One skill format. Every agent runtime.

[A/B Testing Framework] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets experiment spec / variant assignment / metric definition / statistical analysis script. Known failure pattern: Running A/B tests with sample sizes too small to reach statistical significance, leading to decisions based on noise.. Best practice: Use an online sample size calculator before starting the test. Define the minimum detectable effect and ensure the test runs for at least one full business cycle..

[Accessibility ARIA Patterns] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets ARIA attribute refactor / keyboard navigation / focus management / screen reader test script. Known failure pattern: Adding ARIA attributes that conflict with native HTML semantics (e.g., role='button' on a <button> element), confusing screen readers.. Best practice: Use native HTML elements whenever possible. Only use ARIA to supplement missing semantics, never to override existing ones. Test with a real screen reader..

[Agent Tool Binding & Dispatch] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets router tool / domain group / dynamic tool injection / tool usage statistics. Known failure pattern: Giving the agent too many tools at once, causing it to spend more time choosing than executing, and increasing token usage significantly.. Best practice: Group tools by domain and offer a 'router' tool first. The agent picks a domain, then that domain's tools are injected. This reduces the tool set per step..

[Analytics Metric Definitions] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets metric definition / dbt model / SQL logic / dashboard tile / documentation. Known failure pattern: Different teams computing the same metric (e.g., 'daily active users') with different SQL logic, producing conflicting numbers.. Best practice: Define every metric in a central repository as a dbt model or LookML view with a single source of truth, and document its logic explicitly..

[Architecture Decision Records] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets ADR document / decision log / template / review workflow. Known failure pattern: Making important architectural decisions without documenting the context, alternatives, and rationale, leaving future team members confused about why something was done.. Best practice: Write an ADR for every non-trivial decision. Include the context, considered alternatives (with pros/cons of each), the chosen option, and the consequences..

[AWS Lambda Cold Starts] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets handler refactor / SnapStart config / Provisioned Concurrency / warmer function. Known failure pattern: Cold starts lasting multiple seconds because the function loads heavy dependencies or initialises database connections outside the handler.. Best practice: Move initialisation (DB connections, config loading) outside the handler. Use Lambda SnapStart for Java or .NET. Consider Provisioned Concurrency for latency-sensitive functions..

[Azure Bicep Infrastructure] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets main.bicep / module / parameter file / azd template. Known failure pattern: Manually creating resources in the portal without infrastructure-as-code, making environments inconsistent and hard to reproduce.. Best practice: Always define Azure resources in Bicep or Terraform. Use parameters and modules to keep the code DRY and environment-agnostic..

[Browser DevTools & Debugging] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets debugging workflow / breakpoint guide / performance recording / memory snapshot. Known failure pattern: Trying to debug frontend issues by guessing instead of using the Elements, Console, Network, and Sources panels systematically.. Best practice: Start with the Network panel to confirm the request/response are correct, then use Sources to set breakpoints, then Elements to inspect the DOM..

[CLI Tool Design Patterns] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets CLI scaffolding / argument parser / exit code handler / --json output mode. Known failure pattern: Building CLI tools that print output without usable exit codes (always exits 0) or swallow error messages, making them impossible to script with.. Best practice: Always exit 0 on success, non-zero on failure. Print errors to stderr, output to stdout. Support --json flag for machine-readable output..

[Cloud Cost Optimisation] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets right-sizing recommendation / auto-stop schedule / reserved instance plan / unused resource report. Known failure pattern: Running oversized instances 'just in case', or leaving development/staging resources running 24/7 when they are only needed during working hours.. Best practice: Right-size instances based on actual usage metrics (not peak theoretical load). Use auto-stop schedules for non-production environments..

[Code Review Checklist] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets review checklist / automated review comment / risk classification / diff summary. Known failure pattern: Reviewers focusing only on code style and missing architectural issues like missing error handling, security vulnerabilities, or performance regressions.. Best practice: Use a structured review checklist: correctness, security, performance, test coverage, error handling, and code style — in that order..

[Convex Functions & Mutations] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets mutation / query / action / component / scheduler job. Known failure pattern: Accidentally creating OCC (Optimistic Concurrency Control) conflicts by reading and writing the same document in rapid succession from multiple clients.. Best practice: Use patch() for partial updates and batch mutations for atomic multi-document writes. Avoid reading a document before immediately writing it back..

