Slash commands in skill arguments are not parsed as skill invocations

Bug Description

When invoking a skill with another skill referenced in the arguments (e.g., /drawio_v2 /drawio-formatting create a diagram...), only the first skill (drawio_v2) is loaded. The second /drawio-formatting is passed as plain argument text — it is never loaded via the Skill tool or expanded.

Steps to Reproduce

1. Have two skills registered (e.g., drawio_v2 and drawio-formatting)
2. Invoke: /drawio_v2 /drawio-formatting create a diagram of a network
3. Observe that only drawio_v2 SKILL.md is expanded
4. /drawio-formatting appears as literal text in the arguments — the skill is never loaded

Expected Behavior

All /skill-name references in a prompt should be recognized and loaded as skills, or the agent should be explicitly instructed to detect /skill-name patterns in arguments and invoke them via the Skill tool.

Actual Behavior

Only the first /command is processed as a skill invocation. Subsequent /skill-name tokens in the arguments are treated as inert plain text. The agent receives no signal to load them.

Impact

Users who want to compose multiple skills in a single prompt (e.g., a diagram generator + a formatting/theming skill) get silently degraded output because the companion skill's knowledge is never available to the agent.