@lythos/skill-arena 0.9.40 → 0.9.41

package/README.md

@@ -49,23 +49,23 @@ Note: Claude `-p` mode has known issues with web tools in Bun.spawn (deferred to
 ```bash
 bun add -d @lythos/skill-arena
 # or use directly
-bunx @lythos/skill-arena@0.9.40 <command>
+bunx @lythos/skill-arena@0.9.41 <command>
 ```
 
 ## Quick Start
 
 ```bash
 # Single: test a deck with one agent (exec shortcut)
-bunx @lythos/skill-arena@0.9.40 single \
+bunx @lythos/skill-arena@0.9.41 single \
   --brief "Generate auth flow diagram" \
   --deck ./examples/decks/documents.toml
 
 # Vs: compare multiple decks side by side (declarative)
-bunx @lythos/skill-arena@0.9.40 vs \
+bunx @lythos/skill-arena@0.9.41 vs \
   --config examples/arena/research-compare/arena.toml
 
 # Visualize results
-bunx @lythos/skill-arena@0.9.40 viz tmp/arena-<id>/
+bunx @lythos/skill-arena@0.9.41 viz tmp/arena-<id>/
 ```
 
 ## Commands
@@ -74,16 +74,16 @@ bunx @lythos/skill-arena@0.9.40 viz tmp/arena-<id>/
 
 ```bash
 # Print execution plan without running
-bunx @lythos/skill-arena@0.9.40 vs --config arena.toml --dry-run
+bunx @lythos/skill-arena@0.9.41 vs --config arena.toml --dry-run
 
 # Execute with per-side runs_per_side and statistical aggregation
-bunx @lythos/skill-arena@0.9.40 vs --config arena.toml
+bunx @lythos/skill-arena@0.9.41 vs --config arena.toml
 ```
 
 ### CLI-flag mode (backward compat)
 
 ```
-bunx @lythos/skill-arena@0.9.40 run \
+bunx @lythos/skill-arena@0.9.41 run \
   --task ./TASK-arena.md \
   --players ./players/claude.toml \
   --decks ./decks/run-01.toml,./decks/run-02.toml \
@@ -93,13 +93,13 @@ bunx @lythos/skill-arena@0.9.40 run \
 ### Scaffold mode (legacy, manual execution)
 
 ```
-bunx @lythos/skill-arena@0.9.40 scaffold --task "..." --decks a.toml,b.toml
+bunx @lythos/skill-arena@0.9.41 scaffold --task "..." --decks a.toml,b.toml
 ```
 
 ### Viz
 
 ```bash
-bunx @lythos/skill-arena@0.9.40 viz runs/arena-<id>/
+bunx @lythos/skill-arena@0.9.41 viz runs/arena-<id>/
 ```
 
 ## Skill Documentation
@@ -113,7 +113,7 @@ The agent-visible **Skill** layer documentation is here:
 Part of the [lythoskill](https://github.com/lythos-labs/lythoskill) ecosystem — the thin-skill pattern separates heavy logic (this npm package) from lightweight agent instructions (SKILL.md).
 
 ```
-Starter (this package) → npm publish → bunx @lythos/skill-arena@0.9.40 ...
+Starter (this package) → npm publish → bunx @lythos/skill-arena@0.9.41 ...
 Skill   (packages/<name>/skill/)     → build → SKILL.md + thin scripts
 Output  (skills/<name>/)             → git commit → agent-visible skill
 ```

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lythos/skill-arena",
-  "version": "0.9.40",
+  "version": "0.9.41",
   "description": "Skill Arena — benchmark skill effectiveness with controlled-variable comparison",
   "keywords": [
     "ai-agent",

package/src/cli.ts

@@ -152,6 +152,27 @@ async function singleRun(args: string[]) {
    Format: frontmatter (name, description, timeout) + Given/When/Then/Judge sections.
    Example: see playground/arena-one-shot/TASK-arena.agent.md`); process.exit(1) }
     scenarioOpt.scenarioPath = taskPath
+    // Quick validation: check frontmatter presence
+    const raw = readFileSync(taskPath, 'utf-8')
+    if (!raw.startsWith('---')) { console.error(`❌ Invalid .agent.md: missing frontmatter (must start with "---")
+   Correct format:
+   ---
+   name: my-scenario
+   description: what this tests
+   timeout: 120000
+   ---
+   ## Given
+   ...
+   ## When
+   ...
+   ## Then
+   ...
+   ## Judge
+   ...
+   Template: playground/arena-one-shot/TASK-arena.agent.md`); process.exit(1) }
+    if (!raw.includes('## When')) { console.error(`❌ Invalid .agent.md: missing "## When" section.
+   The ## When section defines what the agent should do.
+   See template: playground/arena-one-shot/TASK-arena.agent.md`); process.exit(1) }
   } else {
     scenarioOpt.scenario = {
       name: 'ad-hoc task',

package/src/runner.ts

@@ -1,4 +1,4 @@
-import { mkdirSync, writeFileSync, readFileSync } from 'node:fs'
+import { existsSync, mkdirSync, writeFileSync, readFileSync, rmSync } from 'node:fs'
 import { join, resolve } from 'node:path'
 import { tmpdir } from 'node:os'
 import { runAgentScenario, type AgentScenario } from '@lythos/test-utils/agent-bdd'
@@ -57,7 +57,35 @@ export async function runArenaFromToml(opts: {
     if (configDir) return resolve(configDir, p)
     return resolve(p)
   }
-  const taskAbs = resolvePath(taskPath)
+  const resolveOrCreateTask = (): { path: string; cleanup?: () => void } => {
+    const candidate = resolvePath(taskPath)
+    if (existsSync(candidate)) return { path: candidate }
+    // taskPath is inline text — write temp scenario file
+    const tmp = join(tmpdir(), `arena-task-${stamp()}.agent.md`)
+    writeFileSync(tmp, `---
+name: arena task
+description: ${taskPath.slice(0, 80)}
+timeout: 120000
+---
+
+## Given
+- Working directory with an empty project
+- bun is available
+
+## When
+${taskPath}
+
+## Then
+- Complete the task above
+- Write a summary to output.md
+
+## Judge
+- completeness
+- correctness
+`)
+    return { path: tmp, cleanup: () => { try { rmSync(tmp) } catch {} } }
+  }
+  const { path: taskAbs, cleanup: taskCleanup } = resolveOrCreateTask()
   const resolvedToml: ArenaToml = {
     ...toml,
     side: toml.side.map(s => ({ ...s, deck: resolvePath(s.deck) })),
@@ -78,10 +106,13 @@ export async function runArenaFromToml(opts: {
   const resolved = resolveSides(resolvedToml)
 
   // Build manifest
+  const taskContent = existsSync(taskAbs)
+    ? readFileSync(taskAbs, 'utf-8').slice(0, 200)
+    : taskPath // inline description, not a file path
   const manifest = ArenaManifest.parse({
     id: arenaId,
     created_at: new Date().toISOString(),
-    task: readFileSync(taskAbs, 'utf-8').slice(0, 200),
+    task: taskContent,
     mode: 'decks',
     participants: [...new Map(resolved.map(r => [r.side.name, r])).values()].map(r => ({
       id: r.side.name,