LEARN. PRACTICE. GROW.Concept and mechanism
Caches and artifacts solve different needs. A cache reduces repeated dependency preparation work; a partial match requires reconciling state with current requirements. An artifact carries run results and can support promotion when associated with candidate identity and approval. Retention should cover the operational window when those results are needed. A binary rebuilt after the original expires is a new candidate until equality is established. Include the application commit, effective external references, result digest, and decision evidence in the release record. This gives the next support shift a usable starting point when behavior differs between attempts.
Guided application
In a fictional incident after a rerun, compare the reusable workflow too: rerunning all jobs can resolve a moving tag again; rerunning a specific or failed job preserves the first attempt’s SHA. Caller env does not automatically cross that contract, so inputs and outputs should be explicit. Templates create initial files and copies need maintenance in consumers. To suspend a faulty workflow, disable new triggers and retain investigation history. When collecting logs, avoid printing secrets, especially transformed values that automatic redaction may not recognize. Tell the manager the observed cause, confirmed impact, and what remains unproven.
Same application commit, different moving reusable tag: two runs can have different inputs.
Common pitfalls
Cache as approval; rebuild as identical bytes; equivalent reruns; template as synchronization; redaction as guarantee.
Related topics: Events, filters, and dependencies · Data, outputs, and service networking · Actions with contracts and versions
Preserve evidence and compare relevant inputs before attributing cause.
Reference: Workflow artifacts · GH-200 skills measured January2026;study guide updated2026-02-05