Skip to content

Instantly share code, notes, and snippets.

@kingargyle
Last active March 14, 2026 20:38
Show Gist options
  • Select an option

  • Save kingargyle/6afdaa0f5641724e5ca9c1fafc9e8d83 to your computer and use it in GitHub Desktop.

Select an option

Save kingargyle/6afdaa0f5641724e5ca9c1fafc9e8d83 to your computer and use it in GitHub Desktop.
Skill Capability for SDD

Using the protocol in prompts/sdd_implementation_v1.md as a base, perform a "Phase 0: Agent Skills Integration" update to that file.

Update the document to Version 2.2.0 with the following mandatory architectural changes to support the agentskills.io standard:

  1. FORMAL AGENT SKILLS DISCOVERY (v2.2.0):

    • Protocol Update: Define a specific "Skill Discovery" phase for agents that do NOT natively support external skill loaders (e.g., Android Studio Otter, Gemini, Copilot).
    • Storage Standard: All skills must reside in prompts/skills/<skill-name>/.
    • File Structure: Each folder MUST contain a single SKILL.md file (the interface) and optional scripts/ or references/ directories.
    • The Discovery Step: At the start of every session, the agent MUST execute list_files on prompts/skills/ to identify available expertise.
  2. METADATA SCHEMA DEFINITION:

    • Explicitly define the required YAML frontmatter for SKILL.md so the agent knows exactly what to parse:
      • name: (String) The unique identifier for the skill.
      • description: (String) A concise summary of when this skill should be used. Used for "Task Matching."
      • compatibility: (List/Array) The agents or models this skill is tested for (e.g., [gemini-1.5-pro, claude-3.5-sonnet]).
      • version: (String) Semantic version of the skill.
  3. MODULAR "PULL-ON-DEMAND" REGISTRATION:

    • Metadata Extraction: For every discovered skill folder, the agent MUST read ONLY the YAML frontmatter of the SKILL.md file.
    • Context Economy: The agent is PROHIBITED from reading the body content (instructions, scripts) during discovery. This ensures the context window is preserved for the main task.
  4. ON-DEMAND ACTIVATION WORKFLOW:

    • Define the trigger logic: a. Task Match: When a user's request matches a skill's description field. b. Announce: The agent must state: "Activating Skill: [Name] from prompts/skills/[name]/." c. Load: The agent reads the full SKILL.md and any necessary assets in references/. d. Persona Shift: The agent adopts the specialized expertise and constraints defined in that skill's body content.
  5. COMPATIBILITY FOR NON-NATIVE AGENTS:

    • Add a specific section called "Manual Skill Emulation".
    • Mandate that for IDE-integrated assistants, this file-based discovery simulates the "plugin" architecture of Claude Code, allowing modular expertise without code-bloat.
  6. CLEANUP:

    • Maintain the existing "Memory Protocol" and "Phased GSD" requirements as the core execution framework.

Output: Provide the full, updated markdown content for prompts/sdd_implementation_v1.md.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment