docs: upgrade debate example — 3 roles, oneOf routing, bounded termination

Based on xiaonuo's battle-tested debate workflow. Changes:
- 3 roles (proponent/opponent/host) instead of 2
- oneOf + const schema (speak/conceded/final) instead of enum
- Bounded: 3rd speech forces final statement
- Host role delivers summary/highlights/verdict
- Critical thinking framework in procedure
- Edge prompts pass structured data via {{{argument}}}/{{{closing}}}

.changeset/bootstrap-onboarding.md

@@ -1,9 +0,0 @@
----
-"@united-workforce/cli": patch
----
-
-fix: expand bootstrap prompt with full onboarding and upgrade guide
-
-Bootstrap now covers two scenarios:
-- Fresh install: CLI + adapter installation, `uwf setup` configuration, skill installation, end-to-end verification
-- Upgrade: package update, skill regeneration, breaking change migrations (e.g. $START new/resume)

.changeset/bootstrap-preflight.md

@@ -1,8 +0,0 @@
----
-"@united-workforce/cli": patch
----
-
-fix: bootstrap adds Step 0 environment pre-flight check
-
-- Pre-flight checks for node, pnpm/npm, global bin PATH, hermes CLI with FIX instructions (#112)
-- Install commands changed from npm to pnpm (with npm fallback)

.changeset/bootstrap-schema-path.md

@@ -1,9 +0,0 @@
----
-"@united-workforce/cli": patch
-"@united-workforce/util": patch
----
-
-fix: workflow-authoring flat schema example uses enum, bootstrap adds PATH guidance
-
-- workflow-authoring: flat schema example uses `enum: [done]` instead of bare `const` (#110.3)
-- bootstrap: adds `which hermes` check and PATH guidance for venv installs (#110.4)

.changeset/bootstrap-ux-3.md

@@ -1,14 +0,0 @@
----
-"@united-workforce/cli": patch
----
-
-fix: improve bootstrap docs — agent discovery, pnpm/npm parity, preset provider table (#118, #120)
-
-- Step 1: detect installed agents (hermes/claude) before choosing adapter
-- Step 1: clarify adapter versions are independent from CLI — install @latest
-- Step 1: show pnpm and npm side-by-side
-- Step 1: add "adapter must be installed before `uwf setup --agent`" note
-- Step 1: add ACP verification step (hermes acp --help)
-- Step 2: `--agent` takes adapter command name (e.g. `uwf-hermes`), not npm package
-- Step 2: preset providers listed as a table with names and default base URLs
-- Remove uwf-builtin from supported adapters (not ready yet)

.changeset/bootstrap-ux-fixes.md

@@ -1,10 +0,0 @@
----
-"@united-workforce/cli": patch
----
-
-fix: preset provider base-url auto-fill, bootstrap ACP docs, friendlier name mismatch error
-
-- `uwf setup --provider dashscope` now auto-fills `--base-url` from preset list (#106)
-- Bootstrap guide documents uwf-hermes ACP dependency (`pip install hermes-agent[acp]`) (#107)
-- Bootstrap verify step uses inline workflow instead of missing `examples/eval-simple.yaml` (#107)
-- Workflow filename mismatch error now suggests how to fix it (#108)

.changeset/inject-thread-progress.md

@@ -0,0 +1,12 @@
+---
+"@united-workforce/agent-hermes": patch
+"@united-workforce/agent-claude-code": patch
+"@united-workforce/util-agent": patch
+---
+
+feat: inject thread progress into agent prompt (#127)
+
+Agents now receive a "Thread Progress" section in their prompt showing the
+current step number and how many times the current role has spoken before.
+This eliminates the need for agents to make tool calls (terminal, delegate_task)
+just to count their own turn history.

