Package completed data analysis and ML work so an independent recipient can reproduce the claimed results, verify artifact lineage, and operate the handoff within its stated scope. Use when finalizing a project, study, model package, or review bundle; not for deploying to a live system.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Ship an evidence-bearing result, not merely code that once ran. The package must connect the approved question to immutable inputs, an executable workflow, generated outputs, validation evidence, and an explicit handoff boundary.
Before packaging, record:
Do not silently expand a validated result from a sample to a population, from retrospective analysis to prospective use, or from one environment to another.
Make each authoritative output traceable through:
source identity -> immutable snapshot or query -> transformation -> configuration
-> code revision -> environment -> model artifact -> evaluation -> published output
Record content hashes for immutable local artifacts and durable identifiers for remote sources. Preserve raw inputs when permitted; otherwise preserve the exact query, retrieval time, schema, row counts, access requirements, and a safe fixture sufficient to test the pipeline.
Separate source data, intermediate artifacts, final outputs, caches, and human edits. Never require a recipient to guess which file is canonical.
Provide one documented entry point from a clean state. Pin direct dependencies and runtime versions; record relevant system libraries and hardware. Seed every controllable source of randomness and document remaining nondeterminism rather than promising bitwise equality when the stack cannot provide it.
Packaging does not authorize restricted-data access, material compute, credential use, external transfer, or publication. Verify the relevant authority before those actions. When execution is not authorized or available, prepare the reproducible contract and label the unexecuted stages instead of running them.
Define acceptable reproduction precisely:
Fail loudly on missing inputs, schema drift, stale caches, or incompatible artifacts.
Run from the documented clean entry point without relying on notebook state, shell history, undeclared credentials, or files outside the package. Verify every material claim with evidence of matching scope.
A green unit test is not proof of complete reproduction. The completion matrix must distinguish:
If a full rerun is impossible because data, compute, or credentials are unavailable, ship the maximum verifiable subset and mark the unexecuted stages as unproven. Do not call the package fully reproducible.
Document how to run, verify, interpret, update, and safely retire the result. Include known limitations, prohibited uses, external dependencies, expected costs, credential ownership, failure signals, and escalation contacts. Preserve rejected hypotheses and failed experiments when they affect future maintenance or interpretation.
Use assets/reproduction-handoff.md as the release document and assets/run-manifest.yaml as a machine-readable starting point. For verification levels, lineage requirements, and clean-room acceptance rules, read references/reproduction-contract.md.
Release only when the package contains no secrets, unintended or unauthorized private data, machine-specific absolute paths, undeclared dependencies, or ambiguous canonical artifacts. Keep approved restricted artifacts in their authorized secure location and publish only permitted references or sanitized fixtures. External sharing and public release require separate authority. The final status must be one of:
State the status prominently. Never upgrade it based on intent or partial checks.
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Audit datasets, joins, labels, and ground truth before analysis or modeling. Use when data meaning, row grain, time semantics, source-of-truth reliability, or label construction may invalidate conclusions; not for general model evaluation after the evidence base is already trusted.
日本語の概要は準備中です。原文の説明を表示しています。
Design leakage-safe machine learning experiments that mirror real deployment and support fair model comparisons. Use when defining prediction timing, feature eligibility, train-validation-test splits, baselines, metrics, or controlled model iterations; not for auditing whether raw labels are trustworthy.
日本語の概要は準備中です。原文の説明を表示しています。
Isolate the root cause of ML performance drops, inconsistent evaluations, prediction errors, and training-serving mismatches across data, labels, splits, pipelines, models, metrics, and runtime behavior. Use when investigating a reproducible failure or regression, not routine model selection or general performance validation.
日本語の概要は準備中です。原文の説明を表示しています。
Orchestrate an end-to-end data analysis or machine learning project from decision framing through reproducible handoff. Use when a request spans multiple lifecycle stages or an ambiguous modeling request must become a decision-ready result; use narrower audit or experiment-design skills for isolated reviews.
日本語の概要は準備中です。原文の説明を表示しています。
Route data analysis and machine learning work to the right skill in this suite. Use when starting any analysis, modeling, validation, or reproducibility task and the matching specialized skill is not yet clear; not needed when one specific skill already clearly applies.
日本語の概要は準備中です。原文の説明を表示しています。
Validate trained models and analytical claims against their intended decision, independent evidence, and human-reviewed ground truth. Use when reviewing model performance, analysis conclusions, launch claims, or evaluation reports; use failure diagnosis instead when the main task is locating a known defect.
日本語の概要は準備中です。原文の説明を表示しています。