[Cron Job & Scheduled Task Reliability] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets crontab entry / log rotation / idempotency guard / failure alert integration. Known failure pattern: Cron jobs failing silently because output is not logged, or running the same job multiple times when the system is down at the scheduled time.. Best practice: Redirect cron output to a log file with timestamp. Use || to send failure alerts. Implement job idempotency so running it multiple times has no side effects..

[CSS Layout & Responsiveness] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets CSS layout refactor / responsive grid / container query implementation. Known failure pattern: Over-reliance on media queries when container queries or flex/grid intrinsic sizing would be simpler and more maintainable.. Best practice: Design for the content, not the viewport. Use clamp(), minmax(), and auto-fit/auto-fill before reaching for breakpoints..

[CSV Data Cleaning Pipeline] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets CSV parser / row validator / column type mapper / error report / cleaned output. Known failure pattern: Assuming CSV values are clean and consistent, then hitting parsing errors or silent data corruption when encountering commas inside quoted fields, missing headers, or inconsistent newlines.. Best practice: Always use a proper CSV parser (Python's csv module, Papa Parse in JS) instead of splitting on commas. Validate column count and types for every row..

[Database Migration Safety] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets batch migration / expand-contract pattern / zero-downtime migration / rollback plan. Known failure pattern: Running a long-running migration (e.g., adding a column with a default value) that locks the table and causes downtime for active users.. Best practice: Use PostgreSQL's ADD COLUMN DEFAULT (no-rewrite in recent versions) or break the migration into steps: add column without default, backfill in batches, then add default..

[Data Warehouse Schema Design] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets star schema / fact table / dimension table / ETL pipeline spec. Known failure pattern: Using a highly normalised OLTP schema (3NF) directly in a data warehouse, causing complex JOINs and slow analytical queries.. Best practice: Use a star schema (one fact table, multiple dimension tables) or a wide-column denormalised table for analytical queries. Pre-join at loading time..

[Design Token Systems] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets token JSON / CSS custom properties / theme switcher / token documentation. Known failure pattern: Hardcoding colors, spacing, or typography values in components instead of referencing design tokens, making theming impossible without changing every file.. Best practice: Define all visual primitives as CSS custom properties or JSON tokens. Reference them in components via token names, not literal values..

[Docker Compose Networking] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets docker-compose.yml / network config / healthcheck / depends_on condition. Known failure pattern: Services unable to reach each other because they are on different Docker networks, or using 'localhost' instead of the service name.. Best practice: All services in the same docker-compose.yml are on a shared network by default. Reference other services by their service name, not 'localhost'..

[Docker Multi-Stage Builds] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets multi-stage Dockerfile / .dockerignore / slim base image switch. Known failure pattern: Including the entire node_modules and build toolchain in the final production image, making it unnecessarily large and insecure.. Best practice: Use at least two stages: one for installing dev dependencies and building, another for copying only the production artefacts and running the app..

[Drizzle Schema Design] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets schema.ts / relation map / migration SQL / Drizzle query builder. Known failure pattern: Over-using relations() when simple foreign key columns with manual joins would be clearer and faster.. Best practice: Define relations only for eagerly loaded nested data. For simple lookups, just reference the foreign key column directly..

[Error Monitoring & Alerting Setup] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets Sentry project config / alert rule / error grouping / source map upload / performance monitoring. Known failure pattern: Setting up error monitoring (Sentry, Datadog) but configuring no alerts, so errors accumulate silently until a user complains.. Best practice: Configure at least two alerts: one for new errors (errors appearing for the first time) and one for error spikes (error count exceeding a threshold)..

[FastAPI Dependency Injection] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets dependency / lifespan handler / override for testing. Known failure pattern: Re-initialising the same database connection or HTTP client inside every route instead of using FastAPI's dependency injection.. Best practice: Define shared resources (DB pool, HTTP client) as lifespan-managed dependencies and inject them via Depends()..

[Feature Flags & Gradual Rollouts] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets flag provider config / gradual rollout target / flag cleanup plan / A/B test flag. Known failure pattern: Leaving feature flag code in the codebase permanently, making the codebase harder to read and maintain, and never removing old flags.. Best practice: Treat feature flags as temporary. After a flag has been fully rolled out and stable for one release cycle, remove the flag code and the flag condition entirely..

