Team training: multi-persona project decomposition not implemented

[joelteply]

airc-queue card

Coordinates work via the AIRC queue substrate (airc#562). Edit this card by commenting OR by running airc queue claim/airc queue release/airc queue heartbeat (later PRs).

{
  "kind": "airc-queue-card-v1",
  "id": "#384",
  "owner": "claude-tab-2",
  "status": "claimed",
  "evidence": "Adopted existing GitHub issue into airc queue.",
  "next_action": "Triage, claim, or close this adopted backlog card."
}

Close this issue when the work is done (status=merged/abandoned).

Original issue body

Pre-adoption body

Problem

The README promises: "Give the Academy a project — 'build a side-scrolling game with mushroom people' — and it decomposes it into roles (game designer, engineer, artist), trains each persona for their role, then orchestrates collaborative building."

This is the most ambitious Academy feature and doesn't exist yet. Current Academy can run single-persona challenges (teacher/student pairs). Multi-persona project decomposition with role assignment, parallel training, and collaborative building is not implemented.

What needs to happen

1. Project-level recipe type that decomposes into sub-tasks
2. Role assignment based on persona capabilities/genome
3. Parallel training sessions (one per role)
4. Collaborative building phase where trained personas work together
5. Teacher grades both the project AND individual role performance
6. Shared visibility in academy chat room (peer learning)

Prerequisites

• Academy: no full training session proven end-to-end #377 — Full single-persona academy session must work first
• First HuggingFace adapter publication from the system #378 — Adapter publication (team members need to share trained skills)
• Genome paging: activateSkill/evictLRU not wired end-to-end #382 — Genome paging (personas need to load role-specific adapters)