ? { max_attempts: initial.max_attempts } : {}),
         ...(typeof initial?.max_loops === "number" ? { max_loops: initial.max_loops } : {}),
@@ -573,11 +609,8 @@ export async function promptSingleCompiledConfig(options = {}) {
             initialMaxLoops: initial?.max_loops,
             hasChecks: checks.length > 0,
             intro: [
-                ...(options.selectedWorkflowLabel
-                    ? [`Workflow: ${options.selectedWorkflowLabel}`]
-                    : []),
                 "Compile mode",
-                "Choose whether this dataset should compile once or keep revising the workflow until it passes the same truth checks.",
+                "Choose whether this dataset should compile once, retry the same workflow, or try self-improving workflow variations on the same truth checks.",
             ],
         });
         if (compileLoopSelection === null)

package/dist/commands/test.d.ts

@@ -1,9 +1,3 @@
 import type { CommandModule } from "yargs";
-import type { SourceDatasetConfig } from "../lib/schema.js";
-export declare function resolveConfiguredDatasetSelection(options: {
-    sourcePath: string;
-    requestedDatasetName?: string | null;
-    hintedDatasetConfig?: SourceDatasetConfig | null;
-}): SourceDatasetConfig | null;
 export declare const testCommand: CommandModule;
 export declare function runTestCommand(argv?: Record<string, unknown>): Promise<boolean>;

package/dist/commands/test.js

@@ -3,19 +3,10 @@ import * as p from "@clack/prompts";
 import { detectInterf, resolveSourceControlPath, } from "../lib/interf.js";
 import { fingerprintTruthChecks, sourceDatasetConfigFromInterfConfig, } from "../lib/source-config.js";
 import { addExecutionProfileOptions, executionProfileFromArgv, } from "../lib/execution-profile.js";
-import { chooseCompiledConfigToBuild, findBuiltCompiledPath, findSavedCompiledConfig, } from "./compiled-flow.js";
+import { chooseCompiledConfigToBuild, findBuiltCompiledPath, } from "./compiled-flow.js";
+import { resolveConfiguredDatasetSelection } from "./dataset-selection.js";
 import { printAgentTestFailures, printAgentTestMatrix, printSavedTestComparisonState, readSavedTestComparison, runSavedRawTest, runSavedCompiledTest, saveTestComparisonRun, } from "./test-flow.js";
 import { listRunAgentOptions, promptForTestAgents, resolveNamedLocalExecutor, resolveOrConfigureLocalExecutor, } from "./executor-flow.js";
-export function resolveConfiguredDatasetSelection(options) {
-    if (options.hintedDatasetConfig &&
-        (!options.requestedDatasetName || options.hintedDatasetConfig.name === options.requestedDatasetName)) {
-        return options.hintedDatasetConfig;
-    }
-    if (!options.requestedDatasetName) {
-        return null;
-    }
-    return findSavedCompiledConfig(options.sourcePath, options.requestedDatasetName);
-}
 export const testCommand = {
     command: "test",
     describe: "Compare files as-is and a compiled dataset on saved truth checks",
@@ -156,17 +147,17 @@ async function resolveSelectedTestAgents(options) {
     ];
 }
 async function runModeForAgent(options) {
-    const rawOutcome = options.mode === "raw" || options.mode === "both"
-        ? await runSavedRawTest({
+    const rawPromise = options.mode === "raw" || options.mode === "both"
+        ? runSavedRawTest({
             sourcePath: options.sourcePath,
             datasetConfig: options.datasetConfig,
             executor: options.executor,
             preserveSandboxes: options.preserveSandboxes,
             runSuffix: options.executor.name,
         })
-        : null;
-    const compiledOutcome = options.mode === "compiled" || options.mode === "both"
-        ? await runSavedCompiledTest({
+        : Promise.resolve(null);
+    const compiledPromise = options.mode === "compiled" || options.mode === "both"
+        ? runSavedCompiledTest({
             sourcePath: options.sourcePath,
             datasetConfig: options.datasetConfig,
             compiledPath: options.builtCompiledPath,
@@ -174,7 +165,8 @@ async function runModeForAgent(options) {
             preserveSandboxes: options.preserveSandboxes,
             runSuffix: options.executor.name,
         })
-        : null;
+        : Promise.resolve(null);
+    const [rawOutcome, compiledOutcome] = await Promise.all([rawPromise, compiledPromise]);
     return {
         agentLabel: options.executor.displayName,
         rawOutcome,

