Skip to content

skill-template.md

Latest commit

 

History

History
56 lines (39 loc) · 1.47 KB

skill-template.md

File metadata and controls

56 lines (39 loc) · 1.47 KB
Raw

Skill Template

Place the skill at .claude/skills-impl/<layer>/<scope>/<skill-name>/SKILL.md. The skill name MUST match the parent <scope>/ directory:

  • torch/quark-torch-*
  • onnx/quark-onnx-*
  • shared/quark-* (no infix)

For user-facing skills, also add a thin entry stub at .claude/skills/<skill-name>/SKILL.md. The stub's description MUST include backend keywords so the harness can route by prompt-match (torch: PyTorch / HuggingFace safetensors / transformers; onnx: .onnx / onnxruntime).

---
name: <skill-name>             # matches the <scope> infix rule above
description: <trigger phrases + backend keywords>
layer: <l0-foundation | l1-atomic | l2-workflows | l3-recipes | meta>
backend: <shared | torch | onnx>   # must equal the parent <scope>/ directory
primary_artifact: <artifact file name or report name>
source_knowledge:
  - <Quark upstream doc or source file>
---

Purpose

One paragraph describing the stable responsibility boundary of the skill.

Inputs

  • required inputs
  • optional defaults

Outputs: <primary_artifact filename>

Schema: <primary_artifact>.schema.json

{ "...": "example matching the schema" }

Interaction Flow

  1. Intake
  2. Route
  3. Plan
  4. Confirm
  5. Execute or Summarize

Recovery

  • common failure modes
  • recovery advice

Notes

  • traceability to upstream docs, CLI, or code
  • guardrails that should not be silently skipped