@topce/pizx 0.5.0 → 0.6.0

package/README.md

@@ -4,7 +4,7 @@
 
 ![pizx — zx fork with native Pi AI integration](github-social-banner.png)
 
-> **zx fork with native Pi AI integration** — 15 template tags for shell scripting, AI text generation, coding agents, agentic patterns, communication, and orchestration topologies.
+> **zx fork with native Pi AI integration** — 15 agent pattern tags (plus shell and AI core) for shell scripting, AI text generation, coding agents, agentic patterns, multi-agent communication, and orchestration topologies.
 
 ## Quick Start
 
@@ -87,6 +87,27 @@ await fleet`review all files in src/`
 > `pi` (alias for `π`), `Pi` (alias for `Π`), `fleet` (alias for `Φ`), `ralph` (alias for `Ρ`),
 > `pipeline` (alias for `Λ`), etc. — use whichever style you prefer. See [full mapping](#english-aliases) below.
 
+### Global Access (pizx/globals)
+
+Import the `pizx/globals` module to make **all tags and English aliases** available without explicit imports — matching the `#!/usr/bin/env pizx` shebang experience inside scripts loaded via `import()`:
+
+```js
+import '@topce/pizx/globals'
+
+// All Greek tags are available without imports:
+const answer = await π`explain async/await`
+await Π`fix the lint issues`
+await Φ`review all files`
+
+// English aliases too:
+const docs = await fleet`check all .ts files`
+const plan = await orchestrator`design the architecture`
+
+// Helpers:
+configurePi({ model: 'anthropic/claude-sonnet-4-5' })
+closeAgent()
+```
+
 ### CLI Quick Queries
 
 ```bash
@@ -193,7 +214,10 @@ await Ω({ system: 'You are a senior security architect.' })`design an auth syst
 
 ### Quality Validation
 
-All 15 patterns support an optional `qualityCheck` flag. When enabled, the pattern runs a post-execution LLM review that scores the final output (0.0–1.0), provides an assessment, and recommends improvements:
+12 patterns support an optional `qualityCheck` flag. When enabled, the pattern runs a post-execution LLM review that scores the final output (0.0–1.0), provides an assessment, and recommends improvements:
+
+Supported by: `Ω` (Orchestrator), `Φ` (Fleet), `Σ` (Subagents), `Δ` (Debate), `Λ` (Pipeline), `Θ` (Thread), `Μ` (Memory), `Β` (Broadcast), `Γ` (Graph), `Ν` (Nu/Team), `Χ` (Chi/Learn), `Τ` (Tau/Store).
+Not applicable to: `Ρ` (Ralph Loop — has its own review phase), `Α` (Adaptive), `Ψ` (Critique).
 
 ```js
 const result = await Ω({ qualityCheck: true })`design the system architecture`
@@ -219,7 +243,32 @@ await Ω({ confirm: true })`design the system`
 // → "Proceed? [Y/n] "
 ```
 
-Supported by: `Ω`, `Σ`, `Φ`, `Λ` (more patterns coming).
+Supported by: `Ω`, `Σ`, `Φ`, `Λ`.
+
+### Agent Mode (File Tools for Any Pattern)
+
+By default, all patterns (except `Pi` and `ralph`) use **text generation** — they can read files only if you pass content in via template interpolation. `ralph` already uses coding agent tools when `useTools: true` (default).
+
+Set `mode: 'agent'` to give every subtask the same **coding agent tools** as `Pi`:
+
+```js
+// Fleet workers can read files
+await fleet({ mode: 'agent' })`read package.json and analyze the project`
+
+// Pipeline stages can edit code
+await pipeline({ mode: 'agent' })`read src/ and refactor the error handling`
+
+// Orchestrator workers can run commands
+await orchestrator({ mode: 'agent' })`check the test coverage and report gaps`
+
+// Debate perspectives can research the codebase
+await debate({ mode: 'agent' })`read the architecture docs and debate the design`
+```
+
+Available tools: `read`, `bash`, `edit`, `write`, `grep`, `ls`.
+
+Supported by: all patterns (`fleet`, `orchestrator`, `pipeline`, `debate`, `subagent`, `critique`, `thread`, `memory`, `broadcast`, `adaptive`, `graph`, `team`, `learn`, `store`).
+Not applicable to: `pi`/`π` (always text), `Pi`/`Π` and `ralph` (already use coding agent).
 
 ### Option Chaining & Quiet Mode
 
@@ -234,7 +283,7 @@ await Θ({ agents: 4, turns: 3 })`debate the architecture`
 await Γ({ graph: { nodes: [...], edges: [...] } })`execute workflow`
 ```
 