package/dist/index.d.ts

@@ -1,5 +1,5 @@
 export { createCompiled, compileCompiled, runCompiledSummarize, runCompiledCompile, } from "./lib/workflows.js";
-export { createRawTestTarget, listTestSpecs, loadTestSpec, loadTestSpecFromFile, writeTestSpec, listTestTargets, runTargetTests, runTargetTestsWithJudge, runTargetTestsAuto, saveTargetTestRun, } from "./lib/test.js";
+export { createRawTestTarget, createCompiledTestTarget, listTestSpecs, loadTestSpec, loadTestSpecFromFile, writeTestSpec, listTestTargets, runTargetTests, runTargetTestsWithJudge, runTargetTestsAuto, saveTargetTestRun, } from "./lib/test.js";
 export { computeCompiledHealth, } from "./lib/state.js";
 export { SOURCE_FOLDER_CONFIG_FILE, loadSourceFolderConfig, buildTestSpecFromSourceFolderConfig, } from "./lib/source-config.js";
 export { validateCompiledSummarize, validateCompiledCompile, } from "./lib/validate.js";

package/dist/index.js

@@ -1,5 +1,5 @@
 export { createCompiled, compileCompiled, runCompiledSummarize, runCompiledCompile, } from "./lib/workflows.js";
-export { createRawTestTarget, listTestSpecs, loadTestSpec, loadTestSpecFromFile, writeTestSpec, listTestTargets, runTargetTests, runTargetTestsWithJudge, runTargetTestsAuto, saveTargetTestRun, } from "./lib/test.js";
+export { createRawTestTarget, createCompiledTestTarget, listTestSpecs, loadTestSpec, loadTestSpecFromFile, writeTestSpec, listTestTargets, runTargetTests, runTargetTestsWithJudge, runTargetTestsAuto, saveTargetTestRun, } from "./lib/test.js";
 export { computeCompiledHealth, } from "./lib/state.js";
 export { SOURCE_FOLDER_CONFIG_FILE, loadSourceFolderConfig, buildTestSpecFromSourceFolderConfig, } from "./lib/source-config.js";
 export { validateCompiledSummarize, validateCompiledCompile, } from "./lib/validate.js";

package/dist/lib/agent-args.d.ts

@@ -1,3 +1,4 @@
 import type { WorkflowExecutionProfile } from "./executors.js";
 import type { Agent } from "./agent-types.js";
+export declare function buildAgentEnv(agent: Agent, baseEnv?: NodeJS.ProcessEnv): NodeJS.ProcessEnv;
 export declare function buildAgentArgs(agent: Agent, prompt: string, executionProfile?: WorkflowExecutionProfile): string[];

package/dist/lib/agent-args.js

@@ -1,4 +1,14 @@
 const CODEX_SANDBOX_MODE = `work${"space-write"}`;
+export function buildAgentEnv(agent, baseEnv = process.env) {
+    const env = { ...baseEnv };
+    if (agent.name === "codex") {
+        // Codex executes Bash-tool commands inside its own sandbox. Let Codex
+        // choose a sandbox-safe default shell instead of inheriting a host-only
+        // login shell path such as /bin/zsh.
+        delete env.SHELL;
+    }
+    return env;
+}
 export function buildAgentArgs(agent, prompt, executionProfile = {}) {
     if (agent.name === "claude-code") {
         const args = [

