Skills

A skill is a playbook for the agent: a folder containing a SKILL.md that describes how to do a certain kind of work. When activated, its instructions are injected into the task, and the agent can read the skill's reference files and scripts on demand.

Skill format

A skill is just a folder — at minimum, one SKILL.md:

review-code/
├── SKILL.md          # required: skill definition
├── references/       # optional: reference material (checklists, specs…)
├── scripts/          # optional: executable scripts
├── assets/           # optional: static assets
└── templates/        # optional: template files

SKILL.md uses YAML frontmatter for metadata:

---
name: review-code
description: Review code changes for correctness, security, regressions,
  and maintainability.
version: 1.0.0
---

# Review workflow

Inspect the changed files, read `references/checklist.md` when needed,
run focused validation, and report findings in priority order.

Tip: the description decides when the agent reaches for the skill — write it as "when to use + what it does" for the best results.

Where skills come from

VibeSpace scans the following directories and aggregates what it finds on the Skills page:

Same-named skills from different sources are all kept, listed by source order (workspace skills first, bundled last) — enable whichever you prefer.

Enabling and disabling

1. Open the Skills navigation;
2. Browse the list: name, description, source, version;
3. Toggle each skill — only enabled skills become selectable in the task composer;
4. Built-in skills require a one-time acknowledgment (skills never widen permissions silently).

When the list changes (new folders, edits), hit refresh to rescan.

Practical scenarios

• Team standards: turn "code review checklist", "commit message rules", "release steps" into skills under the workspace's .claude/skills/ and version them;
• Complex flows: capture multi-step, multi-file processes (e.g. "pre-release check") so you never re-explain them in prompts;
• Cross-tool reuse: existing Claude / Codex skill folders work as-is — no migration needed.

Writing tips

• One skill, one job — avoid kitchen-sink skills;
• Put long reference material in references/ and point to it from SKILL.md so the agent loads it on demand instead of burning context;
• Scripts in scripts/ still obey the permission mode — commands go through approvals like any other.