@lythos/skill-arena 0.9.48 → 0.9.49

package/README.md

@@ -49,20 +49,20 @@ 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@0.9.48 <command>
+bunx @lythos/skill-arena@0.9.49 <command>
 ```
 
 ## Quick Start
 
 ```bash
 # Single: test a deck with one agent
-bunx @lythos/skill-arena@0.9.48 single \
+bunx @lythos/skill-arena@0.9.49 single \
   --deck https://raw.githubusercontent.com/lythos-labs/lythoskill/main/examples/decks/scout.toml \
   --brief "Generate auth flow diagram"
 
 # Vs: compare multiple decks side by side
 curl -fsSL https://raw.githubusercontent.com/lythos-labs/lythoskill/main/examples/arena/research-compare/arena.toml > arena.toml
-bunx @lythos/skill-arena@0.9.48 vs --config ./arena.toml
+bunx @lythos/skill-arena@0.9.49 vs --config ./arena.toml
 ```
 
 ## Commands
@@ -71,23 +71,23 @@ bunx @lythos/skill-arena@0.9.48 vs --config ./arena.toml
 
 ```bash
 # Print execution plan without running
-bunx @lythos/skill-arena@0.9.48 vs --config arena.toml --dry-run
+bunx @lythos/skill-arena@0.9.49 vs --config arena.toml --dry-run
 
 # Execute with per-side runs_per_side and statistical aggregation
-bunx @lythos/skill-arena@0.9.48 vs --config arena.toml
+bunx @lythos/skill-arena@0.9.49 vs --config arena.toml
 ```
 
 ### Scaffold mode (legacy, manual execution)
 
 ```
-bunx @lythos/skill-arena@0.9.48 scaffold --task "Generate auth flow diagram" \
+bunx @lythos/skill-arena@0.9.49 scaffold --task "Generate auth flow diagram" \
   --decks https://raw.githubusercontent.com/lythos-labs/lythoskill/main/examples/decks/scout.toml,https://raw.githubusercontent.com/lythos-labs/lythoskill/main/examples/decks/documents.toml
 ```
 
 ### Viz
 
