Agent Skills
› NeverSight/learn-skills.dev
› image-pipeline
image-pipeline
GitHub用于规范图像流水线中RAW解码与色彩处理流程的变更实施。涵盖操作顺序调整、新参数引入及视觉回归测试，确保非破坏性编辑与色彩一致性，防止源文件篡改及色域假设冲突。
触发场景
更改RAW解码流或编辑操作顺序
引入新的图像调整或调优默认值
诊断色彩偏移、裁剪、带状伪影或高光恢复问题
安装
npx skills add NeverSight/learn-skills.dev --skill image-pipeline -g -y
SKILL.md
Frontmatter
{
"name": "image-pipeline",
"description": "Use when implementing or validating RAW decode and color\/edit processing order, including non-destructive parameter application and visual regression checks."
}
Image Pipeline
Use This Skill When
- Changing RAW decode flow or edit operation order.
- Introducing new image adjustments or tuning defaults.
- Diagnosing color shifts, clipping, banding, or highlight recovery issues.
Required Inputs
- Pipeline stage being changed.
- Expected visual behavior and tolerance.
- Test image set or sample corpus path.
Workflow
- Document exact processing order before changes.
- Implement change in one stage at a time.
- Validate with golden images or reference snapshots.
- Confirm edits stay parameter-only (non-destructive).
- Record known visual deltas and rationale.
Guardrails
- Do not mutate source RAW files.
- Keep parameter mapping stable across preview and export.
- Reject mixed color-space assumptions across stages.
Output Contract
- Pipeline code change and updated stage documentation.
- Regression comparison notes (pass/fail per test image).
版本历史
- c3c0a1e 当前 2026-07-23 10:50
