Epic: Agent Skills - Standardized Expertise & Progressive Disclosure

[mattKorwel]

Agent Skills allow users to extend Gemini CLI with specialized expertise, procedural workflows, and task-specific resources. Based on the Agent Skills open standard, a "skill" is a self-contained directory that packages instructions and assets into a discoverable capability.

Unlike general context files (GEMINI.md), which provide persistent background, Skills represent on-demand expertise. This allows Gemini to maintain a vast library of specialized capabilities—such as security auditing, cloud deployments, or codebase migrations—without cluttering the model's immediate context window.

Key Objectives:

• Adhere to the Agent Skills Standard: Support the SKILL.md format, including YAML frontmatter for discovery and Markdown body for procedural guidance.
• Implement Progressive Disclosure: Utilize a three-tier loading strategy:
  1. Discovery: Only metadata (name/description) is visible to the model initially.
  2. Activation: The full instruction set is loaded upon autonomous model request.
  3. Execution: Supporting assets (scripts, references) are made available once the skill is active.
• Tiered Resolution: Support Project-level (.gemini/skills), User-level (~/.gemini/skills), and Extension-provided expertise.
• Security-First Lifecycle: Ensure all skill activations and extension-provided expertise require explicit user consent through the policy engine and installation flows.