@lythos/skill-arena 0.9.20 → 0.9.22

package/README.md

@@ -49,26 +49,26 @@ 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@0.9.20 <command>
+bunx @lythos/skill-arena@0.9.22 <command>
 ```
 
 ## Quick Start
 
 ```bash
 # Mode 1: Compare two skills on the same task
-bunx @lythos/skill-arena@0.9.20 \
+bunx @lythos/skill-arena@0.9.22 \
   --task "Generate auth flow diagram" \
   --skills "design-doc-mermaid,mermaid-tools" \
   --criteria "syntax,context,token"
 
 # Mode 2: Compare full deck configurations
-bunx @lythos/skill-arena@0.9.20 \
+bunx @lythos/skill-arena@0.9.22 \
   --task "Generate auth flow diagram" \
   --decks "./decks/minimal.toml,./decks/rich.toml" \
   --criteria "quality,token,maintainability"
 
 # Visualize results
-bunx @lythos/skill-arena@0.9.20 viz tmp/arena-<id>/
+bunx @lythos/skill-arena@0.9.22 viz tmp/arena-<id>/
 ```
 
 ## Commands
@@ -77,16 +77,16 @@ bunx @lythos/skill-arena@0.9.20 viz tmp/arena-<id>/
 
 ```bash
 # Print execution plan without running
-bunx @lythos/skill-arena@0.9.20 run --config arena.toml --dry-run
+bunx @lythos/skill-arena@0.9.22 run --config arena.toml --dry-run
 
 # Execute with per-side runs_per_side and statistical aggregation
-bunx @lythos/skill-arena@0.9.20 run --config arena.toml
+bunx @lythos/skill-arena@0.9.22 run --config arena.toml
 ```
 
 ### CLI-flag mode (backward compat)
 
 ```
-bunx @lythos/skill-arena@0.9.20 run \
+bunx @lythos/skill-arena@0.9.22 run \
   --task ./TASK-arena.md \
   --players ./players/claude.toml \
   --decks ./decks/run-01.toml,./decks/run-02.toml \
@@ -96,13 +96,13 @@ bunx @lythos/skill-arena@0.9.20 run \
 ### Scaffold mode (legacy, manual execution)
 
 ```
-bunx @lythos/skill-arena@0.9.20 scaffold --task "..." --skills a,b
+bunx @lythos/skill-arena@0.9.22 scaffold --task "..." --skills a,b
 ```
 
 ### Viz
 
 ```bash
-bunx @lythos/skill-arena@0.9.20 viz runs/arena-<id>/
+bunx @lythos/skill-arena@0.9.22 viz runs/arena-<id>/
 ```
 
 ## Skill Documentation
@@ -116,7 +116,7 @@ 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@0.9.20 ...
+Starter (this package) → npm publish → bunx @lythos/skill-arena@0.9.22 ...
 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": "0.9.20",
+  "version": "0.9.22",
   "description": "Skill Arena — benchmark skill effectiveness with controlled-variable comparison",
   "keywords": [
     "ai-agent",

package/src/cli.ts

@@ -109,7 +109,7 @@ async function agentRun(args: string[]) {
     taskPath = join(tmpDir, 'TASK.md')
     const briefTask = `---
 name: ad-hoc task
-description: ${opts.brief!.slice(0, 80)}
+description: ${opts.brief!.replace(/"/g, '\\"').slice(0, 80)}
 timeout: 120000
 ---
 
@@ -136,7 +136,7 @@ Evaluate whether the output is complete, accurate, and well-structured.
 
   const player = resolvePlayer(opts.player ?? 'kimi')
   const agent = useAgent(player)
-  const outDir = opts.out ? resolve(opts.out) : join(process.cwd(), 'agent-run-output')
+  const outDir = opts.out ? resolve(opts.out) : join(process.cwd(), `agent-output-${new Date().toISOString().replace(/[:.]/g, '-').slice(0, 19)}`)
   mkdirSync(outDir, { recursive: true })
 
   console.log(`🤖 agent-run: ${player} × ${deckPath}`)
@@ -164,16 +164,17 @@ Evaluate whether the output is complete, accurate, and well-structured.
   if (result.agentResult.stderr) writeFileSync(join(outDir, 'agent-stderr.txt'), result.agentResult.stderr, 'utf-8')
   if (result.verdict) writeFileSync(join(outDir, 'judge-verdict.json'), JSON.stringify(result.verdict, null, 2) + '\n', 'utf-8')
 
-  // Copy agent-produced files from workdir (output.md, output.docx, etc.)
+  // Copy all agent-produced files from workdir (output.md, output.docx, etc.)
+  // Skip .claude/ (symlink dir) and deck artifacts. Recursive so docx/pdf work.
   if (agentWorkdir) {
-    const { readdirSync, statSync, copyFileSync } = await import('node:fs')
+    const { cpSync, readdirSync } = await import('node:fs')
+    const skipSet = new Set(['.claude', 'skill-deck.toml', 'skill-deck.lock'])
     try {
       for (const entry of readdirSync(agentWorkdir)) {
-        if (entry.startsWith('.') || entry === 'skill-deck.toml' || entry === 'skill-deck.lock') continue
+        if (skipSet.has(entry)) continue
         const src = join(agentWorkdir, entry)
-        try {
-          if (statSync(src).isFile()) copyFileSync(src, join(outDir, entry))
-        } catch {}
+        const dest = join(outDir, entry)
+        try { cpSync(src, dest, { recursive: true }) } catch {}
       }
     } catch {}
   }