Claude Code ↔ OpenClaw Skill Sync

The skill declares it will scan ~/.claude/skills/ and ~/.openclaw/workspace/skills/ and read/write SKILL.md files; all required actions (directory listing, reading/writing SKILL.md, creating _meta.json) are directly related to the stated purpose of syncing skills.

Instructions are narrowly scoped to listing one-level subdirectories, reading SKILL.md contents, comparing them, and copying/overwriting files with explicit user prompts. Two items to note: (1) the security rules say 'Never output token values, API keys, or credentials that may appear in skill files' — SKILL.md files may legitimately contain secrets, and the skill must read full contents to copy them, so there's a risk of accidental exposure if the agent were later instructed to transmit content elsewhere; (2) the instruction 'Always use ~ and never expand ~ to an absolute path' is unusual (it prevents the skill from resolving absolute locations) and could interfere with auditing or access-control checks — it's not necessary for the stated purpose and is worth questioning.

Instruction-only skill with no install steps or downloads. No files are written as part of an installation process beyond the operations the skill explicitly instructs at runtime.

No environment variables, binaries, or external credentials are requested. The skill only needs filesystem access to the two specified directories, which is proportional to its stated function.

The skill does not request always:true and does not modify other skills' configs beyond writing SKILL.md and _meta.json into the specified Claude/OpenClaw skill directories — these writes are the expected behavior for a sync tool.