-### Timeout & Retry
+### Timeout, Retry & API Key
 
 All tags accept `timeoutMs` and `maxRetries` to control LLM call resilience. When unset, the provider SDK defaults apply (typically 10 min timeout, 2 retries).
 
@@ -249,6 +298,23 @@ await π({ timeoutMs: 15000 })`summarize this document`
 configurePi({ timeoutMs: 60000, maxRetries: 3 })
 ```
 
+Use `apiKey` to specify a provider API key directly, bypassing environment variable lookup:
+
+```js
+await π({ apiKey: 'sk-...' })`analyze this data`
+await Ω({ apiKey: 'sk-...' })`design the system`
+```
+
+### Streaming (π.stream)
+
+For real-time streaming, use `π.stream` as an async generator:
+
+```js
+for await (const chunk of π.stream`tell me a long story`) {
+  process.stdout.write(chunk)
+}
+```
+
 ### Token, Cost & Phase Tracking
 
 Every pattern output and π call includes an execution trace with token usage, cost, and a structured phase log. All collected automatically — no extra flags needed.
@@ -326,6 +392,9 @@ All tags accept `system` (replaces default) and `appendSystemPrompt` (appended a
 // π: custom system prompt
 await π({ system: 'You are a security auditor' })`review this code`
 
+// π: with appendSystemPrompt
+await π({ appendSystemPrompt: 'Respond in JSON format' })`list all .ts files`
+
 // Π: set system prompt and append extra instructions
 await Π({ system: 'You are a test engineer', appendSystemPrompt: 'Write tests first' })`add tests for auth`
 
@@ -333,6 +402,15 @@ await Π({ system: 'You are a test engineer', appendSystemPrompt: 'Write tests f
 await Ω({ system: 'Prioritize security over performance' })`design login flow`
 ```
 
+### closeAgent()
+
+Dispose the shared Pi coding agent session (`Π`). Useful in long-running scripts or tests to reset state:
+
+```js
+import { closeAgent } from '@topce/pizx'
+await closeAgent()  // dispose shared agent session
+```
+
 ### Thinking Budgets
 
 Fine-grained token budgets per reasoning level. Passes through to providers via `thinkingBudgets`.
@@ -381,15 +459,15 @@ pizx --help                   # Print help
 **Options:**
 - `-p, --prompt <text>` — Run a quick pi-ai query (no script needed)
 - `-m, --model <id>` — Specify AI model to use
-- `--quiet` — Suppress output except errors
-- `--shell <path>` — Shell to use (default: auto-detect)
+- `-q, --quiet` — Suppress status output
+- `--system <text>` — System context for pi-ai (print mode only)
 