.changeset/setup-ux-2.md

@@ -1,14 +0,0 @@
----
-"@united-workforce/cli": patch
-"@united-workforce/agent-hermes": patch
-"@united-workforce/agent-claude-code": patch
-"@united-workforce/agent-builtin": patch
-"@united-workforce/agent-mock": patch
----
-
-fix: suppress ExperimentalWarning, PEP 668 pip guidance, setup help (#116)
-
-- All CLI bins use shebang `#!/usr/bin/env -S node --disable-warning=ExperimentalWarning`
-- Remove NODE_OPTIONS injection from spawn (shebang handles it)
-- Bootstrap pip install guidance covers venv/pipx/source options for PEP 668 systems
-- `uwf setup --help` mentions interactive wizard mode

.changeset/setup-ux-improvements.md

@@ -1,12 +0,0 @@
----
-"@united-workforce/cli": patch
----
-
-fix: setup UX improvements (#114)
-
-- Setup validates adapter availability and prints install command if missing
-- Setup prints "Config saved to <path> ✓" on success
-- Spawn ENOENT gives actionable error ("not found in PATH" + which command)
-- SQLite ExperimentalWarning suppressed via NODE_OPTIONS in spawned processes
-- Bootstrap VERSION reads cli package version (was reading util version)
-- Bootstrap PATH guidance is shell-agnostic (no hardcoded .bashrc/.profile)

.changeset/start-new-resume.md

@@ -1,9 +0,0 @@
----
-"@united-workforce/cli": minor
-"@united-workforce/util": patch
----
-
-feat: replace $START `_` status with `new`/`resume` semantics
-
-BREAKING: All workflow YAML files must update `$START._` to `$START.new` + `$START.resume`.
-The `resume` edge prompt replaces the previously hardcoded resume message in the CLI.

.changeset/workflow-authoring-docs.md

@@ -1,15 +0,0 @@
----
-"@united-workforce/cli": patch
-"@united-workforce/util": patch
----
-
-fix: unify $status to const-only, drop enum support (#123)
-
-Breaking: `$status` in frontmatter now requires `const` everywhere.
-`enum` is no longer accepted and will be rejected by the validator.
-
-- Validator: `hasStatusConst()` / `getConstStatuses()` replace enum-based checks
-- Error message: "must define $status as const (or oneOf with const)"
-- workflow-authoring docs: all examples use `const`, enum explicitly noted as unsupported
-- bootstrap hello.yaml: `$status: { const: done }`
-- All test fixtures migrated from enum to const/oneOf

