Explore Prompts

Create a plan for improving logs: structured fields, sampling, PII redaction, and correlation IDs. Require that changes do not reduce incident investigability.

Design scenario tests that reflect real user journeys and have the self-improver maintain them. Include a process to add scenarios from support tickets.

Create a blueprint for self-updating policy files (rulesets) with tests, staging, and canary enforcement. Require explicit approval for tightening/loosening security.

Design ownership enforcement: CODEOWNERS mapping, required reviewers per module, and automatic routing. The agent must not merge without required approvals.

Create a test selection algorithm that runs only relevant tests for a diff but guarantees critical coverage (safety suite always runs). Include calibration and drift checks.

Design a tool registry: versioned MCP tools, deprecation, and audit trails. Include a process for approving new tools and revoking compromised ones.

Design a routing agent that selects specialized agents (tests, security, perf, docs) based on task intent and risk. Include confidence thresholds and fallbacks.

Create a linter for agent prompts: ban risky instructions, require citations for factual claims, enforce “no tool calls from untrusted text” policies.

Design a data quality gate: schema, drift detection, missingness, and label quality checks. Integrate with self-improving code changes to ML pipelines safely.

Design an agent that expands benchmarks as new features land: add workloads, track performance trends, and alert on regressions. Include benchmark governance.

Create a plan for generating API clients: typed SDKs, version pinning, integration tests, and docs. Include a policy against generating unsafe dynamic code.

Design an agent that writes release notes from diffs and PR descriptions, focusing on user-visible impact, migrations, and known issues. Include accuracy checks.