AI PLATE/ DocsBack to AI Plate
Documentation / Skills
YOUR AI PLATE HANDBOOK

Skill format & activation

Frontmatter, filter modes, trigger matching, persistence, and known limits.

Desktop source reviewed October 5, 2026

Skill file reference

FieldImport behavior
name / titleSets the visible skill name. The filename is used when omitted.
idOptional stable ID. Lowercase letters, digits, hyphens, and underscores are preserved; other characters become hyphens.
description / descVisible description. If omitted, the first directive paragraph is used.
categorywriting_voice, engineering_discipline, architecture_planning, quality_security, reasoning_verification, or custom. Unknown values are retained as the category value.
triggers / keywords / tagsComma- or semicolon-separated phrases. If absent, the imported ID becomes the trigger.
alwaysActiveUse camelCase for import. When true, it applies in Smart Filter mode even if no trigger matches.
Body after frontmatterStored as promptInstructions and injected verbatim when active.

How activation works

Only enabled skills are eligible. In Smart Filter mode, a skill activates when alwaysActive is true or one trigger matches the prompt. Single-word triggers use word-boundary matching after lowercasing and punctuation cleanup. Multi-word triggers use phrase matching. In Always Active mode, every enabled skill activates for every prompt.

Examples
Prompt: "Please debug this error"
Trigger: "debug"              → matches

Prompt: "Write a unit test"
Trigger: "unit test"          → matches

Prompt: "refactoring"
Trigger: "refactor"           → does not match as a single-word trigger
text

Editing, import, export, and reset

Skills are stored in the application database under the AI Plate skills configuration metadata. Built-in skills can be enabled or disabled, but cannot be deleted. Custom skills can be edited, exported, or deleted in the Skills & Directives settings tab. Reset restores built-ins and removes custom skills.

Implementation sources

Paths in your AI Plate source checkout:

  • core/skills-manager.ts
  • ui/app.js
  • electron/ipc-bridge.ts

Writing useful directives

  • State the task boundary, ordering, and stop conditions directly.
  • Say what needs confirmation before any irreversible action.
  • Avoid provider-specific API keys, personal data, and machine-specific paths in a distributable skill.
  • Keep the directive short enough to apply predictably and test it against several representative prompts.
  • Use a plugin or connector for execution. A skill should shape decisions and output, not pretend to have capabilities it does not have.
AI Plate product & developer documentation · Static reference, no API keys collected.