Answer: First agree on the agent, one to three fixed inputs, the metric to improve, and the behavior that must not change; then run the agent the way a user would on saved or local inputs without deploying or posting live, read the trajectory and name the main failure in one sentence, change one lever, rerun the same fixtures, compare correctness, efficiency, and contract fit, and keep, revert, or narrow the change. Status: Confirmed Entry: tool/cursor/agents-and-automation/run-one-measured-round Revision: 1 Digest: sha256:fd32c80e89a10e77 (changes whenever the content changes) Timeline: Created: 2026-11-13T03:08:41Z Last changed: 2026-11-13T03:08:41Z Last verified: 2026-11-20T12:33:20Z Next check: 2026-11-21T12:33:20Z Check interval: Daily Expected change: No fixed schedule Data: Field | Value | Source | As of fixed inputs | one to three fixed inputs | [1] | 2026-11-20 step 1 | Run the agent the way a user would, on saved or local inputs, without deploying or posting live. | [1] | 2026-11-20 step 2 | Read the trajectory and name the main failure in one sentence. | [1] | 2026-11-20 step 3 | Change one lever. | [1] | 2026-11-20 step 4 | Rerun the same fixtures. | [1] | 2026-11-20 step 5 | Compare correctness, efficiency, and contract fit. | [1] | 2026-11-20 step 6 | Keep, revert, or narrow the change. | [1] | 2026-11-20 warning | Don't swap a fixture mid-round, or the comparison means nothing. | [1] | 2026-11-20 Sources: [1] Anysphere. "cursor.com/docs/agent-sdk/hillclimbing.md". https://cursor.com/docs/agent-sdk/hillclimbing.md Role: primary. Accessed: 2026-11-20T12:33:20Z Quote: "First agree on the agent, one to three fixed inputs, the metric to improve" (Supports: fixed inputs) Quote: "Run the agent the way a user would, on saved or local inputs, without deploying or posting live." (Supports: step 1) Quote: "Read the trajectory and name the main failure in one sentence." (Supports: step 2) Quote: "Change one lever." (Supports: step 3) Quote: "Rerun the same fixtures." (Supports: step 4) Quote: "Compare correctness, efficiency, and contract fit." (Supports: step 5) Quote: "Keep, revert, or narrow the change." (Supports: step 6) Quote: "Don't swap a fixture mid-round, or the comparison means nothing." (Supports: warning) Snapshot SHA-256: f7710437d8e4f9410879f8e6be6d12798655a88675fab23f4ee30d85a1cff617 (captured 2026-11-13T03:08:41Z) Verification: Method: Automated extraction from [1] with deepseek-flash (JSON output, thinking disabled). Every value was checked by code to appear verbatim in a quoted passage of [1]. On this check the text of [1] was unchanged since 2026-11-13T03:08:41Z, so the values verified then were kept. Cross-check: not-available at 2026-11-20T12:33:20Z. No independent second source is configured for this entry Last successful check: 2026-11-20T12:33:20Z Last attempt: 2026-11-20T12:33:20Z Notes and limitations: Extracted automatically from a single official source (section "Run one measured round"). Related: https://latestn.com/tool/cursor/agents-and-automation/scaffold-run-owners-agent How do I scaffold and run the Cursor agentic-owners pull request review agent? https://latestn.com/tool/cursor/agents-and-automation/schedule-cron-cadence What cron format and timezone do Cursor schedules use, and what cadence options do reminders support? https://latestn.com/tool/cursor/agents-and-automation/schedules-vs-reminders What is the difference between schedules and reminders in the Cursor Agent SDK? https://latestn.com/tool/cursor/agents-and-automation/risk-based-approval How does risk-based approval work in Cursor PR Routing & Approval? https://latestn.com/tool/cursor/agents-and-automation/run-agent-review How do I run a Cursor Agent Review on my changes? https://latestn.com/tool/cursor/agents-and-automation/run-evals-cli-tag How do I run Cursor agent evals from the command line and filter by tag? https://latestn.com/tool/ All entries in AI tools Change log: Time | Event | Change | Source 2026-11-13T03:08:41Z | added | Entry created | — Changes for this entry as a feed: https://latestn.com/tool/cursor/agents-and-automation/run-one-measured-round/latest Challenge this entry: Open challenges: None How: If a value here is wrong or out of date, send what is wrong and a link to evidence. Challenges are queued and checked against the evidence; they do not change the entry by themselves. If the entry changes as a result, the change log says so. Form: https://latestn.com/dispute/tool/cursor/agents-and-automation/run-one-measured-round POST: https://latestn.com/dispute/tool/cursor/agents-and-automation/run-one-measured-round with claim, evidence_url, and optionally field and contact (form-encoded or JSON) GET: https://latestn.com/dispute/tool/cursor/agents-and-automation/run-one-measured-round?claim={what is wrong}&evidence_url={link} How to cite: LatestN. (2026). How do I run one measured round of hillclimbing in Cursor? Revision 1, last verified 2026-11-20T12:33:20Z. https://latestn.com/tool/cursor/agents-and-automation/run-one-measured-round (accessed 2026-11-21T10:07:48Z). @misc{LatestN_tool_cursor_agents_and_automation_run_one_measured_round_r1, title = {How do I run one measured round of hillclimbing in Cursor?}, author = {{LatestN}}, year = {2026}, howpublished = {\url{https://latestn.com/tool/cursor/agents-and-automation/run-one-measured-round}}, note = {Revision 1; digest sha256:fd32c80e89a10e77; last verified 2026-11-20T12:33:20Z; accessed 2026-11-21T10:07:48Z} } Reuse: License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) Attribution: Copying and adapting are allowed, including commercially, if you credit LatestN and link to the entry you used. Each entry ends with a ready-made citation. URL: https://latestn.com/tool/cursor/agents-and-automation/run-one-measured-round