salesforce-performance-tuning

v2026.09.24

Analyze and tune Salesforce query, pagination, batching, caching, automation, and asynchronous performance using measured org evidence. Use when resolving latency or throughput problems. Trigger with "tune Salesforce performance".

GitHub
安装命令
npx skhub add jeremylongshore/salesforce-performance-tuning
Markdown
SKILL.md

Evidence-Driven Salesforce Performance Tuning

Overview

Optimize the end-to-end workload without trading correctness, permission enforcement, shared capacity, or recovery for a synthetic benchmark.

Prerequisites

  • Representative workload, latency and throughput objectives, business invariant, and current telemetry
  • Queries, object volumes, selectivity, pagination, automation, locks, API mode, limits, and downstream timing
  • Authorized performance org or sandbox, synthetic dataset, stop thresholds, and rollback owner

Tool Discipline

Use Read, Glob, and Grep to inspect approved repository and evidence files, WebFetch to re-check current first-party Salesforce documentation, and Write or Edit only for secretless plans, fixtures, configuration, and redacted receipts.

Current Contract

Salesforce offers versioned query and limits resources; query performance feedback through REST explain is explicitly Beta. Query plans, indexes, object cardinality, automation cost, and limits depend on the target org and must be measured.

Authentication

Use an approved read-only performance principal first and a separately approved principal for canary writes. Do not collect broad record payloads or use administrator access to mask real permission cost.

Instructions

  1. Define the business invariant and measure end-to-end latency, throughput, error rate, API consumption, locks, retries, and downstream lag.
  2. Trace time across client, network, authentication, query, pagination, automation, async jobs, event delivery, and reconciliation.
  3. Inspect query shape, selected fields, predicates, cardinality, selectivity, ordering, pagination, and current metadata.
  4. Use documented explain or query-plan tooling only under its current status and corroborate recommendations with measured tests.
  5. Compare bounded changes to query design, caching, conditional requests, composite or bulk modes, concurrency, and scheduling.
  6. Canary one change at a time with synthetic or approved non-production data and fixed stop thresholds.
  7. Verify correctness, permissions, limits, event and business outcomes; roll back regressions and publish the evidence.

Approval Boundaries

Do not request indexes, change automation, increase concurrency, add caches, alter consistency, or load production for a benchmark without owners.

Output

Return the baseline trace, bottleneck evidence, tested alternatives, capacity effect, correctness checks, canary result, rollback, and recommendation.

Error Handling

ConditionResponse
Explain output is treated as production proofLabel it Beta guidance and require representative measured validation.
Optimization changes returned recordsReject the change and restore the original business invariant.
Faster client behavior consumes unsafe shared capacityApply backpressure or scheduling before increasing rollout.

Example

A redacted completion receipt might look like this:

workload=account-sync; baseline_p95=recorded; bottleneck=query; change=selective-filter; limits=safe; correctness=exact; canary=pass

Resources

Next Steps

Run the workflow first in the lowest-risk authorized org and preserve its redacted receipt. Schedule a review against the next Salesforce seasonal release and the customer change calendar.

发现
标签

此技能尚未发布标签。

版本
最新版本元数据

版本

v2026.09.24

发布时间

2026年9月24日

分类

未分类

许可证

MIT

源路径

skills/.curated/salesforce-performance-tuning

默认分支

main

最新提交

e5a6c3b

Tree SHA

c2dc8e8