Explore Prompts

Design an accessibility agent that runs audits (linting, contrast checks where applicable), proposes fixes, and validates with automated tests. Require human review for UX changes.

Create a blueprint for updating localization strings: ensure placeholders match, run i18n checks, and verify no abusive/unsafe text. Include review steps.

Design a comment improvement agent that updates misleading comments but requires linking to code references and tests to avoid stale documentation.

Create a policy for change budgets: cap number/size of automated PRs per day, prioritize highest impact, and avoid overwhelming reviewers.

Design a global kill switch that halts all automated edits/deploys across environments. Include manual override process and audit logging.

Design an agent that updates runbooks based on incidents and recurring questions. Include change review, versioning, and a “proof of usefulness” metric.

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.