brainstorming
brainstorming
You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).
Explorar Skills
Desarrollo · SKILL.md
sampling-and-generation
sampling-and-generationValidate pix2code trained artifacts and generate DSL from screenshots using greedy or beam search workflows.
Independent GitHub snapshot · Automated classification · Original documentation governs
Documentación de la Skill
GitHubname: sampling-and-generation description: "Validate pix2code trained artifacts and generate DSL from screenshots using greedy or beam search workflows." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0
Sampling and Generation
Use this sub-skill when the user wants to turn one screenshot or a directory of screenshots into pix2code DSL output, or when they need to check that a trained artifact directory is complete before sampling.
Read first
- references/artifact-contract.md for the required files inside a trained model directory.
- references/inference-workflows.md for greedy versus beam search usage and the single-image versus batch scripts.
- references/troubleshooting.md for missing weights, mismatched metadata, OpenCV preprocessing issues, and stale wrapper problems.
- scripts/check_pix2code_artifacts.py to validate a trained-artifact directory before any sampling command.
Quick workflow
- Verify the artifact directory first:
python sub-skills/sampling-and-generation/scripts/check_pix2code_artifacts.py --artifacts bin
-
Sample a single screenshot when the artifacts are complete by adapting the legacy argument order described in
inference-workflows.md. -
Sample a directory of screenshots when you want a batch of
.guioutputs by using the same artifact-directory and search-mode conventions. -
If the task is only to diagnose missing files or stale artifacts, use the bundled checker and stop there.
Boundaries
This sub-skill owns artifact validation and screenshot-to-DSL generation guidance. It does not own dataset preparation; route that to ../data-and-training/SKILL.md. It does not own DSL compilation; route that to ../dsl-compilation/SKILL.md.
Validation checklist
- The artifact directory contains
pix2code.json,pix2code.h5,meta_dataset.npy, andwords.vocab. meta_dataset.npyholds the input shape and output size expected by the sampler.- The screenshot input exists and is readable by OpenCV.
- Greedy search is the default path; beam search is selected by passing an integer beam width.
- The generated DSL output is a
.guifile with the original tokens stripped of<START>and<END>.