| name | shared |
| description | Shared resources for universal-creator generator skills: prompting technique registry (techniques.json), the technique-selector subagent, the 18 worked prompt-technique examples (examples/), and the reusable agent trio (agents/) — validation-reviewer, artifact-router, prompt-strategist — plus the shared memory-guardrails snippet. Installed automatically when any dependent generator skill is installed. DO NOT USE directly — consumed by agent-generator, skill-generator, hook-generator, and prompt-generator. |
| license | MIT |
Shared Resources
Shared assets used by every universal-creator generator skill. This directory
is dependency-only: it ships when any generator skill that declares
dependencies: [shared] is installed.
Contents
techniques.json — Registry of all 18 prompting techniques with
structural signals, use case mappings, and example snippets.
technique-selector.md — Subagent that classifies a generation request
into one of 5 use case categories and returns the technique to embed plus
concrete structural requirements.
examples/ — 18 worked .prompt.md files, one per technique (zero-shot,
few-shot, cot, react, reflexion, rag, self-consistency, generate-knowledge,
tree-of-thoughts, ape, active-prompt, dsp, pal, multimodal-cot,
graph-prompting, meta-prompting, prompt-chaining, art). Read at runtime by
any generator that needs to apply or reference a technique example.
agents/ — Reusable agent definitions: validation-reviewer.agent.md,
artifact-router.agent.md, prompt-strategist.agent.md, and the shared
_memory-guardrails.md snippet they all link to. These are sync-fanned in
from the canonical .github/agents/ source by scripts/sync_agents.py;
do not hand-edit.
scripts/ — Canonical helper scripts (package_skill.py,
quick_validate.py, run_eval.py, run_loop.py, utils.py, etc.) used
by every generator skill's scripts/ directory; kept in parity via
universal-creator sync.
Usage
Generator skills (agent-generator, skill-generator, hook-generator,
prompt-generator) spawn technique-selector.md as a subagent during their
workflow to determine which prompting technique to embed. Agents and examples
are referenced by relative path from the consumer skill (e.g.
../shared/examples/cot.prompt.md, ../shared/agents/validation-reviewer.agent.md).
The techniques.json registry is read by technique-selector at runtime —
it is the single source of truth for all technique definitions.
Edit protection
The CLI installer (universal-creator install ...) writes a per-file SHA256
manifest at <scope>/.universal-creator/shared.lock on first install. On
reinstall, files whose on-disk hash diverges from the manifest are preserved
and skipped with a warning. Pass --force-shared to override the protection
and reset the directory to bundled content.