Skills
Specialized instruction sets that turn the AI into a domain expert for each type of task.
A skill is what makes an agent behave differently in Feature mode versus Bug mode versus Review mode. Skills are structured instruction sets — Markdown files that give the AI deep, domain-specific knowledge about a specific type of task, its workflow, and how to communicate progress back to you.
When you activate a mode, the corresponding skill loads automatically. The agent knows exactly what to do, how to do it, and how to show you what's happening in real time.
10 mode skills
Each work mode is powered by a dedicated skill:
| Skill | What it enables |
|---|---|
| Feature | Action-first feature development with sidebar progress tracking |
| Bug | Systematic investigation → diagnosis → fix, read-only until confident |
| Plan | Codebase exploration, plan design, approval before any code changes |
| Review | Autonomous code review with structured inline comments |
| PR | Git operations: commit, push, create PR, resolve conflicts, rebase |
| Bisect | Binary search through git history to find regression-causing commits |
| Specforge | Transform a vague idea into a structured Epic + Stories specification |
| Analyze | Deep codebase analysis: quality, security, architecture, tech debt |
| Test Plan | Generate manual test plans from code changes with prioritized test cases |
| Deep Search | AI-powered codebase exploration with annotated, ranked results |
5 utility skills
Background helpers that activate automatically when needed:
| Skill | When it activates |
|---|---|
| Branch Rename | When a branch starts with a generic prefix — gives it a meaningful name |
| Conversation Name | When a conversation is "Untitled" — names it based on the task |
| Edit | When you highlight code and request an inline change |
| Question | When you ask about a selected code snippet |
| Toast | On task completion or errors — sends a notification to the IDE |
How skills are loaded
Skills are bundled with Innovatic. When a new worktree is created, the full set of skills is synced into that workspace automatically. Each mode injects the appropriate skill into the agent's context before your first message arrives.
You don't configure anything — the right skill for the right mode is always there.
Extending skills
Adding a new skill is straightforward:
- Create a new directory with a
SKILL.mdfile - Write the instructions in Markdown — workflow phases, constraints, CLI integration
- The skill is automatically available in all new worktrees
Custom per-repository instructions can extend any skill's behavior without modifying the skill itself — useful for project-specific conventions and rules.
Skills define what the agent does. Rules (always loaded) define the defaults that apply to every conversation regardless of mode.
Related features
- Work Modes — the 10 modes powered by these skills
- Documents — the structured output each skill produces