examples/debate.yaml

@@ -1,63 +1,140 @@
-name: "debate"
+name: debate
-description: "Structured debate between two sides. Tests cross-process session resume."
+description: "Structured two-side debate with host summary. Demonstrates multi-role coordination, oneOf output routing, and bounded termination."
+
 roles:
-  against:
+  proponent:
-    description: "Argues against the proposition"
+    description: "Argues FOR the proposition"
-    goal: |
+    goal: "Build a compelling case for the proposition through logical reasoning and evidence"
-      You are a skilled debater arguing AGAINST the proposition.
+    capabilities: []
-      Be logical, cite evidence, and directly address your opponent's points.
-      Keep each argument concise (under 200 words).
-    capabilities:
-      - argumentation
-      - critical-thinking
     procedure: |
-      1. If this is the opening, present your strongest argument against the proposition.
+      You are an experienced scholar arguing FOR the proposition.
-      2. If responding to the other side, directly counter their points with evidence and logic.
+
-      3. If you find yourself genuinely convinced by the other side, you may concede.
+      ## Critical Thinking Framework (apply before every response)
-    output: |
+
-      Provide your argument in the frontmatter.
+      ### A. Pre-response reflection (internal, do not output)
-      Set status to "conceded" ONLY if you are genuinely convinced and wish to stop debating.
+      - Does every step in my argument chain hold? Any hidden assumptions?
-      Otherwise set status to "continue".
+      - If I were my opponent, how would I attack this? Where am I weakest?
+      - Does my evidence actually support my claim, or could it backfire?
+
+      ### B. Evidence discipline
+      - Verify key numbers — watch for order-of-magnitude errors
+      - Assess data freshness — fast-moving fields have short half-lives
+      - Distinguish primary data from secondary citations, expert opinion, and common assumptions
+
+      ### C. Anti-fragility
+      - Anticipate counterarguments; preemptively strengthen or strategically abandon weak points
+      - Catch logical gaps, data misuse, or outdated claims in your opponent's reasoning
+
+      ## Rules
+      1. Check Thread Progress to see how many times you have spoken.
+      2. On your 3rd speech, you MUST output $status: final (closing statement).
+      3. If genuinely convinced by the opponent, output $status: conceded.
+      4. Otherwise output $status: speak and counter the opponent's points.
+      5. Be rigorous, cite evidence, stay concise.
+    output: "Debate argument"
+    frontmatter:
+      type: object
+      oneOf:
+        - properties:
+            $status: { const: speak }
+            argument: { type: string }
+          required: [$status, argument]
+        - properties:
+            $status: { const: conceded }
+            reason: { type: string }
+          required: [$status, reason]
+        - properties:
+            $status: { const: final }
+            closing: { type: string }
+          required: [$status, closing]
+
+  opponent:
+    description: "Argues AGAINST the proposition"
+    goal: "Build a compelling case against the proposition through logical reasoning and evidence"
+    capabilities: []
+    procedure: |
+      You are an experienced scholar arguing AGAINST the proposition.
+
+      ## Critical Thinking Framework (apply before every response)
+
+      ### A. Pre-response reflection (internal, do not output)
+      - Does every step in my argument chain hold? Any hidden assumptions?
+      - If I were my opponent, how would I attack this? Where am I weakest?
+      - Does my evidence actually support my claim, or could it backfire?
+
+      ### B. Evidence discipline
+      - Verify key numbers — watch for order-of-magnitude errors
+      - Assess data freshness — fast-moving fields have short half-lives
+      - Distinguish primary data from secondary citations, expert opinion, and common assumptions
+
+      ### C. Anti-fragility
+      - Anticipate counterarguments; preemptively strengthen or strategically abandon weak points
+      - Catch logical gaps, data misuse, or outdated claims in your opponent's reasoning
+
+      ## Rules
+      1. Check Thread Progress to see how many times you have spoken.
+      2. On your 3rd speech, or when the proponent has issued a final statement, you MUST output $status: final.
+      3. If genuinely convinced by the proponent, output $status: conceded.
+      4. Otherwise output $status: speak and counter the proponent's points.
+      5. Be rigorous, cite evidence, stay concise.
+    output: "Debate argument"
+    frontmatter:
+      type: object
+      oneOf:
+        - properties:
+            $status: { const: speak }
+            argument: { type: string }
+          required: [$status, argument]
+        - properties:
+            $status: { const: conceded }
+            reason: { type: string }
+          required: [$status, reason]
+        - properties:
+            $status: { const: final }
+            closing: { type: string }
+          required: [$status, closing]
+
+  host:
+    description: "Debate moderator — delivers impartial summary and verdict"
+    goal: "Objectively review the debate, analyze both sides, and deliver a verdict"
+    capabilities: []
+    procedure: |
+      You are an experienced academic debate moderator.
+
+      ## Task
+      1. Outline each side's core arguments
+      2. Evaluate reasoning quality and evidence use
+      3. Highlight the most impactful exchanges
+      4. Analyze the deeper significance of the topic
+      5. Deliver an overall verdict
+
+      ## Style
+      - Impartial but with independent judgment
+      - Substantive, not superficial
+    output: "Debate summary report"
     frontmatter:
       type: object
       properties:
-        $status:
+        $status: { const: done }
-          enum: ["continue", "conceded"]
+        summary: { type: string }
-        argument:
+        highlights: { type: string }
-          type: string
+        verdict: { type: string }
-      required: [$status, argument]
+      required: [$status, summary, highlights, verdict]
-  for:
+
-    description: "Argues for the proposition"
-    goal: |
-      You are a skilled debater arguing FOR the proposition.
-      Be logical, cite evidence, and directly address your opponent's points.
-      Keep each argument concise (under 200 words).
-    capabilities:
-      - argumentation
-      - critical-thinking
-    procedure: |
-      1. Read the opposing side's latest argument carefully.
-      2. Counter their points with evidence and logic.
-      3. If you find yourself genuinely convinced by the other side, you may concede.
-    output: |
-      Provide your argument in the frontmatter.
-      Set status to "conceded" ONLY if you are genuinely convinced and wish to stop debating.
-      Otherwise set status to "continue".
-    frontmatter:
-      type: object
-      properties:
-        $status:
-          enum: ["continue", "conceded"]
-        argument:
-          type: string
-      required: [$status, argument]
 graph:
   $START:
-    new: { role: "against", prompt: "Present your opening argument against the proposition." }
+    new: { role: proponent, prompt: "The debate begins. You are arguing FOR the proposition. Present your opening argument." }
-    resume: { role: "against", prompt: "Review the previous debate output and continue the argument against the proposition." }
+    resume: { role: proponent, prompt: "The debate continues." }
-  against:
+
-    conceded: { role: "$END", prompt: "The against side conceded. Debate over." }
+  proponent:
-    continue: { role: "for", prompt: "Counter the opposing argument: {{{argument}}}" }
+    speak: { role: opponent, prompt: "Proponent argues:\n\n{{{argument}}}\n\nYou are the opponent. Counter this argument." }
-  for:
+    conceded: { role: host, prompt: "The proponent conceded: {{{reason}}}\n\nPlease summarize the debate." }
-    conceded: { role: "$END", prompt: "The for side conceded. Debate over." }
+    final: { role: opponent, prompt: "Proponent's closing statement:\n\n{{{closing}}}\n\nYou are the opponent. Deliver your final response." }
-    continue: { role: "against", prompt: "Counter the opposing argument: {{{argument}}}" }
+
+  opponent:
+    speak: { role: proponent, prompt: "Opponent argues:\n\n{{{argument}}}\n\nYou are the proponent. Counter this argument." }
+    conceded: { role: host, prompt: "The opponent conceded: {{{reason}}}\n\nPlease summarize the debate." }
+    final: { role: host, prompt: "Opponent's closing statement:\n\n{{{closing}}}\n\nThe debate is over. Please summarize." }
+
+  host:
+    done: { role: "$END", prompt: "Summary complete." }

