how subagent activate the skill

[jixian01]

Hey all,
in cc official docu https://code.claude.com/docs/en/sub-agents, it state:
"skills No Comma-separated list of skill names to auto-load when the subagent starts. Skills are loaded into the subagent’s context automatically"
my understand if the skills listed frontmatter will be auto loaded and subagent could use the skill during conversation session. not listed skills are unknown in the subagent context and can't be used.

i saw almost all of the subagents in this repo don't have skills config in frontmatter. does the subagent works with skill? or the subagents don't need skill.

BR

[wshobson]

Hey @jixian01, good observation — the absence of skills: in frontmatter is intentional.

The skills: field in the subagent docs preloads specific skills into a subagent's context at startup. This repo deliberately doesn't use it because Claude Code already auto-discovers skills based on their description: triggers (phrases like "Use when…", "Use PROACTIVELY when…", etc.). When a subagent encounters a relevant task, it matches against those descriptions and invokes the skill on demand — no preload needed.

Concretely: if a subagent is doing Python work and a python-development:* skill's description matches, it gets pulled in automatically. Preloading via skills: would force the skill into context whether the subagent needs it or not, which costs tokens unnecessarily.

See the "Skills (cross-harness)" section in AGENTS.md for the discovery model across harnesses, and docs/agent-skills.md for the progressive-disclosure rationale. The short answer: yes, subagents can use skills, and they activate them by description-matching rather than explicit declaration.