[clavery]

Summary

Agents and users get more guidance for production triage and for moving work between instances, and failed job commands now say why they failed more often.

Observability

• New b2c-performance-triage and b2c-integration-triage runbooks, plus a metrics signals catalog. Together they map slow, erroring, or silent storefront and external-service symptoms to Metrics API categories and dimensions. The checkout, edge-traffic, and production triage runbooks now point to these signals.
• metrics_get (MCP) accepts summarize: true, which returns per-series statistics instead of every data point, so an agent can scan a whole category without hitting output limits. Its output is now compact JSON.
• Fix: Metrics API series tags (apiFamily, controller, host, ocapiCategory, statusClass) no longer include the <realm>_<environment> prefix. This is fixed in both the TypeScript and Python SDKs.
• MCP evals: routing cases for the new runbooks, and an opt-in, read-only live tier that runs against a real instance.

Skills

• b2c-replication: automate code and data replication from staging with a custom job built on the ExecuteCodeReplication / ExecuteDataReplication steps. Includes the task-group reference and the global-then-site order needed the first time a site is replicated.
• b2c-storefronts: choosing between creating and importing a Storefront Next storefront. Covers moving a storefront to another instance in the same MRT project (export, then job import --wait-for-storefront) and the import requirements. Cross-linked from b2c-job and b2c-custom-job-steps.

Job log diagnostics (shared by job run, job wait, job import, job export, job import-set, cap install, cap uninstall)

• If a failed execution hasn't flagged its log as written yet, the command re-checks the execution up to twice before giving up.
• [DATAERROR] lines from the log are printed next to the final error. The generic Internal fatal error summary usually hides the real cause, which those lines carry.
• When the log download fails, the warning now gives the reason (for example 401 Unauthorized) and a b2c webdav get --root logs ... --user-auth command that can read it. Before, it only said Could not retrieve job log.

Everything is best-effort: if any of these lookups fails, the output is the same as before.

Platform dependencies

• The Metrics API is in closed beta. Without access, metrics_get and the metrics steps in the runbooks return an authorization error.
• ExecuteCodeReplication requires B2C Commerce 26.11. The skill documents the preconfigured-process step for earlier releases.
• Storefront import with --wait-for-storefront requires 26.10 on both instances.

Testing

Manual testing:

• Run b2c job wait sfcc-site-archive-import <id> --show-log on a failed site import, using client credentials whose API client has no WebDAV access to Logs. The warning should include the 401 and the webdav get --user-auth command, and that command should download the log.
• With an API client that can read logs over WebDAV, a failed site import (for example, a storefront archive whose sites don't exist on the target) should show a Data errors: block.
• Follow b2c-replication on a staging instance: import the job definitions, then run the code, global, and site jobs in order.
• Follow b2c-storefronts to export a storefront from staging and import it on production. Production should get a new environment in the same MRT project.
• With Metrics API access, run metrics_get with summarize: true and confirm the tags have no realm prefix.

[github-actions]

📘 Docs preview

🧹 The preview for this PR has been removed (PR closed).