Ski Output Claude Code

Claude Code Platform Output

Takes a generated GA entity (.agents/) and produces the Claude Code equivalent (.claude/), applying all platform-specific transformations.

Input / Output

Input:

• entity_type: workflow | agent | command | skill | rule | knowledge-base | resources | script | hook
• entity_name: the entity filename (e.g. age-spe-email-classifier.md)
• source_content: the full content of the generated GA entity
• export_path: base export path (e.g. exports/{name})
• model_ga: the GA model value from frontmatter (e.g. gemini-3.1)
• settings_json: current state of the system's settings.json (for hook appending)

Output:

• Entity file written to the correct .claude/ path
• Updated settings.json if entity is a Hook
• Summary of what was generated

Procedure

1. Determine destination path

Read the directory mapping from ../resources/res-entity-templates-claude-code.md §5:

2. Transform content based on entity type

For Workflows, Agents, Commands (behavioral entities)

Path transformations:

• Replace ./age-spe- with ../agents/age-spe- in all references
• Replace ./age-sup- with ../agents/age-sup- in all references

Frontmatter injection (Agents only):

• Map model value using default mapping: gemini-3-flash → haiku, gemini-3.1 → sonnet. Valid CC values: opus, sonnet, haiku. Note: model and effort defaults may be overridden later by the user in §7.5 Model & Effort Selection of the entity builder.
• Add effort field if specified (valid values: max, high, medium, low). max is only available on opus and sonnet. Controls reasoning depth.
• Add CC-specific fields if specified in the handoff JSON: tools, disallowedTools, permissionMode

Protection: Do NOT transform paths inside:

• Code blocks documenting architecture of generated systems
• Generic templates/placeholders with [name] brackets
• Paths that already use ../agents/ format

For Skills

• Direct copy — content is identical on both platforms
• Verify subdirectory structure: ski-[name]/SKILL.md (never flat file)

For Rules, Knowledge-base, Resources

• Direct copy — content is identical on both platforms

For Scripts

• GA format: .md procedural document
• CC format: .sh or .py executable
• Extract the executable logic from the GA .md and generate the corresponding script file
• Preserve the same frontmatter in a companion documentation file if needed

For Hooks

• Copy the .md documentation file to .claude/hooks/
• Generate the settings.json hook entry:
  1. Read event and matcher from the hook frontmatter
  2. Read platform_behavior.claude_code for the entry structure
  3. Append to the hooks key in settings.json, grouped by event type

3. Write the output file

Write the transformed content to {export_path}/.claude/{destination_path}.

4. Return summary

CC output: {entity_name} → .claude/{destination_path}
  Transforms: [list of transforms applied, e.g. "path transforms", "model mapping", "settings.json hook"]

Error Handling

• Destination file already exists: Overwrite silently (the GA version is the source of truth).
• Unknown entity type: Skip and warn — do not block the pipeline.
• Model value not in mapping table: Use sonnet as default and warn.
• Hook without event field in frontmatter: Skip settings.json entry and warn.