openwork/pr/soul-starter-observability

Soul Starter Observability + Steering

What changed

• Updated the bundled give-me-a-soul starter template to seed memory from real workspace context (AGENTS + sessions + todos + transcript snippets) instead of empty placeholders.
• Expanded the starter to create explicit interaction/observability commands (soul-status, steer-soul) alongside soul-heartbeat and take-my-soul-back.
• Strengthened Soul dashboard steering prompts so they explicitly pull from .opencode/soul.md, AGENTS guidance, and OpenCode sqlite activity.
• Added an Improvement sweep action in the Soul dashboard.
• Changed Enable soul mode on the Soul dashboard to inject the full bundled starter template prompt.

Why

• Soul mode check-ins were becoming stale/generic because the starter flow often left memory under-seeded.
• Users needed a clearer, built-in way to observe and steer Soul behavior without ad-hoc prompting.

Evidence

• pr/soul-starter-observability/evidence/soul-dashboard-steering.png
• pr/soul-starter-observability/evidence/session-enable-soul-prefilled.png

Validation

• pnpm --filter @different-ai/openwork-ui typecheck
• pnpm --filter @different-ai/openwork-ui test:health
• Brought up Docker stack for manual verification: packaging/docker/dev-up.sh
• Verified Soul dashboard renders new controls and state card in browser, then shut stack down.