 ## Commands
 
 ```bash
 npm run build                  # Build (JS + DTS)
 npm run check                  # Lint and format with Biome
-npm test                       # 223 unit tests (no network)
+npm test                       # 311 unit tests (no network)
 npm run test:integration       # Integration tests (requires Pi credentials)
 npm run test:quality           # Run qualityCheck example
 npm run test:confirm           # Run confirm gate example
@@ -409,10 +487,40 @@ See [`examples/`](examples/) for runnable examples of every pattern and feature:
 - [`hello-pizx.mjs`](examples/hello-pizx.mjs) — Basic script with shell + AI
 - [`basic-pi.mjs`](examples/basic-pi.mjs) — π text generation
 - [`basic-capital-pi.mjs`](examples/basic-capital-pi.mjs) — Π coding agent
-- [`pattern-ralph.mjs`](examples/pattern-ralph.mjs) — Ralph Loop
+- [`quick-ask.mjs`](examples/quick-ask.mjs) — Quick π query
+- [`ralph-loop.mjs`](examples/ralph-loop.mjs) — Ralph Loop (detailed)
+- [`pattern-ralph.mjs`](examples/pattern-ralph.mjs) — Ralph Loop (concise)
 - [`pattern-fleet.mjs`](examples/pattern-fleet.mjs) — Fleet parallel execution
+- [`pattern-subagent.mjs`](examples/pattern-subagent.mjs) — Subagents delegation
 - [`pattern-debate.mjs`](examples/pattern-debate.mjs) — Multi-perspective debate
-- ... and more for every pattern
+- [`pattern-orchestrator.mjs`](examples/pattern-orchestrator.mjs) — Orchestrator
+- [`pattern-pipeline.mjs`](examples/pattern-pipeline.mjs) — Pipeline chain
+- [`pattern-critique.mjs`](examples/pattern-critique.mjs) — Critique loop
+- [`pattern-thread.mjs`](examples/pattern-thread.mjs) — Thread conversation
+- [`pattern-memory.mjs`](examples/pattern-memory.mjs) — Memory blackboard
+- [`pattern-broadcast.mjs`](examples/pattern-broadcast.mjs) — Broadcast messaging
+- [`pattern-adaptive.mjs`](examples/pattern-adaptive.mjs) — Adaptive workflow
+- [`pattern-graph.mjs`](examples/pattern-graph.mjs) — DAG execution
+- [`pattern-nu.mjs`](examples/pattern-nu.mjs) — Self-organizing teams
+- [`pattern-chi.mjs`](examples/pattern-chi.mjs) — Cross-agent learning
+- [`pattern-tau.mjs`](examples/pattern-tau.mjs) — Tool-mediated store
+- [`pattern-five-whys.mjs`](examples/pattern-five-whys.mjs) — Five Whys analysis
+- [`pattern-tracking.mjs`](examples/pattern-tracking.mjs) — Token/cost tracking
+- [`pattern-quality.mjs`](examples/pattern-quality.mjs) — Quality check demo
+- [`pattern-timeout-retry.mjs`](examples/pattern-timeout-retry.mjs) — Timeout & retry demo
+- [`pattern-system-propagation.mjs`](examples/pattern-system-propagation.mjs) — System prompt propagation
+
+### English Aliases Examples
+
+See [`english-examples/`](english-examples/) for runnable examples using all English aliases:
+
+- [`english-hello.mjs`](english-examples/english-hello.mjs) — Hello world with English aliases
+- [`english-fleet.mjs`](english-examples/english-fleet.mjs) — Fleet via English aliases
+- [`english-debate.mjs`](english-examples/english-debate.mjs) — Debate via English aliases
+- [`english-orchestrator.mjs`](english-examples/english-orchestrator.mjs) — Orchestrator via English aliases
+- [`english-pipeline.mjs`](english-examples/english-pipeline.mjs) — Pipeline via English aliases
+- [`english-all-patterns.mjs`](english-examples/english-all-patterns.mjs) — All patterns via English aliases
+- [`english-import-verify.mjs`](english-examples/english-import-verify.mjs) — Verify all imports
 
 ### New Feature Demos
 

package/dist/cli.js

@@ -101,6 +101,7 @@ function loadPiAuth() {
 // src/patterns/types.ts
 import { createInterface } from "node:readline";
 import { completeSimple } from "@earendil-works/pi-ai";
+import { createAgentSession } from "@earendil-works/pi-coding-agent";
 
 // src/model-picker.ts
 import {
@@ -393,6 +394,40 @@ ${skillContext}` : skillContext;
   }
