SuperAPI AI Directory

b2c-production-triage

Coding & development

View repository

Updated 10-16 08:44

About this skill

Production triage

Connect the toolkit's logs, job executions, business data, and configuration into an evidence-backed assessment and next action. Investigation does not authorize remediation. Use the symptom routes below rather than collecting every available signal.

Scope and impact

  • Resolve the task's project/instance with config_inspect; CLI: b2c setup inspect. Preserve the resolved target across calls. Ask for missing site, time window, timezone, symptom, or expected behavior; never invent IDs or silently switch tenants.
  • Start with the reported interval; if unspecified, propose the last two hours and label that assumption. Use fixed UTC boundaries for comparisons and paging.
  • Record current impact (sites, checkout availability, order/data integrity, workaround) and the incident owner. Apply the team's severity policy; do not invent a priority or support SLA. Escalate active material impact early without waiting for root cause; agree who communicates and the next update time.

Access and tools

  • Establish access: MCP, terminal/CLI, logs, data, source. Operator work does not require a cartridge checkout. Credentials/setup issues: MCP mcp/b2c-config section setup-and-access; native b2c-config/b2c-auth, or the authentication guide.

Prefer dedicated MCP tools; discover available names once. Read skill://mcp/scapi/SKILL.md before using scapi_search/scapi_execute, as those tools require. Read only the matching runbook/reference, not every linked skill. Native assistants use the installed skill name; MCP can read these URIs or use skills_read with collection: "b2c-ops" and a task query.

Symptom Toolkit path Decision / detail
Error spike or integration errors logs_list_files -> returned prefix -> logs_get_recent with count: 20, since, and a relevant search Defaults cover error/customerror, not every custom-* integration log. CLI: b2c logs list / b2c logs get.
Failed job or stale data Code mode codemode.describe("builtin/failed-job-triage"), then run with fixed from/to -> execution IDs and logFilePath -> log tools Job health covers continuation, step details, exact WebDAV logs, and green-but-incomplete runs.
Checkout failures or orders down Custom logs -> determine whether failures produce order records -> relevant data/traffic evidence Checkout triage distinguishes checkout symptoms from failed-order records.
Find/count FAILED orders or investigate affected products Verify Admin filter support; otherwise external OCAPI Shop order search -> safe IDs -> logs and selected Admin product reads Failed-order triage owns enumeration, count/sample limits, and the workaround. No OCAPI helper exists inside code mode.
Suspected code change Code mode builtin/code-version-inspect: active/rollback versions and activation/modification timestamps Current metadata is not a complete activation history. Corroborate with release records.
Runtime-only defect MCP debugger on an authorized reproduction target Operator handoff is valid without source access. Do not halt production requests for routine triage.

Checks and decisions

  1. Carry the returned configuration resolution into the evidence record. config_inspect stays masked; do not read dw.json or request tokens merely to make managed MCP calls. Pass the same project/instance context to CLI fallbacks.
  2. Join signals using execution IDs, log timestamps, and affected record IDs. For exact job logs, pass the returned logFilePath to webdav_get; use byte continuation or download locally. Recent-log tools remain preferred for filtered samples/watches. A bounded result is not full-interval coverage.
  3. Reuse code-mode snippet totals/window/nextOffset; do not return entire schemas or records. The failed-job snippet omits step details and healthy runs: use builtin/job-execution-inspect / builtin/job-execution-review selectively. Compare expected business data with a healthy input before calling a green job healthy.
  4. Preserve failed stages: HTTP ok/status and diagnostics, or thrown auth/transport/safety errors. An OCAPI fallback error may mask an earlier SCAPI rejection; CLI --api-backend scapi isolates it. POST searches can be blocked by Safety Mode despite being reads. Do not infer missing scopes from every 403 or treat unavailable counts as zero.

Mitigation and recovery

Recommend the smallest supported mitigation, its blast radius and rollback limit. The authorized owner decides; inspecting an incident is not permission to change it. Restore service when justified without waiting for a full causal explanation. After intervention, recheck business behavior and comparable signals over an agreed observation interval. Report recovered, partial, pending, or unknown; record follow-up prevention separately from immediate recovery.

Escalation and handoff

Report: target/site; UTC window; observed impact and denominator; evidence locations (execution IDs, log path/timestamps, minimal excerpts); suspected component/input/change; alternative explanations; access gaps; next owner/action. Exclude tokens, payment details, and unnecessary customer data.

Route custom-code evidence to the developer/implementation partner, grant or schedule issues to the administrator, provider failures to the integration owner, and suspected platform failures to Salesforce Support. Read escalation and case template when preparing a handoff. Do not submit a case/message without authorization or claim a confirmed cause when evidence establishes only a lead.