Skip to main content

Skills

Skills let you extend the agent with reusable, on-demand expertise. A skill is a folder that contains instructions (and optionally scripts and reference materials) that the agent can load when it’s relevant.

Unlike pasting huge guides into the chat, skills follow a progressive disclosure pattern:

  • On startup, the agent only preloads each skill’s metadata (name + description), which is token-efficient.
  • When a task matches, the agent can load the skill by reading its SKILL.md (and any referenced files) from disk.

What Skills are good for​

Skills work especially well for:

  • Repeatable workflows you want the agent to follow consistently (e.g., “how we do releases”, “how we write PR descriptions”, “our brand guidelines”).
  • Domain-specific playbooks that don’t belong in the model’s general knowledge (internal processes, project conventions).
  • Tasks that benefit from executable helpers: skills can include scripts that the agent can run for deterministic, repeatable steps.
  • Reducing context bloat: instead of loading everything up-front, the agent only loads the skill contents when needed.

Simon Willison has a good overview of why this pattern is powerful and how it compares to heavier integrations: https://simonwillison.net/2025/Oct/16/claude-skills/

For deeper technical context and best practices, see Anthropic’s engineering article: https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills

Where Skills live​

AiderDesk loads skills from these locations:

  • Project-level skills: .aider-desk/skills/ (in your project root)
  • Home-level skills: ~/.aider-desk/skills/
  • Built-in skills: bundled with AiderDesk
  • Extension-provided skills: extensions can supply their own skills

This lets you keep personal skills globally, while sharing project-specific skills with your repository. If multiple skills share the same name, the more specific location wins: extension skills take precedence over project skills, which take precedence over global (home) skills, which in turn take precedence over built-in skills.

Skill structure​

Each skill is a directory, typically shaped like:

{skills-root}/
  {skill-name}/
    SKILL.md
    references/
    scripts/
    ...

At minimum, a skill needs SKILL.md. It should start with YAML frontmatter containing at least a name and description.

Important prerequisites​

Skills only work when Skills Tools are enabled for the active agent profile.

You can enable them in either place:

  1. Agent Settings: Settings → Agent → (select your profile) → Use Skills Tools
  2. Per-task via AgentSelector: AgentSelector → Use skills tools

If skills exist on disk but Skills Tools are disabled, the agent won’t be able to discover or load them.

Using Skills in Agent Mode​

Enable “Use Skills Tools” in Agent Settings​

In Settings → Agent, enable Use Skills Tools for the agent profile you will use.

Agent Settings - Use Skills Tools

Enable “Use skills tools” in AgentSelector​

You can also toggle skills tools per-task from the Agent selector.

Agent Selector - Use skills tools toggle

Skills Tools​

When enabled, the agent can:

  • Discover installed skills (using skill metadata).
  • Activate a skill when it’s relevant by loading its SKILL.md and following its instructions.

In practice this means the agent can selectively load the right “mini-manual” for the job, instead of relying on generic behavior.

  • If you use skills frequently, keep their descriptions specific and action-oriented so the agent can select them correctly.
  • Prefer small, composable skills over one giant document.
  • Move rarely-needed details into additional files referenced from SKILL.md.

Security notes​

Skills can include executable scripts and external links. Treat them like code:

  • Install skills only from trusted sources.
  • If you didn’t write a skill, audit it before using it (especially scripts and any network access).
  • For shared repos, prefer keeping skills project-scoped so they travel with the codebase and are reviewable.

Further reading​