   return text.trim();
 }
+async function executeTask(prompt, opts = {}) {
+  if (opts.mode === "agent") {
+    return runAgentTask(prompt, opts);
+  }
+  return ask(prompt, opts);
+}
+async function runAgentTask(prompt, opts) {
+  const model = pickModel(opts.model);
+  if (!model) throw new Error("pizx/patterns: No AI models configured. Run `pi auth login` first.");
+  const tools = ["read", "bash", "edit", "write", "grep", "ls"];
+  const { session } = await createAgentSession({
+    tools,
+    ...model ? { model } : {}
+  });
+  try {
+    await session.sendUserMessage(prompt);
+    const messages = session.messages;
+    for (let i = messages.length - 1; i >= 0; i--) {
+      const msg = messages[i];
+      if (msg?.role !== "assistant") continue;
+      const c = "content" in msg ? msg.content : void 0;
+      if (typeof c === "string") return c.trim();
+      if (Array.isArray(c)) {
+        const texts = c.filter(
+          (block) => typeof block === "object" && block !== null && "type" in block && "text" in block
+        ).map((block) => block.text);
+        if (texts.length > 0) return texts.join("\n").trim();
+      }
+    }
+    return "";
+  } finally {
+    session.dispose();
+  }
+}
 var QUALITY_REVIEW_SYSTEM = `You are a quality assurance reviewer. Evaluate the final deliverable against the original request.
 
 Output format:
@@ -496,7 +531,7 @@ async function execute(pieces, args, opts) {
 `);
   }
   if (!opts.quiet) process.stderr.write("  \u2192 Planning...\n");
-  const planText = await ask(goal, {
+  const planText = await executeTask(goal, {
     ...opts,
     model: plannerModel,
     thinkingLevel: "high",
@@ -529,12 +564,12 @@ async function execute(pieces, args, opts) {
     if (!opts.quiet)
       process.stderr.write(`  \u2192 Step ${executionStep}: ${currentStep.slice(0, 60)}...
 `);
-    const result = await ask(currentStep, {
+    const result = await executeTask(currentStep, {
       ...opts,
       model: workerModel,
       system: mergeSystem(opts.system, EXECUTE_SYSTEM)
     });
-    const evaluation = await ask(
+    const evaluation = await executeTask(
       `Goal: ${goal}
 Step executed: ${currentStep}
 Result: ${result}
@@ -738,7 +773,7 @@ async function execute2(pieces, args, opts) {
   const broadcastResults = await Promise.allSettled(
     roles.map(async (role) => {
       const prompt = WORKER_PROMPT.replace("{role}", role).replace("{question}", question);
-      const text = await ask(prompt, { ...opts, model: workerModel });
+      const text = await executeTask(prompt, { ...opts, model: workerModel });
       return new BroadcastResponse(role, text, true);
     })
   );
@@ -754,7 +789,7 @@ async function execute2(pieces, args, opts) {
   if (!opts.quiet) process.stderr.write("  \u2192 Synthesizing responses...\n");
   const responsesText = responses.map((wr) => `--- ${wr.role} ---
 ${wr.response}`).join("\n\n");
-  const synthesis = await ask(
+  const synthesis = await executeTask(
     `Original question:
 ${question}
 
@@ -875,7 +910,7 @@ async function execute3(pieces, args, opts) {
     process.stderr.write(`\u03A7: Cross-Agent Learning \u2014 analyzing ${label}
 `);
   }
-  const response = await ask(input, {
+  const response = await executeTask(input, {
     ...opts,
     model: plannerModel,
     system: mergeSystem(opts.system, ANALYSIS_SYSTEM)
@@ -951,12 +986,16 @@ async function execute4(pieces, args, opts) {
   for (let r = 0; r < rounds; r++) {
     if (r === 0) {
       if (!opts.quiet) process.stderr.write("  \u2192 Generating initial content...\n");
-      currentContent = await ask(prompt, { ...opts, model: workerModel, system: opts.system });
+      currentContent = await executeTask(prompt, {
+        ...opts,
+        model: workerModel,
+        system: opts.system
+      });
