@lythos/skill-arena 0.9.23 → 0.9.24

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@0.9.23 <command>
+bunx @lythos/skill-arena@0.9.24 <command>
 ```
 
 ## Quick Start
 
 ```bash
 # Mode 1: Compare two skills on the same task
-bunx @lythos/skill-arena@0.9.23 \
+bunx @lythos/skill-arena@0.9.24 \
   --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@0.9.23 \
+bunx @lythos/skill-arena@0.9.24 \
   --task "Generate auth flow diagram" \
   --decks "./decks/minimal.toml,./decks/rich.toml" \
   --criteria "quality,token,maintainability"
 
 # Visualize results
-bunx @lythos/skill-arena@0.9.23 viz tmp/arena-<id>/
+bunx @lythos/skill-arena@0.9.24 viz tmp/arena-<id>/
 ```
 
 ## Commands
@@ -77,16 +77,16 @@ bunx @lythos/skill-arena@0.9.23 viz tmp/arena-<id>/
 
 ```bash
 # Print execution plan without running
-bunx @lythos/skill-arena@0.9.23 run --config arena.toml --dry-run
+bunx @lythos/skill-arena@0.9.24 run --config arena.toml --dry-run
 
 # Execute with per-side runs_per_side and statistical aggregation
-bunx @lythos/skill-arena@0.9.23 run --config arena.toml
+bunx @lythos/skill-arena@0.9.24 run --config arena.toml
 ```
 
 ### CLI-flag mode (backward compat)
 
 ```
-bunx @lythos/skill-arena@0.9.23 run \
+bunx @lythos/skill-arena@0.9.24 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@0.9.23 run \
 ### Scaffold mode (legacy, manual execution)
 
 ```
-bunx @lythos/skill-arena@0.9.23 scaffold --task "..." --skills a,b
+bunx @lythos/skill-arena@0.9.24 scaffold --task "..." --skills a,b
 ```
 
 ### Viz
 
 ```bash
-bunx @lythos/skill-arena@0.9.23 viz runs/arena-<id>/
+bunx @lythos/skill-arena@0.9.24 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@0.9.23 ...
+Starter (this package) → npm publish → bunx @lythos/skill-arena@0.9.24 ...
 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": "0.9.23",
+  "version": "0.9.24",
   "description": "Skill Arena — benchmark skill effectiveness with controlled-variable comparison",
   "keywords": [
     "ai-agent",
@@ -37,6 +37,7 @@
     "bun": ">=1.0.0"
   },
   "dependencies": {
+    "@lythos/cold-pool": "workspace:*",
     "@lythos/test-utils": "^0.9.1",
     "zod": "^3.24.0",
     "zod-to-json-schema": "^3.25.2"

package/src/cli.ts

@@ -37,8 +37,8 @@ function printHelp(): void {
   console.log(`🎭 lythoskill-arena — Skill comparison runner
 
 Usage:
-  lythoskill-arena agent-run --task <path> --deck <path> [--player kimi] [--out <dir>]
-  lythoskill-arena agent-run --brief "<prompt>" --deck <path> [--out <dir>]
+  lythoskill-arena agent-run --task <path> --deck <path> [--player kimi] [--out <dir>] [--timeout <ms>]
+  lythoskill-arena agent-run --brief "<prompt>" --deck <path> [--out <dir>] [--timeout <ms>]
   lythoskill-arena run --task <path> --players <A.toml,B.toml> --decks <A.toml,B.toml> --criteria <c1,c2,...> [--out <dir>]
   lythoskill-arena scaffold --task "<description>" --skills <skill1,skill2,...>
   lythoskill-arena scaffold --task "<description>" --decks <deck1,deck2,...>
@@ -90,6 +90,7 @@ async function agentRun(args: string[]) {
     else if (args[i] === '--deck' || args[i] === '-d') opts.deck = args[++i]
     else if (args[i] === '--player' || args[i] === '-p') opts.player = args[++i]
     else if (args[i] === '--out' || args[i] === '-o') opts.out = args[++i]
+    else if (args[i] === '--timeout') opts.timeout = args[++i]
   }
 
   if (!opts.deck) {
@@ -127,7 +128,7 @@ async function agentRun(args: string[]) {
     scenarioOpt.scenario = {
       name: 'ad-hoc task',
       description: opts.brief!.slice(0, 80),
-      timeout: 120000,
+      timeout: Number(opts.timeout ?? 120000),
       given: { deck: {} },
       when: opts.brief!,
       then: ['Write your output to output.md', 'The output should be complete and well-structured'],

package/src/preflight.ts

@@ -6,6 +6,8 @@
  * IO is injected via function parameters (e.g., existsFn, readdirFn).
  */
 
+import { ColdPool, parseLocator } from '@lythos/cold-pool'
+
 // ── Types ─────────────────────────────────────────────────────────────────
 
 /** A skill as declared in skill-deck.toml */
@@ -83,8 +85,11 @@ export function parseDeckSkills(
 /**
  * Check each declared skill against the cold pool filesystem.
  *
- * For skills with explicit paths: resolve `<coldPoolDir>/<path>/SKILL.md`
- * For skills without paths (array format): resolve `<coldPoolDir>/<name>/SKILL.md`
+ * Path resolution delegates to @lythos/cold-pool's `parseLocator` and
+ * `ColdPool.resolveDir` so localhost / FQ / standalone forms all map to
+ * the right physical layout (per ADR-20260507021957847). Non-FQ legacy
+ * names (e.g., bare `pdf`) fall back to `<coldPoolDir>/<name>/SKILL.md`.
+ *
  * Skills with HTTP/URL paths are skipped (not local).
  *
  * `existsFn` is the IO injection point — swap for real fs or mock.
@@ -94,11 +99,28 @@ export function checkSkillExistence(
   coldPoolDir: string,
   existsFn: (path: string) => boolean
 ): SkillCheck[] {
+  const pool = new ColdPool(coldPoolDir)
   return skills.map(skill => {
-    const resolvedName = skill.path && !skill.path.startsWith('http')
+    const candidatePath = skill.path && !skill.path.startsWith('http')
       ? skill.path
       : skill.name
-    const expectedPath = `${coldPoolDir}/${resolvedName}/SKILL.md`
+
+    let expectedPath: string
+    const locator = parseLocator(candidatePath)
+    if (!locator) {
+      // Legacy bare-name fallback. Per ADR-20260502012643244 this should
+      // be removed in 0.10.x once arena.toml authors switch to FQ.
+      expectedPath = `${coldPoolDir}/${candidatePath}/SKILL.md`
+    } else if (locator.isLocalhost) {
+      // localhost layout: top-level dir under coldPool, no `localhost/` prefix
+      expectedPath = `${pool.resolveDir(locator)}/SKILL.md`
+    } else if (locator.skill) {
+      expectedPath = `${pool.resolveDir(locator)}/${locator.skill}/SKILL.md`
+    } else {
+      // Standalone repo: SKILL.md at repo root
+      expectedPath = `${pool.resolveDir(locator)}/SKILL.md`
+    }
+
     return {
       name: skill.name,
       expectedPath,