菜单
Diagnostic checklist for a failing build — isolate which stage broke. Use when a build errors out, when validation fails, when fetch returns 403, or any time the user needs to triage a pipeline failure.
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
Walk through adding a new dataset to sources.json and producing its first build. Use when the user wants to onboard a new upstream source, add a slug to the manifest, or extend the catalog.
Walk through writing a new transform handler under scripts/pipeline/handlers/. Use when the default tighten_types/identity paths can't produce the right shape — row-level JSON parsing, streaming to avoid OOM, multi-output splitting, or VARIANT-from-the-start.
Load a Raincloud dataset via the lightweight Python loader API (`raincloud.load(slug)`). Use when the user wants to read a prepared parquet/vortex artifact, inspect a slug's metadata, or pull a dataset for downstream analysis from cache → mirror → local build.
Push locally-built Raincloud artifacts to a mirror (`scripts.pipeline.publish`). Use when the maintainer wants to upload one or more slugs' parquet/vortex bytes to a configured mirror after a successful build, gated on the snapshot's recorded sha256.
Use when the user asks to compute or refresh per-column statistics for a raincloud dataset — produces `outputs/v1/<slug>/profile.json` for the TUI's detail pane and `list_datasets --inspect`.
Run the full Raincloud pipeline (fetch → extract → parse → transform → write → validate → convert) for one or more dataset slugs. Use when the user asks to build a dataset, rebuild a slug, or process a batch.
| name | raincloud-debug-build |
| description | Diagnostic checklist for a failing build — isolate which stage broke. Use when a build errors out, when validation fails, when fetch returns 403, or any time the user needs to triage a pipeline failure. |
| argument-hint | <slug> |
Triage a failing /raincloud-build $ARGUMENTS. Reference: SKILLS.md "Debugging a failing build".
Walk these in order — stop as soon as the cause is found:
Is it a row-count mismatch in validate? Re-run with --loose to convert the error to a warning:
python -m scripts.pipeline.build $ARGUMENTS --loose
If the build now succeeds, update expect.rows in sources.json to the actual count.
Isolate the stage — invoke each independently to see exactly where it breaks:
python -m scripts.pipeline.fetch $ARGUMENTS
python -m scripts.pipeline.extract $ARGUMENTS
(parse, transform, write, validate, convert are not standalone CLIs — once extract runs cleanly, fall through to /raincloud-build.)
Check the raw cache at outputs/raw_downloads/<slug>/. Fetch is idempotent and skips when expected_bytes / expected_sha256 already match. If you suspect a stale or corrupt download, rm -rf outputs/raw_downloads/<slug>/ and re-fetch.
Check _workdir/<slug>/ — extract stage output. If a handler complains "no .xxx files", look here first to see what was actually unpacked.
DuckDB OOM / swap thrash — cap memory and redirect spill:
RAINCLOUD_DUCKDB_MEMORY_LIMIT=8GB \
RAINCLOUD_DUCKDB_TEMP_DIRECTORY=/mnt/scratch/duckdb-tmp \
python -m scripts.pipeline.build $ARGUMENTS --loose
Default DuckDB memory_limit (~80% of system RAM) can swap-thrash on heavily-nested VARIANT shredding. See README.md "DuckDB resource limits".
Kaggle 403 — see /raincloud-add-kaggle-tos. The fetch error message points at the URL to click through.
Vortex panic on convert — known type-support gaps in vortex 0.69 (FixedSizeBinary(16), VARIANT round-trip blow-up). See /raincloud-convert and SKILLS.md.
When in doubt, prefer Read → Grep over guessing. The pipeline has hidden contracts (streaming handlers returning [], raw_downloads being unversioned, VARIANT requiring v1.5.0) that aren't obvious from any single file — see AGENTS.md.