packages/agent-claude-code/src/claude-code.ts

@@ -7,6 +7,7 @@ import {
   type AgentRunResult,
   buildContinuationPrompt,
   buildRolePrompt,
+  buildThreadProgress,
   createAgent,
   getCachedSessionId,
   setCachedSessionId,
@@ -27,6 +28,10 @@ export function buildClaudeCodePrompt(ctx: AgentContext): string {
   if (ctx.outputFormatInstruction !== undefined && ctx.outputFormatInstruction !== "") {
     parts.push(ctx.outputFormatInstruction, "");
   }
+
+  // Inject thread progress so the agent knows step count and role visit count
+  parts.push(buildThreadProgress(ctx.steps, ctx.role), "");
+
   parts.push(rolePrompt, "", "## Task", ctx.start.prompt);
 
   if (!ctx.isFirstVisit) {

packages/agent-hermes/src/hermes.ts

@@ -6,6 +6,7 @@ import {
   type AgentRunResult,
   buildContinuationPrompt,
   buildRolePrompt,
+  buildThreadProgress,
   createAgent,
 } from "@united-workforce/util-agent";
 import type { AcpUsage } from "./acp-client.js";
@@ -60,6 +61,9 @@ export function buildHermesPrompt(ctx: AgentContext): string {
     parts.push(ctx.outputFormatInstruction, "");
   }
 
+  // Inject thread progress so the agent knows step count and role visit count
+  parts.push(buildThreadProgress(ctx.steps, ctx.role), "");
+
   if (!ctx.isFirstVisit) {
     // Re-entry: show only steps since last visit, meta only
     parts.push(buildContinuationPrompt(ctx.steps, ctx.role, ctx.edgePrompt));

