omniagents

Use when writing or reviewing Python 3.14+ with strict type-safety enforced: generics with PEP 695 syntax, annotationlib / deferred annotations, Protocol vs ABC, ParamSpec decorators, TypeIs / TypeGuard narrowing, TypedDict / ReadOnly, magic strings / numbers vs Literal / StrEnum / Final, NewType vs aliases, or any time mypy / pyright / pyrefly / ty is configured, run, or failing.

Use when writing or reviewing TypeScript 6.0+ with strict type-safety: tsconfig defaults and deprecations, explicit global types, rootDir / paths / moduleResolution, import attributes, generics with T-prefix names, branded types, discriminated unions with `never` exhaustiveness, Result types, magic strings / numbers vs literal unions and named constants, Zod boundaries, `satisfies`, assertion functions, inferred type predicates, `using`, standard decorators, `unknown` vs `any`, or `// @ts-expect-error`.

Use when reviewing code for maintainability issues such as coupling, cohesion, abstraction leaks, boundary drift, code smells, naming problems, duplicated logic, premature generalization, or design-pattern misuse.

Use when producing or consuming structured code review findings. Defines the shared output contract, severity rubric, finding schema, per-specialist report shape, and final aggregated report shape that all reviewer specialist agents must follow and the verifier agent enforces.

Always use when user asks to create, generate, draw, or design a diagram, flowchart, architecture diagram, ER diagram, sequence diagram, class diagram, network diagram, mockup, wireframe, or UI sketch, or mentions draw.io, drawio, drawoi, .drawio files, or diagram export to PNG/SVG/PDF.

Use when the user is entering unfamiliar territory — a new area of the codebase, a new domain, tool, or problem space — or is about to scope substantial work in one, and the gap between what they asked for and what the territory actually demands is likely to be large. Also use when the user asks what they are missing, says they don't know what good looks like, or wants help prompting better before delegating work.

Generate standalone HTML + inline-SVG technical diagrams that come out format-clean every time — text that always fits its box, arrows that actually connect, no overlaps, no drift. Works by planning a coordinate grid first and sizing every box with monospace arithmetic instead of guessing pixels, then self-checking the geometry before delivery. Output is a single self-contained .html file (no build step, no dependencies) that opens in any browser.

Use when reviewing code for logic errors, type-safety violations, null/undefined handling, off-by-one, race conditions, incorrect API usage, broken contracts, missing error handling, edge-case gaps, or state-management bugs.

Use when reviewing code for operability issues such as missing or noisy logging, absent metrics or tracing on new code paths, unreported errors, unbounded timeouts or retries, unsafe migrations, missing rollout/rollback paths, ungraceful shutdown, or config and secrets handling problems.

Use when reviewing code for performance issues: algorithmic complexity, N+1 queries, unnecessary allocations, unbounded concurrency, cache misuse, blocking I/O in async context, hidden materialization, ReDoS, cold-start cost, missing database indexes, connection pool exhaustion, or memory leaks.

Use when reviewing code for security vulnerabilities: injection (SQL, command, template, log), authentication/authorization gaps, secrets in code, unsafe deserialization, path traversal, SSRF, dependency vulnerabilities, cryptographic misuse, data exposure, or OWASP Top 10 patterns.

Use when reviewing code for test adequacy issues such as missing regression coverage, weak assertions, brittle mocks, fixture misuse, untested edge cases, snapshot overuse, flaky timing, or tests that miss user-visible behavior.

Use when designing or reviewing cross-service, cross-process, regional, or operator-facing system shape: timeouts, retries, circuit breakers, bulkheads, rate limits, idempotency, DLQs, REST/gRPC/GraphQL/pub-sub, service mesh, webhooks, Saga, Raft/Paxos, distributed locks, CQRS, event sourcing, outbox, sharding, replication, caching, autoscaling, sidecars, strangler migrations, distributed monoliths, cascading failures, retry storms, or shared databases.

Use when writing or reviewing Python that wraps a network service or authors a client SDK: client construction, async lifecycle, granular timeouts, retry policy, idempotency keys, structured exception hierarchies, per-request overrides, structured outputs with Pydantic, streaming via async context managers, request-id propagation, or pagination.

Use when writing or reviewing TypeScript that wraps a network service or authors a client SDK: client construction, per-request options, retry policy with maxNetworkRetries, idempotency keys, `APIError` hierarchies, structured outputs via `zodResponseFormat` / `zodTextFormat`, streaming with AbortController, cursor pagination, or telemetry hooks.

Use when the user wants to deeply understand a bug, PR, refactor, feature, codebase flow, or design decision through structured Socratic dialogue. Triggers on "teach me", "explain deeply", "I want to understand", "walk me through", "help me learn", "deep dive", or /coding-teacher.

Use when designing or reviewing Python code shape inside one service: module/class boundaries, GoF patterns, DDD, functional Result/Option/ADT shapes, in-process concurrency, package architecture, code smells, anti-patterns, refactors, "what pattern?", "where should X live?", or "how should I model this?". For cross-service or distributed concerns, use the system skill instead.

Use when writing or reviewing TypeScript API documentation for exported modules, classes, functions, methods, types, interfaces, properties, TSDoc tags, TypeDoc output, compile-checkable examples, or JSDoc interop in legacy JavaScript.

Use when authoring or reviewing Markdown, CommonMark, GitHub-Flavored Markdown, MyST, README files, docs pages, Markdown tables, fenced code examples, links, images, citations, or markdownlint / Prettier output.

Use when writing technical strategy memos, proposals, experiment designs, RFCs, investment justifications, executive narratives, or persuasive documents where skeptical decision-makers need evidence, caveats, tradeoffs, and a measured, intellectually honest tone.

Use when Python code may fail at scale: profiling hot paths, memory pressure, unbounded async fan-out, backpressure, queue design, hidden materialization, cache growth, expensive logging, timeouts, cancellation, cProfile, tracemalloc, or performance reviews that need measurement before refactoring.

Use when writing or reviewing Python docstrings for public modules, classes, functions, methods, properties, generators, overloads, doctest examples, or generated Sphinx API documentation, especially when choosing NumPy vs Sphinx style or deciding which Parameters, Returns, Yields, Raises, Warns, Attributes, Notes, See Also, and Examples sections belong.

Use when writing or reviewing modern Pydantic v2 or pydantic-settings code: BaseModel boundaries, SettingsConfigDict, BaseSettings, nested environment configuration, Annotated Field constraints, validators, computed fields, TypeAdapter, RootModel, discriminated unions, SecretStr, serialization, or production validation patterns.