Versions 1.82.7 and 1.82.8 of litellm were published on March 24, 2026
and contained malicious code that exfiltrated credentials (SSH keys,
cloud credentials, .env files, API keys) to an attacker-controlled domain.
Pin the dependency to >=1.70.0,<1.82.7 in both pyproject.toml and
requirements.txt as a stopgap until litellm can be replaced with direct
provider SDK calls.
See: https://github.com/HKUDS/OpenSpace/issues/31
Ref: PYSEC-2026-2, BerriAI/litellm#24521
Fixes#29. When a skill is already registered, register_skill_dir()
returned None, which caused fix_skill() to incorrectly report a failure.
Now returns the existing SkillMeta instead of None when the skill_id
is already present in the registry, making register_skill_dir() truly
idempotent as its callers (fix_skill, _auto_register_skill_dirs) expect.
Per reviewer feedback: keep strict \byes\b / \bno\b word boundaries to
prevent false positives, but widen confirm/reject/skip to stem-style
\bconfirm\w*\b etc. so common LLM variants like "confirmed", "rejected",
"skipping" still parse correctly instead of falling through to the
default False path.