packages/util-agent/__tests__/build-thread-progress.test.ts

@@ -0,0 +1,59 @@
+import type { StepContext } from "@united-workforce/protocol";
+import { describe, expect, test } from "vitest";
+import { buildThreadProgress } from "../src/build-thread-progress.js";
+
+function makeStep(role: string): StepContext {
+  return {
+    role,
+    output: {},
+    detail: "0000000000000" as string,
+    agent: "uwf-mock",
+    edgePrompt: "",
+    startedAtMs: 0,
+    completedAtMs: 0,
+    cwd: "",
+    assembledPrompt: null,
+    usage: null,
+    content: null,
+  };
+}
+
+describe("buildThreadProgress", () => {
+  test("first step of thread", () => {
+    const result = buildThreadProgress([], "proponent");
+    expect(result).toContain("## Thread Progress");
+    expect(result).toContain("first step");
+    expect(result).toContain("first time");
+    expect(result).toContain("proponent");
+  });
+
+  test("second step, role not seen before", () => {
+    const steps = [makeStep("opponent")];
+    const result = buildThreadProgress(steps, "proponent");
+    expect(result).toContain("Thread step 2");
+    expect(result).toContain("spoken 0 times");
+  });
+
+  test("role has spoken once before", () => {
+    const steps = [makeStep("proponent"), makeStep("opponent")];
+    const result = buildThreadProgress(steps, "proponent");
+    expect(result).toContain("Thread step 3");
+    expect(result).toContain("spoken 1 time before");
+    // singular "time" not "times"
+    expect(result).not.toContain("1 times");
+  });
+
+  test("role has spoken multiple times", () => {
+    const steps = [
+      makeStep("proponent"),
+      makeStep("opponent"),
+      makeStep("proponent"),
+      makeStep("opponent"),
+      makeStep("proponent"),
+      makeStep("opponent"),
+    ];
+    const result = buildThreadProgress(steps, "proponent");
+    expect(result).toContain("Thread step 7");
+    expect(result).toContain("spoken 3 times");
+  });
+});

packages/util-agent/src/build-thread-progress.ts

@@ -0,0 +1,27 @@
+import type { StepContext } from "@united-workforce/protocol";
+
+/**
+ * Build a compact thread-progress summary so the agent knows where it is
+ * in the conversation without making tool calls to count steps.
+ *
+ * Example output:
+ *   ## Thread Progress
+ *   Thread step 6. You (proponent) have spoken 2 times before this turn.
+ */
+export function buildThreadProgress(steps: StepContext[], role: string): string {
+  const totalSteps = steps.length;
+  const roleVisits = steps.filter((s) => s.role === role).length;
+
+  const parts = [`## Thread Progress`];
+  if (totalSteps === 0) {
+    parts.push(
+      `This is the first step of the thread. You (${role}) are speaking for the first time.`,
+    );
+  } else {
+    parts.push(
+      `Thread step ${totalSteps + 1}. You (${role}) have spoken ${roleVisits} time${roleVisits === 1 ? "" : "s"} before this turn.`,
+    );
+  }
+
+  return parts.join("\n");
+}

packages/util-agent/src/index.ts

@@ -1,6 +1,7 @@
 export { buildContinuationPrompt } from "./build-continuation-prompt.js";
 export { buildOutputFormatInstruction } from "./build-output-format-instruction.js";
 export { buildRolePrompt } from "./build-role-prompt.js";
+export { buildThreadProgress } from "./build-thread-progress.js";
 export type { BuildContextMeta } from "./context.js";
 export { buildContext, buildContextWithMeta } from "./context.js";
 export type { ExtractResult, ResolvedLlmProvider } from "./extract.js";

packages/util/src/usage-reference.ts

@@ -140,5 +140,18 @@ For specific scenarios, run the corresponding \`uwf prompt\` command:
 |----------|---------|-------------|
 | Writing workflow YAML | \`uwf prompt workflow-authoring\` | Designing roles, conditions, graphs, and edge prompts |
 | Building a new agent adapter | \`uwf prompt adapter-developing\` | Creating a new \`uwf-<name>\` CLI adapter |
+
+## Upgrading
+
+\`\`\`bash
+# Install the latest version
+pnpm add -g @united-workforce/cli@latest @united-workforce/agent-hermes@latest
+# or: npm install -g @united-workforce/cli@latest @united-workforce/agent-hermes@latest
+
+# Verify
+uwf --version
+
+# Then run uwf prompt bootstrap and follow the upgrade instructions
+\`\`\`
 `;
 }