@lythos/skill-arena 0.14.3 → 0.14.4

package/README.md

@@ -16,7 +16,7 @@
 ```bash
 bun add -d @lythos/skill-arena
 # or use directly
-bunx @lythos/skill-arena@0.14.3 <command>
+bunx @lythos/skill-arena@0.14.4 <command>
 ```
 
 ## Quick Start

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lythos/skill-arena",
-  "version": "0.14.3",
+  "version": "0.14.4",
   "description": "Skill Arena — benchmark skill effectiveness with controlled-variable comparison",
   "keywords": [
     "ai-agent",
@@ -42,13 +42,15 @@
     "bun": ">=1.0.0"
   },
   "dependencies": {
-    "@lythos/cold-pool": "^0.14.3",
-    "@lythos/infra": "^0.14.3",
-    "@lythos/test-utils": "^0.14.3",
+    "@lythos/cold-pool": "^0.14.4",
+    "@lythos/infra": "^0.14.4",
+    "@lythos/test-utils": "^0.14.4",
     "zod": "^3.24.0",
     "zod-to-json-schema": "^3.25.2"
   },
   "optionalDependencies": {
-    "@lythos/agent-adapter-claude-sdk": "^0.14.3"
+    "@lythos/agent-adapter-claude-sdk": "^0.14.4",
+    "@lythos/agent-adapter-deepseek-serve": "^0.14.4",
+    "@lythos/agent-adapter-codex": "^0.14.4"
   }
 }

package/src/arena-toml.ts

@@ -66,6 +66,7 @@ export interface ExecutionPlan {
   total_runs: number
 }
 
+/** Plan builder — pure data transform. Execution: arena/runner.ts spawns subagents. */
 export function buildExecutionPlan(toml: ArenaToml): ExecutionPlan {
   const cells: ExecutionCell[] = []
   for (const side of toml.side) {

package/src/comparative-judge.ts

@@ -53,6 +53,7 @@ export function computePareto(vectors: { participant_id: string; scores: Record<
 
 // ── Comparative Judge Prompt ──────────────────────────────────────────────
 
+/** Pure prompt builder — no IO. Execution: arena vs mode subagent spawn (cli.ts vsRun). */
 export function buildComparativePrompt(opts: {
   manifest: ArenaManifest
   verdicts: { participantId: string; verdict: unknown }[]

package/src/player.test.ts

@@ -21,11 +21,11 @@ deck = "./decks/rich.toml"
 
 describe('resolvePlayer', () => {
   test('maps claude-code → claude', () => {
-    expect(resolvePlayer('claude-code')).toBe('claude')
+    expect(resolvePlayer('claude-code')).toBe('claude-sdk')
   })
 
   test('maps Claude → claude (case insensitive)', () => {
-    expect(resolvePlayer('Claude')).toBe('claude')
+    expect(resolvePlayer('Claude')).toBe('claude-sdk')
   })
 
   test('maps kimi → kimi', () => {
@@ -37,7 +37,7 @@ describe('resolvePlayer', () => {
   })
 
   test('trims whitespace', () => {
-    expect(resolvePlayer('  claude-code  ')).toBe('claude')
+    expect(resolvePlayer('  claude-code  ')).toBe('claude-sdk')
   })
 })
 
@@ -45,7 +45,7 @@ describe('resolveSides', () => {
   test('resolves all sides in arena.toml', () => {
     const sides = resolveSides(toml)
     expect(sides).toHaveLength(2)
-    expect(sides[0].platform).toBe('claude')
+    expect(sides[0].platform).toBe('claude-sdk')
     expect(sides[1].platform).toBe('expert-architect')
     expect(sides[0].playerName).toBe('claude-code')
   })
@@ -63,7 +63,7 @@ describe('groupBySide', () => {
     expect(groups).toHaveLength(2)
     expect(groups[0].runs).toBe(3) // runs_per_side
     expect(groups[1].runs).toBe(3)
-    expect(groups[0].platform).toBe('claude')
+    expect(groups[0].platform).toBe('claude-sdk')
   })
 
   test('control flag preserved', () => {

package/src/player.ts

@@ -12,10 +12,12 @@ export interface ResolvedSide {
 
 /** Built-in player registry. Player names that map directly to useAgent platforms. */
 const BUILTIN_PLAYERS: Record<string, string> = {
-  'claude': 'claude',
-  'claude-code': 'claude',
+  'claude': 'claude-sdk',           // SDK mode (claude-cli is deprecated — ADR-20260518145235543)
+  'claude-code': 'claude-sdk',
+  'claude-cli': 'claude',           // legacy CLI spawn — explicitly opt-in
   'kimi': 'kimi',
   'deepseek': 'deepseek',
+  'codex': 'codex',
   'cursor': 'cursor',
   'gemini': 'gemini',
 }

