[DOCS] Cowork resolves Skills from the claude.ai deployment mount, not local ~/.claude/skills — undocumented, breaks scheduled tasks

Documentation Type

Unclear/confusing documentation

Documentation Location

_No response_

Section/Topic

Doc regarding skills and different surfaces

Current Documentation

The documentation is unclear on how Cowork resolves skills

What's Wrong or Missing?

Environment: Cowork desktop app (Windows 11); also Claude Code CLI. Two identically-configured machines with ~/.claude/skills kept in sync via git.

Summary: Skill resolution differs by runtime, and the Cowork behavior is undocumented. A scheduled Cowork task failed with "skill not found" for a skill whose file is physically present on the machine.

Observed:

Claude Code CLI reads the real local ~/.claude/skills git repo — a local-only skill (daily-summary) resolves fine.
Cowork (both interactive and scheduled) reads a mount that mirrors the account's claude.ai deployment, not the local repo. ls inside a Cowork session showed 13 skills (system + deployed) — the local-only daily-summary was absent, even though the file exists on disk. The local repo contains more skills than the mount.
Net effect: a scheduled Cowork task invoking daily-summary fails, despite the file being present locally.
Expected: Either Cowork reads local ~/.claude/skills like the CLI, or the docs clearly state that Cowork/scheduled tasks resolve skills only from the claude.ai-deployed set (and that local-only skills are invisible there).

Impact: Silent, confusing failure. The file exists, the interactive CLI works, but the scheduled Cowork run can't find the skill — with no indication that a separate deployment mount is in play.

Suggested Improvement

Asks:

Document the skill-resolution model per runtime (CLI = local disk; Cowork = deployment mount; Chat = deployment).
Consider making the mismatch discoverable (e.g. a warning when a scheduled task references a skill not in the mount), or letting Cowork fall back to local ~/.claude/skills.

Impact

High - Prevents users from using a feature

Additional Context

_No response_