Auto-activate for AlloyDB in GCP configs or docs. Google AlloyDB expertise: PostgreSQL-compatible managed database on GCP. Produces AlloyDB cluster configurations, connection patterns, and columnar engine setups on GCP. Use when: provisioning AlloyDB clusters, configuring read pools, using columnar engine, Private Service Access networking, or migrating from Cloud SQL. Not for AlloyDB Omni (see alloydb-omni) or vanilla PostgreSQL without AlloyDB features.

Auto-activate for alloydb-omni in compose/k8s configs. AlloyDB Omni expertise: run AlloyDB anywhere (local, on-prem, other clouds) with container-based deployment. Produces container-based AlloyDB Omni deployments for local dev and non-GCP environments. Use when: running AlloyDB locally for development, deploying Omni containers, configuring Kubernetes operators, or testing AlloyDB features without GCP. Not for GCP-managed AlloyDB (see alloydb) or vanilla PostgreSQL.

Use when any task depends on external API/framework behavior, SDK references, library versions, breaking changes, migration guides, changelogs, or release notes. Auto-activate for uncertainty about third-party docs, version compatibility, deprecations, or current syntax before implementation decisions. Produces version-verified documentation references with links to official sources, staleness warnings, and version gap notes.

Auto-activate when evaluating system architecture, reviewing component boundaries, assessing coupling between modules, planning large refactors, introducing new layers or abstractions, or when design decisions have long-term structural consequences. Produces structural assessment with boundary evaluation, coupling analysis, and time-horizon risk — what will be painful to change in 6-12 months. Use when: architecture review needed, evaluating maintainability of a design, checking for premature abstraction or missing abstraction, or assessing whether component boundaries are in the right place. Not for code style, naming conventions, or implementation details within well-bounded components.

Use when a workflow depends on Beads persistence and you need to choose or adapt between official Beads, beads_rust compatibility, or a no-Beads fallback

Auto-activate for Cloud SQL gcloud commands, cloud-sql-proxy usage, or Cloud SQL connection strings. Google Cloud SQL expertise: fully managed PostgreSQL, MySQL, and SQL Server on GCP. Produces Cloud SQL instance configurations, connection patterns, backup strategies, and replication setups. Use when: provisioning Cloud SQL instances, configuring Auth Proxy connections, setting up read replicas, managing backups and PITR, or migrating to Cloud SQL. For higher performance PostgreSQL workloads see flow:alloydb. For GKE deployment patterns see flow:gke.

Auto-activate for dishka imports. Dishka dependency injection framework: Provider, Scope, Container, FromDishka, Inject. Use when: setting up DI containers, defining providers/scopes, or integrating dependency injection with Litestar or FastAPI. Produces Dishka DI container configurations with providers, scopes, and framework integrations. Not for manual dependency injection or other DI frameworks.

Auto-activate for .duckdb files, duckdb imports. Comprehensive DuckDB expertise: advanced analytical SQL patterns, performance tuning, data engineering (ETL, multi-source reads, cloud storage), client APIs for Python/Node/Rust/Java/R/Go/WASM, extension development, function reference, and configuration/administration. Use when: writing DuckDB queries, optimizing performance, building data pipelines, connecting from any language, developing extensions, importing/exporting CSV/Parquet/JSON/Delta/Iceberg, or configuring DuckDB for production analytics workloads. Not for OLTP databases (see postgres/mysql) or traditional data warehouses.

Use at task, phase, flow, sync, archive, finish, revise, or failure checkpoints to keep Flow specs clean, capture learnings and failures, elevate durable patterns, and refine this skill with project-specific nuances

Auto-activate for shared memory, ring buffer, SPSC/MPMC patterns. Zero-copy IPC patterns: shared memory regions, SPSC/MPMC ring buffers, platform sync primitives, notification mechanisms, and cross-process coordination. Use when implementing IPC primitives or high-performance data transfer. Not for network IPC (gRPC, REST) or message queues.