package/dist/lib/agent-execution.js

@@ -1,7 +1,7 @@
 import { spawn } from "node:child_process";
 import chalk from "chalk";
 import { CODEX_NOISE_PATTERNS } from "./agent-constants.js";
-import { buildAgentArgs } from "./agent-args.js";
+import { buildAgentArgs, buildAgentEnv } from "./agent-args.js";
 import { appendAgentEventLog, appendAgentStatusLog } from "./agent-logs.js";
 import { displayAgentEvent, emitVisibleAgentText, summarizeAgentToolActivity, } from "./agent-render.js";
 import { classifyTerminalVisibleStatus, extractAgentFailureStatus, hasAgentStalled, } from "./agent-status.js";
@@ -20,6 +20,7 @@ export function spawnAgent(agent, dirPath, prompt, options = {}) {
         appendAgentStatusLog(options.statusLogPath, launchLine);
         const proc = spawn(agent.command, args, {
             cwd: dirPath,
+            env: buildAgentEnv(agent),
             stdio: ["ignore", "pipe", "pipe"],
         });
         const timeoutMs = options.executionProfile?.timeoutMs ?? null;

package/dist/lib/agent-preflight.js

@@ -3,7 +3,7 @@ import { spawnSync } from "node:child_process";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { CODEX_NOISE_PATTERNS } from "./agent-constants.js";
-import { buildAgentArgs } from "./agent-args.js";
+import { buildAgentArgs, buildAgentEnv } from "./agent-args.js";
 const successfulPreflightAgents = new Set();
 export function buildAgentPreflightPrompt() {
     return [
@@ -22,6 +22,7 @@ export function runAgentPreflight(agent, options = {}) {
         const result = spawnSyncImpl(agent.command, buildAgentArgs(agent, buildAgentPreflightPrompt()), {
             cwd: dirPath,
             encoding: "utf8",
+            env: buildAgentEnv(agent),
             timeout: timeoutMs,
         });
         const stdout = typeof result.stdout === "string" ? result.stdout.trim() : "";

package/dist/lib/agent-shells.d.ts

@@ -1,4 +1,4 @@
-import type { RuntimeContractType, WorkflowImprovementContext, WorkflowZoneId } from "./schema.js";
+import type { RuntimeContractType, SourceTruthCheck, WorkflowImprovementContext, WorkflowZoneId } from "./schema.js";
 export interface NativeStageDefinition {
     id: string;
     label: string;
@@ -8,6 +8,7 @@ export interface NativeStageDefinition {
     reads: WorkflowZoneId[];
     writes: WorkflowZoneId[];
 }
+export declare function writeNativeAgentSurface(rootPath: string, agentsContent: string, skillName: string, skillContent: string): boolean;
 export declare function renderCompiledAgents(compiledPath: string, name: string, workflowId: string, about?: string, options?: {
     workflowOriginSelected?: string | null;
     workflowLocalDraft?: boolean;
@@ -15,6 +16,30 @@ export declare function renderCompiledAgents(compiledPath: string, name: string,
 export declare function renderCompiledQuerySkill(): string;
 export declare function syncStageExecutionShellWrites(compiledPath: string, shellRoot: string, stage: NativeStageDefinition, writeArtifacts?: readonly string[]): void;
 export declare function renderClaudeBootstrap(content: string): string;
+export interface WorkflowAuthoringPreviewInfo {
+    compiledPath: string;
+    compileResult: {
+        ok: boolean;
+        failedStage: string | null;
+    };
+}
+export declare function createWorkflowAuthoringShell(options: {
+    workflowPath: string;
+    workflowId: string;
+    label: string;
+    baseWorkflowId: string;
+    datasetPath: string;
+    taskPrompt: string;
+    checks: SourceTruthCheck[];
+    preview?: WorkflowAuthoringPreviewInfo | null;
+}): {
+    rootPath: string;
+    workflowBeforePath: string;
+    workflowAfterPath: string;
+    promptLogPath: string;
+    eventLogPath: string;
+    statusLogPath: string;
+};
 export declare function pruneStageExecutionShells(compiledPath: string): void;
 export declare function projectCompiledQueryShell(compiledPath: string, compiledName: string, workflowId: string, about?: string, options?: {
     workflowOriginSelected?: string | null;