菜单
Add a new data-transformation shaper to RING-5's pipeline (subclass Shaper, add a discriminated-union TypedDict config, register in ShaperFactory with a camelCase key, add the UI config). Use when the user wants a new transform/shaper/pipeline step.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Add a new plot/chart type to RING-5 (subclass BasePlot, emit engine-agnostic traces, register in PlotFactory, wire config UI, test). Use when the user asks to add or modify a visualization/chart/plot type.
Gate the ApplicationAPI facade contract — the single Web→Core entry point, its three sub-APIs, and the async parse/scan protocol. Use when verifying API usage, reviewing facade/service changes, or checking that the UI talks to core only through the facade.
Gate the RING-5 architecture — the 7 hard rules and the strict 3-layer dependency direction (Web → Core ← Parsing). Use before merging, after any structural/refactor change, or when asked to "check architecture", "run the arch gate", or verify layer boundaries.
Sweep code and docs for dead, duplicated, or stale paths/symbols and remove them — repoint drifted paths, delete docs for removed subsystems, and enforce canon-only docs (no "was X, now Y" history). Use when asked to clean up stale references, dead paths, doc debt, or legacy shims/aliases.
Write or debug RING-5's Streamlit UI tests — the no-browser AppTest/ui_logic/ui_unit layers and the Playwright browser e2e/visual suite (Page Object Model, data-testid selectors, and Streamlit-specific gotchas). Use for any UI test, Playwright, AppTest, screenshot, or flaky-UI-test task.
Gate the gem5 parsing/scanning subsystem for correctness — no data fabrication, async-only API, the StatType lifecycle invariant, Perl regex duplication, and worker-pool lifecycle. Use to verify parser/scanner/Perl/stat-type changes before merging (complements the parsing-and-variable-types build skill).
| name | add-shaper |
| description | Add a new data-transformation shaper to RING-5's pipeline (subclass Shaper, add a discriminated-union TypedDict config, register in ShaperFactory with a camelCase key, add the UI config). Use when the user wants a new transform/shaper/pipeline step. |
Shapers are the no-code transform steps. They are stateless, immutable, composable: each takes a
DataFrame and returns a new DataFrame (result = data.copy() first; never inplace=True). They
live entirely in Layer B (src/core/services/shapers/) — no Streamlit there.
Best starting move: copy the closest existing impl in
src/core/services/shapers/impl/ (e.g. sort.py, normalize.py, mean.py) and adapt.
Implement — src/core/services/shapers/impl/<name>.py:
from src.core.services.shapers.shaper import Shaper # or UniDfShaper for single-frame ops
class MyShaper(Shaper):
def _verify_params(self) -> bool: # static validation of self params (ctor-time)
...
def _verify_preconditions(self, df: pd.DataFrame) -> bool: # data-dependent validation
...
def __call__(self, df: pd.DataFrame) -> pd.DataFrame:
result = df.copy()
# ... transform ...
return result
The shaper is constructed with its config dict: shaper_class(dict(params)). NaN-handling
convention: mirror impl/mean.py's _safe_gmean/_safe_hmean (dropna, reject non-positive,
return np.nan on edge cases) if doing aggregation.
Typed config — add a MyShaperConfig(BaseShaperConfig) TypedDict to
src/core/models/shaper_models.py, and add it to the ShaperStepConfig union. The type
field is the discriminator.
Register in src/core/services/shapers/factory.py:
ShaperFactory._registry with a camelCase key (e.g. "myShaper");_display_names.Validation — if it needs extra checks, extend src/core/services/shapers/validation.py.
UI config — add src/web/components/shapers/<name>_config.py rendering the widgets and
returning the config dict; wire it into src/web/pages/ui/shaper_config.py /
selector_transformer_configs.py dispatch. Per-step unique keys: f"{prefix}_{field}_{shaper_id}".
Tests — unit in tests/unit/ (transform correctness + both _verify_* paths), plus a
pipeline/integration test if it interacts with other steps. Import path is
src.core.services.shapers.impl.<name> (NOT src.web...).
.copy()), no inplace=True, no bare except.pipeline_service.py which
fingerprint-caches results — keep __call__ pure (no hidden state).ShaperFactory._display_names) — don't duplicate them.make arch-check && ./python_venv/bin/mypy src/ && make test-unit.