Auto-activate for Makefile, GNUmakefile. GNU Make patterns for uv-based Python project automation: .PHONY, targets, recipes. Use when: creating or editing a Makefile, adding development targets (install, clean, test, lint), or setting up self-documenting help. Not for CMake (see cpp), Cargo (see rust), or non-Make build systems.

Flow-specific supplemental patterns for Mojo. Auto-activate for .mojo files, .🔥 files. Mojo development patterns for high-performance computing: SIMD, zero-copy Python interop, GIL-free parallelism, C FFI, and Hatch build integration. Use when: writing Mojo code, .mojo files, SIMD kernels, Python-Mojo hybrid projects, hatch-mojo build hooks, or packaging Mojo extensions into wheels. Produces high-performance Mojo code with SIMD kernels, Python interop, and Hatch build integration. Not for pure Python performance work or C extensions (see python/cpp).

Auto-activate when reviewing code in hot paths, evaluating database queries, assessing memory usage patterns, reviewing loop performance, checking for N+1 queries, evaluating caching strategies, or when code changes affect latency-sensitive operations. Produces bottleneck inventory with estimated impact (critical/moderate/minor), measurement recommendation for each finding, and fix priority. Use when: performance review needed, optimizing slow code, evaluating scaling bottlenecks, or assessing resource efficiency. Not for micro-optimizations on cold paths, premature optimization, or style-level concerns.

Auto-activate for podman commands, Containerfile. Podman expertise: rootless containers, pod management, systemd integration, and Docker CLI compatibility. Use when: running rootless containers, managing pods, using podman-compose, configuring systemd services, or working with OCI images without Docker daemon. Produces rootless container configurations, pod management, and systemd integration patterns. Not for Docker (see docker) or Kubernetes (see gke).

Auto-activate for pyapp build config. Build air-gapped, multi-architecture standalone Python executables using PyApp and uv. Use when: bundling Python runtimes for network-isolated environments, patching PyApp defaults, or compiling single-binary assets. Not for PyInstaller, cx_Freeze, or other Python packaging tools.

Auto-activate for saq imports, SAQ task queue configuration. SAQ (Simple Async Queue): async task queues, background jobs, cron scheduling, worker lifecycle. Produces SAQ task definitions, Worker setup, CronJob scheduling, and queue configuration. Use when: defining background tasks, enqueueing jobs, scheduling cron work, or managing worker lifecycle. Not for Celery, RQ, or Dramatiq -- SAQ has its own async-native patterns. For Litestar integration (SAQPlugin, DI, web UI, CLI), see flow:litestar.

Flow-specific supplemental patterns for shadcn/ui. Auto-activate for components.json (shadcn config), cn() utility. Tailwind component expertise for shadcn/ui. Use when: using cn() utility, Radix primitives, shadcn add, copy-paste components, component CLI, dialogs, forms, data tables, or command palettes. Not for Material UI, Chakra UI, or other component libraries.

Auto-activate when tracing execution paths, mapping dependencies, understanding unfamiliar code, following data flow through a system, investigating how a feature works end-to-end, or when debugging requires understanding call chains. Produces an execution map showing the path from entry point to result — with file paths, function names, data transformations, and dependency relationships at each node. Use when: need to understand how code flows from entry point to result, mapping what depends on what, exploring unfamiliar codebases systematically, or when reading random files isn't building understanding. Not for reading random files hoping to build understanding — every file opened must be because the previous file pointed you there.

Auto-activate for uvicorn imports, uvicorn CLI commands, ASGI server configuration. Uvicorn ASGI server: worker configuration, event loop selection, SSL, lifespan, logging, programmatic API. Produces uvicorn CLI invocations, Config/Server usage, and deployment configurations. Use when: deploying ASGI apps with uvicorn, configuring workers/reload, setting up SSL, or running development servers. Note: Granian is preferred over uvicorn for production workloads — see flow:granian.