agents

Use when the user wants to continue in a fresh session, hand work to another agent, preserve current context near context limits, or create a compact continuation brief for later. Use when they say handoff, next session, fresh agent, continue tomorrow, context limit, or pass this to another agent.

Use when planning features, designing changes, stress-testing ideas, scoping work, or deciding how to approach non-trivial implementation. Use when the user says plan, design, shape, grill, think through, prototype the idea, or asks for an implementation plan.

Use when running an unusually strict code quality review focused on maintainability, structural simplification, abstraction quality, spaghetti growth, giant files, type boundaries, and whether a branch should be blocked despite passing tests. Use when the user says thermo-nuclear review, thermonuclear review, harsh maintainability review, deep code quality audit, or review this branch aggressively.

Use when creating, editing, pruning, splitting, merging, or testing agent skills. Use when deciding skill frontmatter, model-vs-user invocation, description triggers, supporting files, pressure scenarios, or whether a skill should exist.

Use when debugging any bug, failing test, flaky behavior, performance regression, unexpected runtime behavior, or reviewing an existing bug theory/postmortem. Use before reading source code to explain a symptom, before proposing a fix, or when tempted to patch from inspection.

Use when deciding which layer a piece of code belongs in (domain, application, or infrastructure), when business rules are mixed with framework/UI/database/HTTP code, when an external SDK or API response shape is leaking into core logic, when a test forces you to mock your own classes or sibling collaborators, or when unsure whether to introduce an interface/port/abstraction.

Use when implementing features, fixing bugs (including P0 hotfixes and production incidents), refactoring, removing dead code, or deleting legacy paths. Use when user mentions TDD, red-green-refactor, test-first, or when tempted to write code before tests. Use when fixing any bug — even trivial ones. Use when encountering or modifying existing code that lacks tests. Use when writing tests, deciding what to mock, or when tests are full of mocks, coupled to implementation, or break on every refactor.

Use before any code, config, or documentation edit. Use when the request uses vague verbs like "refactor", "clean up", "improve", "modernize", or "make better". Use when adjacent issues (dead code, bad style, smells, TODOs) are visible during a focused task. Use when tempted to do "while I'm here" cleanup.