fix(skills_config): handle null/scalar values in get_disabled_skills

[zhanggttry]

Fix #13026 — get_disabled_skills crash on null/scalar values

Problem

get_disabled_skills() in hermes_cli/skills_config.py crashes or produces incorrect results when the YAML config contains unexpected but plausible values:

1. skills: null → AttributeError: 'NoneType' object has no attribute 'get'
2. skills.disabled: my-skill (scalar string) → set("my-skill") splits into individual characters: {'m', 'y', '-', 's', 'k', 'i', 'l', 'l'}
3. skills.platform_disabled.telegram: my-skill → same character-splitting bug

These edge cases arise from hand-edited YAML configs where a user might write:

skills:
  disabled: my-skill     # scalar instead of list

or where skills: is set to null by another tool.

Solution

• Add _normalize_string_set() helper that safely coerces None, scalar strings, and lists into proper Set[str]. This mirrors the existing _normalize_string_set() in agent/skill_utils.py (L144-168) which already handles these cases for the runtime skill loader.
• Guard skills_cfg against None and non-dict values (config.get("skills") or {})
• Guard platform_disabled against None values (skills_cfg.get("platform_disabled") or {})

Changes

File	Change
hermes_cli/skills_config.py	Add _normalize_string_set(), harden get_disabled_skills()
tests/test_skills_config.py	New: 18 test cases covering all edge cases

Testing

19 passed, 0 failed ✅

All edge cases covered: null config, scalar strings, empty values, normal lists, platform fallbacks, and overrides.

[Bartok9]

Salvaged your fix onto current main in #50648 (credited to you + @YangManBOBO). This branch predates main's global | platform_disabled union rewrite (#46201) and now conflicts; #50648 re-applies the null/scalar normalization on top of current main while preserving that union semantics, and mirrors the existing agent.skill_utils._normalize_string_set helper. Full credit to your original work. 🙏

[teknium1]

Closing — this fix landed on main via PR #61797. You were the earliest submitter of this fix (April 20, nearly three months before the others), and your implementation was essentially the shape that ended up merged: null → empty set, scalar → single skill name, mirroring agent.skill_utils._normalize_string_set. Apologies that this sat unreviewed long enough for the fix to be rebuilt independently — the credit for first identifying and fixing #13026 is yours. Thanks for the contribution!