package/src/runner.test.ts

@@ -0,0 +1,91 @@
+import { describe, test, expect } from 'bun:test'
+import { buildArenaPrompt, formatPlanOutput } from './runner'
+import { buildExecutionPlan, parseArenaToml } from './arena-toml'
+
+describe('buildArenaPrompt — plan-mode (pure, no IO)', () => {
+  test('includes CWD, Deck, and output directory', () => {
+    const prompt = buildArenaPrompt({
+      brief: 'Write a hello world function',
+      cwd: '/tmp/arena-cell',
+      deckPath: '/tmp/test-deck.toml',
+    })
+    expect(prompt).toContain('CWD: /tmp/arena-cell')
+    expect(prompt).toContain('Deck: /tmp/test-deck.toml')
+    expect(prompt).toContain('Produce output to: /tmp/arena-cell/')
+  })
+
+  test('includes decision-log.jsonl mandatory instructions', () => {
+    const prompt = buildArenaPrompt({
+      brief: 'test',
+      cwd: '/tmp/arena-cell',
+      deckPath: '/tmp/test-deck.toml',
+    })
+    expect(prompt).toContain('MANDATORY — write decision-log.jsonl')
+    expect(prompt).toContain('t (seconds elapsed)')
+    expect(prompt).toContain('phase (setup/content/design/output)')
+    expect(prompt).toContain('decision (what you chose)')
+    expect(prompt).toContain('reason (why)')
+  })
+
+  test('includes robustness and tools instructions', () => {
+    const prompt = buildArenaPrompt({
+      brief: 'test',
+      cwd: '/tmp/arena-cell',
+      deckPath: '/tmp/test-deck.toml',
+    })
+    expect(prompt).toContain('ROBUSTNESS')
+    expect(prompt).toContain('TOOLS')
+    expect(prompt).toContain('.claude/skills/')
+  })
+
+  test('task brief appears at the end', () => {
+    const prompt = buildArenaPrompt({
+      brief: 'Generate a dark-mode CSS theme',
+      cwd: '/tmp/arena-cell',
+      deckPath: '/tmp/test-deck.toml',
+    })
+    expect(prompt).toContain('TASK:')
+    expect(prompt).toContain('Generate a dark-mode CSS theme')
+    // Brief should be at the end (after MANDATORY, ROBUSTNESS, TOOLS)
+    const briefIdx = prompt.indexOf('Generate a dark-mode CSS theme')
+    const mandatoryIdx = prompt.indexOf('MANDATORY')
+    const toolsIdx = prompt.indexOf('TOOLS')
+    expect(briefIdx).toBeGreaterThan(mandatoryIdx)
+    expect(briefIdx).toBeGreaterThan(toolsIdx)
+  })
+
+  test('outputDir overrides default output path', () => {
+    const prompt = buildArenaPrompt({
+      brief: 'test',
+      cwd: '/tmp/arena-cell',
+      deckPath: '/tmp/test-deck.toml',
+      outputDir: '/custom/output',
+    })
+    expect(prompt).toContain('Produce output to: /custom/output/')
+  })
+
+  test('preflightReport included when provided', () => {
+    const prompt = buildArenaPrompt({
+      brief: 'test',
+      cwd: '/tmp/arena-cell',
+      deckPath: '/tmp/test-deck.toml',
+      preflightReport: '✅ 3 skills linked, 0 missing',
+    })
+    expect(prompt).toContain('Preflight:')
+    expect(prompt).toContain('✅ 3 skills linked, 0 missing')
+  })
+
+  test('no preflight section when report not provided', () => {
+    const prompt = buildArenaPrompt({
+      brief: 'test',
+      cwd: '/tmp/arena-cell',
+      deckPath: '/tmp/test-deck.toml',
+    })
+    expect(prompt).not.toContain('Preflight:')
+  })
+
+  test('prompt is deterministic — same inputs = same output', () => {
+    const opts = { brief: 'test', cwd: '/tmp/a', deckPath: '/tmp/d.toml' }
+    expect(buildArenaPrompt(opts)).toBe(buildArenaPrompt(opts))
+  })
+})

package/src/runner.ts

@@ -9,6 +9,7 @@ import { ArenaManifest, Player, type JudgeInput, type Evidence, type JudgeVerdic
 import type { ArenaManifest as ArenaManifestType } from '@lythos/test-utils/schema'
 try { await import('@lythos/agent-adapter-claude-sdk') } catch { /* package not installed */ }
 try { await import('@lythos/agent-adapter-deepseek-serve') } catch { /* package not installed */ }
+try { await import('@lythos/agent-adapter-codex') } catch { /* package not installed */ }
 import { runComparativeJudge } from './comparative-judge'
 import { parseArenaToml, buildExecutionPlan, type ArenaToml, type ExecutionPlan } from './arena-toml'
 import { resolvePlayer, resolveSides } from './player'
@@ -54,6 +55,7 @@ function resolveJudgeText(toml: ArenaToml, configDir?: string): string | null {
 
 // ── Prompt template (IoC: brief = variable, template = fixed contract) ────
 
+/** Pure prompt builder — no IO. Execution: arena subagent spawn (cli.ts singleRun / vsRun). */
 export function buildArenaPrompt(opts: {
   brief: string
   cwd: string