     } else {
       if (!opts.quiet) process.stderr.write(`  \u2192 Improving (round ${r + 1})...
 `);
       const prevCritique = critiqueRounds[r - 1]?.critique ?? "";
-      currentContent = await ask(
+      currentContent = await executeTask(
         `Original request: ${prompt}
 
 Critique:
@@ -971,7 +1010,7 @@ Revise the content based on the critique.`,
     }
     if (!opts.quiet) process.stderr.write(`  \u2192 Critiquing (round ${r + 1})...
 `);
-    const critique2 = await ask(currentContent, {
+    const critique2 = await executeTask(currentContent, {
       ...opts,
       model: plannerModel,
       system: mergeSystem(opts.system, CRITIQUE_SYSTEM)
@@ -1102,7 +1141,7 @@ Refine your position. Address the counter-arguments directly. Strengthen your ar
 `;
   }
   if (!opts.quiet) process.stderr.write("  \u2192 Synthesizing perspectives...\n");
-  const conclusion = await ask(
+  const conclusion = await executeTask(
     `${debateHistory}
 
 Synthesize a balanced conclusion from the full debate above. Weigh the evidence from all rounds.`,
@@ -1178,7 +1217,7 @@ function describeTask(task) {
   return task;
 }
 var FLEET_SYSTEM = `You are a focused task specialist. Complete the assigned task concisely and accurately. Output only the result \u2014 no commentary about being an AI.`;
-async function executeTask(task, opts, workerModel) {
+async function executeFleetTask(task, opts, workerModel) {
   if (typeof task === "function") {
     try {
       const text = await task("");
@@ -1189,7 +1228,7 @@ async function executeTask(task, opts, workerModel) {
   }
   const model = workerModel ?? opts.model;
   try {
-    const text = await ask(task, {
+    const text = await executeTask(task, {
       ...opts,
       model,
       system: mergeSystem(opts.system, FLEET_SYSTEM)
@@ -1223,7 +1262,7 @@ async function execute6(pieces, args, opts) {
   for (let i = 0; i < tasks.length; i += concurrency) {
     const batch = tasks.slice(i, i + concurrency);
     const batchResults = await Promise.allSettled(
-      batch.map((task) => executeTask(task, opts, workerModel))
+      batch.map((task) => executeFleetTask(task, opts, workerModel))
     );
     batchResults.forEach((r, idx) => {
       if (r.status === "fulfilled") {
@@ -1370,7 +1409,7 @@ ${depResults}
 Your task: ${node.task}`;
           }
         }
-        const text = await ask(context, {
+        const text = await executeTask(context, {
           ...opts,
           model: workerModel,
           system: mergeSystem(opts.system, NODE_SYSTEM)
@@ -1465,7 +1504,7 @@ async function execute8(pieces, args, opts) {
     const roundResults = await Promise.allSettled(
       roles.map(async (role) => {
         const prompt = buildWriterPrompt(role, topic, blackboard);
-        const text = await ask(prompt, { ...opts, model: workerModel });
+        const text = await executeTask(prompt, { ...opts, model: workerModel });
         return { role, text };
       })
     );
@@ -1479,7 +1518,7 @@ async function execute8(pieces, args, opts) {
     }
   }
   if (!opts.quiet) process.stderr.write("  \u2192 Consolidating findings...\n");
-  const synthesis = await ask(
+  const synthesis = await executeTask(
     `Topic: ${topic}
 
 Blackboard findings:
@@ -1570,7 +1609,7 @@ async function negotiateRoles(task, opts) {
   const min = opts.minAgents ?? 2;
   const max = opts.maxAgents ?? 5;
   const prompt = NEGOTIATE_SYSTEM.replace("{min}", String(min)).replace("{max}", String(max));
-  const response = await ask(`Task: ${task}
+  const response = await executeTask(`Task: ${task}
 
 ${prompt}`, {
     ...opts,
@@ -1602,7 +1641,7 @@ async function decideWorkflow(roles, task, opts) {
   if (roles.length <= 1)
     return { workflow: "parallel", reasoning: "Single role \u2014 no dependencies." };