 ```bash
-bunx @lythos/skill-arena@0.9.48 viz runs/arena-<id>/
+bunx @lythos/skill-arena@0.9.49 viz runs/arena-<id>/
 ```
 
 ## Skill Documentation
@@ -101,7 +101,7 @@ 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@0.9.48 ...
+Starter (this package) → npm publish → bunx @lythos/skill-arena@0.9.49 ...
 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": "0.9.48",
+  "version": "0.9.49",
   "description": "Skill Arena — benchmark skill effectiveness with controlled-variable comparison",
   "keywords": [
     "ai-agent",

package/src/cli.ts

@@ -54,11 +54,9 @@ Options:
       --deck <path>      Deck path (single only)
       --brief "<text>"   Inline task description (single only, alternative to --task)
       --player <name>    Agent player (single only, default: kimi)
-  -c, --criteria <list>  Evaluation criteria (scaffold only, default: syntax,context,logic,token)
       --config <path>    Path to arena.toml (vs only)
       --dry-run          Print execution plan without running (vs --config only)
       --out <dir>        Output directory
-  -d, --dir <dir>        Parent dir (scaffold: defaults to tmp)
   -p, --project <dir>    Project root (default: .)
       --timeout <ms>     Subagent timeout (single only)
 
@@ -202,6 +200,7 @@ async function singleRun(args: string[]) {
   // Optional: register claude-sdk adapter if the package is installed
   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 */ }
   const { runAgentScenario } = await import('@lythos/test-utils/agent-bdd')
   const { resolvePlayer } = await import('./player')
 
@@ -338,8 +337,6 @@ function parseArgs(argv: string[]) {
 
   const options: Record<string, string | undefined> = {
     task: undefined,
-    decks: undefined,
-    criteria: 'syntax,context,logic,token',
     dir: 'tmp',
     project: '.',
     config: undefined,
@@ -352,11 +349,6 @@ function parseArgs(argv: string[]) {
     const arg = argv[i]
     if (arg === '--task' || arg === '-t') {
       options.task = argv[++i]
-    } else if (arg === '--decks') {
-      options.decks = argv[++i]
-    } else if (arg === '--criteria' || arg === '-c') {
-      options.criteria = argv[++i]
-      options.control = argv[++i]
     } else if (arg === '--dir' || arg === '-d') {
       options.dir = argv[++i]
     } else if (arg === '--project' || arg === '-p') {

package/src/path-guard.ts

@@ -0,0 +1,92 @@
+/**
+ * path-guard — Path validation for arena CLI.
+ *
+ * Arena accepts deck/task/player paths from CLI arguments. These must
+ * be validated to prevent arbitrary file reads and directory traversal.
+ *
+ * Reference: arena sweep 2026-05-10 P1 path-trust findings
+ */
+
+import { resolve, isAbsolute } from "node:path"
+import { existsSync } from "node:fs"
+
+/**
+ * Validate a deck path: must point to a file within the project or
+ * be a valid absolute path. Refuse paths containing ".." traversal.
+ */
+export function validateDeckPath(raw: string, projectDir: string): string {
+  if (raw.includes("..")) {
+    throw new Error(`Deck path contains parent traversal (..): ${raw}`)
+  }
+  if (raw.includes("\0")) {
+    throw new Error(`Deck path contains null byte`)
+  }
+
+  const resolved = isAbsolute(raw) ? resolve(raw) : resolve(projectDir, raw)
+
+  if (!resolved.startsWith(resolve(projectDir) + "/") && !isAbsolute(raw)) {
+    throw new Error(
+      `Deck path "${raw}" resolves outside the project directory.\n` +
+      `  Resolved: ${resolved}\n` +
+      `  Project:  ${resolve(projectDir)}`
+    )
+  }
+
+  return resolved
+}
+
+/**
+ * Validate a task path: must be a .md or .agent.md file, must exist,
+ * must not traverse outside the project.
+ */
+export function validateTaskPath(raw: string, projectDir: string): string {
+  if (raw.includes("..")) {
+    throw new Error(`Task path contains parent traversal (..): ${raw}`)
+  }
+  if (raw.includes("\0")) {
+    throw new Error(`Task path contains null byte`)
+  }
+
+  const resolved = isAbsolute(raw) ? resolve(raw) : resolve(projectDir, raw)
+
+  if (!resolved.startsWith(resolve(projectDir) + "/")) {
+    throw new Error(
+      `Task path "${raw}" resolves outside the project directory.\n` +
+      `  Resolved: ${resolved}\n` +
+      `  Project:  ${resolve(projectDir)}`
+    )
+  }
+
+  if (!existsSync(resolved)) {
+    throw new Error(`Task file not found: ${raw}\n  Resolved: ${resolved}`)
+  }
+
+  return resolved
+}
+
+/**
+ * Validate the output directory — must be within the project or /tmp.
+ * Arena writes agent output and judge verdicts here.
+ */
+export function validateOutDir(raw: string, projectDir: string): string {
+  if (raw.includes("..")) {
+    throw new Error(`Output directory contains parent traversal (..): ${raw}`)
+  }
+
+  const resolved = isAbsolute(raw) ? resolve(raw) : resolve(projectDir, raw)
+  const resolvedProject = resolve(projectDir)
+
+  // Allow /tmp as a valid output target
+  if (resolved.startsWith("/tmp/") || resolved === "/tmp") {
+    return resolved
+  }
+
+  if (!resolved.startsWith(resolvedProject + "/")) {
+    throw new Error(
+      `Output directory "${raw}" is outside the project.\n` +
+      `  Use --out with a path under the project or in /tmp.`
+    )
+  }
+
+  return resolved
+}

package/src/preflight.ts

@@ -6,6 +6,7 @@
  * IO is injected via function parameters (e.g., existsFn, readdirFn).
  */
 
+import { join } from 'node:path'
 import { ColdPool, parseLocator } from '@lythos/cold-pool'
 
 // ── Types ─────────────────────────────────────────────────────────────────
@@ -170,8 +171,8 @@ export function buildCopyPlan(
   for (const name of entries) {
     if (skipSet.has(name)) continue
     plan.push({
-      src: `${workdir}/${name}`,
-      dest: `${outDir}/${name}`,
+      src: join(workdir, name),
+      dest: join(outDir, name),
       name,
     })
   }