Five prompt patterns
Turn vague asks into useful work prompts.
Be specific: name the artifact (e.g., PR #1247 /login refactor, q3-dashboard DAX measures, API-outage runbook), the risk (e.g., SQL injection, filter-context bugs, missing paging step), and the output shape (3-line summary + risks + tests, table, diff).
Give it a role: make Claude reason from a real professional lens (e.g., senior backend reviewer on the Car-Mart API team, SRE on the dealer-portal on-call rotation, data analyst on the Snowflake warehouse) - not a generic assistant.
Show your work: force visible reasoning before the answer (suspect code + symptom + ranked causes; or which dashboards + queries + assumptions; or which 3 axes the refactor improves).
Iterate: treat the first answer as a draft. Three iteration shapes: refine the format (table columns), cut the cruft (strip boilerplate, match team template), expand the depth (add the math, cite the source).
Save the prompt: when a prompt produces a useful answer, move it to a Project's prompts/ folder with title (e.g., [pattern: code-review-security]), tags (e.g., code-review, security, PR), and a copy-pasteable body.
Act as a senior backend reviewer for the Car-Mart API team. Review this PR for the /login endpoint refactor. Walk me through your reasoning. Output: Summary / Risk / Test Plan / Rollback. Then extract the prompt pattern I should reuse.In your lane
Use the code-review pattern on the next PR you review.
Use the SQL-query pattern on the next query you write.
Use the incident-triage pattern on the next alert you investigate.
Try one pattern on a real ticket before lunch.