Persona Task Awareness

# Persona Task-Type Awareness Implementation Plan

> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.

**Goal:** Integrate Hu et al.'s persona-accuracy tradeoff findings into the skill ecosystem so assembled personas help alignment tasks without degrading knowledge-retrieval tasks.

**Architecture:** Four-part change — (1) add Hu et al. to the research foundation in authorities.md, (2) add an alignment-vs-knowledge split to persona design guidance, (3) add a dynamic task-type assessment heuristic to the persona template, (4) add context-placement awareness to output-structure.md so knowledge-heavy persona content lands in references (user-context) rather than rules (system-context).

**Research basis:** Hu, Rostami & Thomason, "Expert Personas Improve LLM Alignment but Damage Accuracy" (arXiv:2603.18507, March 2026). Tested 6 LLMs across MT-Bench, MMLU, and safety benchmarks with 12 expert personas at 3 granularity levels.

---

### Task 1: Add Hu et al. to authorities.md

**Files:**
- Modify: `.claude/references/authorities.md:54-68` (Agent Research section)
- Modify: `.claude/references/authorities.md:70-77` (What This Means for Persona Blending section)

This task adds the paper as a cited authority alongside the existing positive-side persona citations (Kong, Xu, Pei). It also updates the "What This Means" section to acknowledge the tradeoff.

- [ ] **Step 1: Add Hu et al. entry to Agent Research section**

Add after the existing Xu et al. entry (line 66), before the Hong et al. entry (line 68):