   const rolesText = roles.map((r, i) => `${i + 1}. ${r.name}: ${r.goal}`).join("\n");
-  const response = await ask(
+  const response = await executeTask(
     `Task: ${task}
 
 Roles:
@@ -1632,7 +1671,7 @@ async function executeRoles(roles, task, workflow, opts) {
   if (workflow === "sequential") {
     let context = task;
     for (const role of roles) {
-      const output = await ask(context, {
+      const output = await executeTask(context, {
         ...opts,
         model: workerModel,
         system: mergeSystem(opts.system, EXECUTE_SYSTEM2(role))
@@ -1800,7 +1839,7 @@ async function execute10(pieces, args, opts) {
   }
   if (!opts.quiet) process.stderr.write("  \u2192 Planning...\n");
   const planStart = Date.now();
-  const planText = await ask(request, {
+  const planText = await executeTask(request, {
     ...opts,
     model: plannerModel,
     thinkingLevel: "high",
@@ -1865,7 +1904,7 @@ async function execute10(pieces, args, opts) {
   const workerTexts = workerResults.map((wr, i) => `Task ${i + 1}: ${wr.task}
 Result: ${wr.output}`).join("\n\n");
   const synthStart = Date.now();
-  const synthesis = await ask(
+  const synthesis = await executeTask(
     `Original request:
 ${request}
 
@@ -2016,7 +2055,7 @@ async function execute11(pieces, args, opts) {
     } else {
       const prompt = customPrompt ?? generateStagePrompt(stage, currentInput, i === 0);
       const systemMessage = i === 0 ? `You are a specialist executing stage ${i + 1}: ${stage}. Focus only on this stage's output.` : `You are a specialist executing stage ${i + 1}: ${stage}. Process the previous stage's output according to your instructions. Maintain all important information from previous stages.`;
-      output = await ask(prompt, {
+      output = await executeTask(prompt, {
         ...opts,
         model: workerModel,
         system: mergeSystem(opts.system, systemMessage)
@@ -2038,7 +2077,7 @@ ${sr.output.slice(0, 200)}${sr.output.length > 200 ? "..." : ""}`
 var \u039B = createPatternTag(defaults11, execute11);
 
 // src/patterns/ralph.ts
-import { createAgentSession } from "@earendil-works/pi-coding-agent";
+import { createAgentSession as createAgentSession2 } from "@earendil-works/pi-coding-agent";
 var defaults12 = {
   maxIterations: 5,
   useTools: true,
@@ -2064,7 +2103,7 @@ async function executeWithTools(goal, opts) {
       `pizx/\u03A1: model not found: "${opts.model}". Run \`pi models\` to see available models.`
     );
   }
-  const { session } = await createAgentSession({
+  const { session } = await createAgentSession2({
     tools: ["read", "bash", "edit", "write", "grep", "ls"],
     ...agentModel ? { model: agentModel } : {}
   });
@@ -2114,13 +2153,13 @@ async function execute12(pieces, args, opts) {
 `);
     }
     if (!opts.quiet) process.stderr.write("  \u2192 Analyzing...\n");
-    const analysis = await ask(currentGoal, {
+    const analysis = await executeTask(currentGoal, {
       ...opts,
       model: plannerModel,
       system: mergeSystem(opts.system, ANALYSIS_SYSTEM2)
     });
     if (!opts.quiet) process.stderr.write("  \u2192 Planning...\n");
-    const plan = await ask(
+    const plan = await executeTask(
       `Goal: ${currentGoal}
 
 Analysis: ${analysis}
@@ -2135,7 +2174,7 @@ ${plan}
 Goal: ${currentGoal}`, {
       ...opts,
       model: workerModel
-    }) : await ask(`Implement this plan:
+    }) : await executeTask(`Implement this plan:
 ${plan}
 
 Goal: ${currentGoal}`, {
@@ -2143,19 +2182,22 @@ Goal: ${currentGoal}`, {
       model: workerModel
     });
     if (!opts.quiet) process.stderr.write("  \u2192 Reviewing...\n");
-    const review = await ask(`Plan:
+    const review = await executeTask(
+      `Plan:
 ${plan}
 
 Result:
 ${result}
 
-Review the implementation.`, {
-      ...opts,
-      model: plannerModel,
