Phase 5 (v0.7.0): Agent integration — skills, e2e demo, mock full-pipeline

[Copilot]

Implements the three remaining Phase 5 items to lower agent onboarding friction: updated skills documentation covering all 5 Harness layers, a runnable end-to-end demo, and mock-mode support for the full pipeline.

.claude/skills/longbridge-trading/SKILL.md

• Rewrote to cover all 5 layers (L1 Research → L2 Signal → L3 Portfolio → L4 Risk → L5 Execution)
• Added correct field names (weight, total_capital_pct, cash_reserve_pct), TWAP/ICEBERG algo order examples, signal-to-order traceability (signal_refs), and a complete pipeline workflow
• Updated file system reference table with ← / → annotations showing agent read/write responsibilities

demo_e2e.sh

New 9-step script demonstrating the full pipeline in mock mode:

init → watchlist → signal defs → portfolio target → risk policy
→ controller (mock) → inspect L1–L3 output → submit orders → stop

Produces visible output at every layer so an agent can trace data flow end-to-end.

internal/research/mock.go + main.go

Previously, --mock mode skipped research entirely (if !useMock gate). Now:

• RefreshFeedsMock(root) — generates synthetic news/topics for watchlist symbols without hitting the Content API
• GenerateMockKlineData(root, symbol, 120) — writes synthetic daily OHLCV bars so builtin signals (SMA_CROSS, RSI, PRICE_CHANGE) can compute in mock mode
• Controller calls both on every tick when useMock == true

// mock path (was a no-op before)
research.RefreshFeedsMock(root)
if wl, err := research.ParseWatchlist(root); err == nil {
    for _, sym := range wl.Symbols {
        research.GenerateMockKlineData(root, sym, 120)
    }
}