SuperAPI AI Directory

shared

General & other

View repository

Updated 10-16 02:03

About this skill

Shared library

Load only the reference or script needed by the current skill. This marker keeps the library discoverable as one installation unit; it does not guarantee identical loading behavior across hosts.

Resolve shared resources

A path such as shared/references/model-roster.md is relative to the skill collection, not the user's project directory.

  1. Use the loaded shared skill's directory when available.
  2. For a source checkout, use skills/shared/ under this repository.
  3. For a flat installation, find shared/ beside the installed skills. If it is missing, report the missing library dependency instead of guessing a path or silently skipping a contract.

Before a shell command, set SHARED_DIR to that resolved absolute directory in the same command. Other skills resolve their own scripts from their loaded skill directory. The scripts/skill_paths.py helper supports repository scripts that must find skills in both nested and flat layouts.

Contents

  1. model-routing.json: the single source for model identities, reasoning settings, and role routes. Use scripts/model_routing.py to inspect, resolve, or validate it.
  2. references/: workflow stages, model routing, runner behavior, evidence, handoffs, and run state. Every direct executable invocation first determines whether its branch dispatches a worker under references/model-preview.md. Coordinator work and deterministic commands use invocation authority; dispatch uses one concrete approved snapshot. Nested skills reuse that snapshot. references/engineering-rules.md owns mode selection.
  3. scripts/: runner discovery, jobs, output handling, and validation.
  4. tests/, schemas, and hooks: executable checks and shared formats.