-      maxTokens: 1024,
-      thinkingLevel: "high",
-      system: mergeSystem(opts.system, REVIEW_SYSTEM)
-    });
+Review the implementation.`,
+      {
+        ...opts,
+        model: plannerModel,
+        maxTokens: 1024,
+        thinkingLevel: "high",
+        system: mergeSystem(opts.system, REVIEW_SYSTEM)
+      }
+    );
     const shouldContinue = review.includes("ITERATE") && !review.includes("DONE");
     iterations.push({
       iteration,
@@ -2224,7 +2266,7 @@ var DECOMPOSE_SYSTEM = `You are a task decomposition specialist. Break down comp
 var SYNTHESIS_SYSTEM5 = `You are a synthesis specialist. Combine the results from multiple sub-agent analyses into a coherent, comprehensive answer. Identify patterns, conflicts, and gaps.`;
 async function decomposeTask(task, opts) {
   if (opts.subdomains && opts.subdomains.length > 0) return opts.subdomains;
-  const result = await ask(
+  const result = await executeTask(
     `Decompose this task into ${opts.maxSubTasks ?? 4} independent sub-tasks that can be worked on in parallel:
 
 ${task}
@@ -2314,7 +2356,7 @@ async function execute13(pieces, args, opts) {
   const subResultsText = subResults.map((sr, i) => `Sub-task ${i + 1}: ${sr.subTask}
 Result: ${sr.text}`).join("\n\n");
   const synthStart = Date.now();
-  const synthesis = await ask(
+  const synthesis = await executeTask(
     `Original task:
 ${task}
 
@@ -2432,7 +2474,7 @@ var CONSOLIDATE_SYSTEM = `You are a research director. Consolidate the structure
 async function defineSchema(task, opts) {
   const agentCount = opts.agents ?? 3;
   const prompt = SCHEMA_SYSTEM.replace("{agentCount}", String(agentCount));
-  const response = await ask(`Task: ${task}
+  const response = await executeTask(`Task: ${task}
 
 ${prompt}`, {
     ...opts,
@@ -2487,7 +2529,7 @@ async function executeRound(roles, assignments, store2, round, opts) {
       const storeText = formatStore(store2);
       const systemPrompt = isWrite ? WRITE_SYSTEM(role, keysStr).replace("{store}", storeText) : UPDATE_SYSTEM(role, keysStr).replace("{store}", storeText);
       const task = isWrite ? `Write your initial findings to your assigned keys: ${keysStr}` : `Review the shared context and update your entries for keys: ${keysStr}`;
-      const response = await ask(task, {
+      const response = await executeTask(task, {
         ...opts,
         model: workerModel,
         system: mergeSystem(opts.system, systemPrompt)
@@ -2658,7 +2700,7 @@ async function execute15(pieces, args, opts) {
     for (let a = 0; a < roles.length; a++) {
       const role = roles[a] ?? `Agent ${a + 1}`;
       const prompt = buildThreadPrompt(role, thread2);
-      const response = await ask(prompt, { ...opts, model: workerModel });
+      const response = await executeTask(prompt, { ...opts, model: workerModel });
       messages.push(new ThreadMessage(role, turn, response));
       thread2 += `
 [${role}] (Turn ${turn}): ${response}
@@ -2666,7 +2708,7 @@ async function execute15(pieces, args, opts) {
     }
   }
   if (!opts.quiet) process.stderr.write("  \u2192 Synthesizing conclusion...\n");
-  const conclusion = await ask(
+  const conclusion = await executeTask(
     `Topic: ${topic}
 
 Conversation thread:
@@ -2881,7 +2923,7 @@ var \u03C0 = makePi();
 
 // src/pi-agent.ts
 import {
-  createAgentSession as createAgentSession2,
+  createAgentSession as createAgentSession3,
   DefaultResourceLoader
 } from "@earendil-works/pi-coding-agent";
 var _agentDefaults = {
@@ -2946,7 +2988,7 @@ async function getSession(opts) {
         loader.extendResources({ skillPaths });
       }
     }
-    const result = await createAgentSession2({
+    const result = await createAgentSession3({
       cwd: opts.cwd,
       thinkingLevel: opts.thinkingLevel,
       tools: opts.tools,