AGENT SKILL

prompt-automation

Configure YiVal data generators and prompt variation generators for synthetic examples, document questions, chain-of-density prompts, and self-exemplar prompts.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
prompt-automation
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Configure YiVal data generators and prompt variation generators for synthetic examples, document questions, chain-of-density prompts, and self-exemplar prompts.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

YiVal prompt automation

Use this sub-skill when the task involves generating input examples, generating prompt variations, caching generated prompts/data, using yival gen/yival task, or wiring OpenAI/document/chain-of-density/self-exemplar generators into a YiVal config.

Read first

Useful helper:

  • python sub-skills/prompt-automation/scripts/inspect_prompt_components.py prints built-in generator ids and default configs after importing their modules.

Built-in generator map

Need YAML location Registry id
Generate synthetic input examples from a function/task description dataset.data_generators openai_prompt_data_generator
Generate questions from text/file/Google Drive document content dataset.data_generators document_data_generator
Generate optimized prompt strings with OpenAI variations[].generator_name openai_prompt_based_variation_generator
Use the fixed chain-of-density summarization prompt variations[].generator_name chain_of_density_prompt_generator
Use a self-exemplar prompt template variations[].generator_name self_exemplar

Minimal auto-prompt shape

description: Auto prompt experiment
custom_function: my_task.headline
dataset:
  source_type: machine_generated
  data_generators:
    openai_prompt_data_generator:
      number_of_examples: 2
      chunk_size: 1000
      model_name: gpt-4
      output_path: generated_examples.pkl
      input_function:
        name: headline_generation
        description: Generate a landing page headline for a tech startup business.
        parameters:
          tech_startup_business: str
variations:
  - name: task
    generator_name: openai_prompt_based_variation_generator
    generator_config:
      number_of_variations: 2
      model_name: gpt-4
      output_path: generated_prompts.pkl
      variables:
        - tech_startup_business
      prompt:
        - role: system
          content: Write concise prompts for GPT-4.
    variations:
      - value_type: str
        value: Generate landing page headline for {tech_startup_business}
        instantiated_value: Generate landing page headline for {tech_startup_business}
        variation_id: null

The custom function should use StringWrapper(name="task", variables={"tech_startup_business": tech_startup_business}, state=state) so generated/manual prompt values are applied.

Workflow guidance

  1. Start with a tiny run: one or two generated examples and one or two variations.
  2. Always set output_path for provider-backed generators so reruns load cached pickles instead of paying/calling again.
  3. Keep variables explicit for prompt variation generation; generated prompt outputs must contain {variable} placeholders when configured.
  4. For document generation, start with source: text before file or Google Drive sources.
  5. Route generated outputs into evaluation-optimization for evaluator, selector, and enhancer design.

Safety and cost boundaries

  • OpenAI generators call network services and can bill. Confirm credentials and budget before running.
  • Document file/drive sources can invoke loaders, OCR, Google APIs, and local file reads. Use only approved documents.
  • Retrieval demos that use embeddings/FAISS/OpenAI are reference patterns, not default runtime paths.
  • If the user only needs a static prompt list, use manual variations instead of provider-backed generation.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351