hermes - 💡(How to fix) Fix fix(cli): unknown skills should warn instead of crash to prevent Kanban worker auto-blocking

Problem

When a Kanban task references a non-existent skill (e.g. skills=["research"] instead of skills=["arxiv"]), the worker process crashes on startup with ValueError: Unknown skill(s): research.

This causes:

1. Worker exits with code 1 (crash)
2. Dispatcher retries, crashes again
3. After 2 consecutive failures, the task is auto-blocked
4. The task can never complete without manual intervention

Root Cause

Both cli.py:14030 and tui_gateway/server.py:1877 raise ValueError when build_preloaded_skills_prompt returns missing skills. This is too aggressive — a typo in a skill name should not crash the entire worker.

Fix

Changed both locations to log a warning and continue with whatever skills loaded successfully:

# Before:
raise ValueError(f"Unknown skill(s): {missing_display}")

# After:
logger.warning(
    "Unknown skill(s) requested, skipping: %s. "
    "Available skills can be listed with `hermes skills list`.",
    missing_display,
)

This is defense-in-depth: even if invalid skills get into the DB through any path (API, manual SQL, agent error), the worker survives and does useful work with the valid skills.

Files Changed

• cli.py — main CLI entry point
• tui_gateway/server.py — TUI gateway session init