[Git Conflict Resolution] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets conflict resolution plan / cherry-pick strategy / rebase workflow / merge commit message. Known failure pattern: Resolving merge conflicts by blindly accepting one side without understanding why the change was made, potentially reintroducing bugs.. Best practice: For each conflicted section, trace the origin of both changes using 'git log --oneline' on the file. Understand the intent before picking a resolution..

[GitHub Actions Pipeline Optimisation] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets workflow YAML / cache config / matrix build / conditional job execution. Known failure pattern: Long CI times caused by not caching dependencies between runs, or running the full test suite on every push regardless of change scope.. Best practice: Cache node_modules (or other dependency folders) using actions/cache with a hash of the lock file. Use paths filter to run only relevant jobs..

[GraphQL N+1 Query Prevention] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets DataLoader instance / batch load function / resolver refactor / query complexity analysis. Known failure pattern: A resolver that fetches a parent entity, then for each child calls a separate database query, resulting in N+1 queries for N children.. Best practice: Use DataLoader to batch and cache child-loading queries. DataLoader groups all child-loading calls into a single IN query per request cycle..

[Jest Test Optimisation] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets jest config optimisation / --changedSince / --onlyChanged / test sharding / module mocking. Known failure pattern: Running the entire test suite on every change, taking minutes even for small incremental code changes.. Best practice: Use jest --changedSince to run only tests related to changed files. Use jest --onlyChanged during development to get instant feedback..

[JSON Schema Validation] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets JSON Schema / validator middleware / type guard / error message / response parser. Known failure pattern: Trusting external API responses without validating their structure, causing runtime errors when the API changes the response format unexpectedly.. Best practice: Always validate external JSON responses against a JSON Schema before accessing properties. Use AJV (JavaScript) or jsonschema (Python) for fast validation..

[Kubernetes Horizontal Pod Autoscaling] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets HPA manifest / custom metric / vertical pod autoscaler / cluster autoscaler config. Known failure pattern: HPA not scaling because metrics-server is not installed, or because resource requests/limits are not set on the target deployment.. Best practice: Always set CPU/memory requests on every container. HPA cannot scale based on resource metrics without requests defined..

[Kubernetes Pod Lifecycle] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets deployment.yaml / startup probe / readiness probe / liveness probe / init container. Known failure pattern: Pods stuck in CrashLoopBackOff because the application exits when a dependency (database, cache) is not yet ready.. Best practice: Implement a startup probe with a longer initial delay and a readiness probe that checks actual dependency health, not just TCP connectivity..

[LLM Context Window Budget Management] Audit the current setup; do NOT modify files; produce a structured inventory and a risk-ranked list of findings Targets trimmed context array / token budget report / sliding window snapshot / semantic retrieval hit list / cache hit dashboard. Known failure pattern: Dumping the entire conversation history plus all file contents into the LLM context window on every turn, causing immediate overflow on multi-hour sessions and burning tens of thousands of tokens on redundant content. Worse: re-reading the same 10MB file 50 times because each tool call rebuilds context from scratch without cache awareness.. Best practice: Use sliding window summarization: keep system prompt + last 5 turns verbatim, compress older turns into a 200-token lossless summary. Aggressively cache stable prefixes (system prompt, tool schemas, file headers). Strip redundant tool outputs after they're acted on. Use semantic search to inject only relevant code chunks, never whole files. Always log token usage per turn so budget overruns are visible..

[MCP Tool Design & Best Practices] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets MCP tool descriptor / resource definition / prompt template / server metadata. Known failure pattern: Designing MCP tool names that are too generic ('search', 'get_data') causing ambiguity when multiple tools are available to the agent.. Best practice: Prefix tool names with a namespace that reflects their domain (e.g., 'github_search_repos', 'jira_get_issue'). Always provide a detailed description of when to use each tool..

[Message Queues & Background Jobs] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets queue producer / worker / dead-letter handler / retry policy. Known failure pattern: Losing messages when a worker crashes before acknowledging completion, because auto-ack is enabled.. Best practice: Disable auto-ack. Acknowledge only after the job has been fully processed and its result has been persisted..

[Multi-Tenant Data Isolation] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets RLS policy / tenant context middleware / session variable injection / tenant-aware query builder. Known failure pattern: Using a single database with a tenant_id column but forgetting to filter by tenant_id in every query, accidentally mixing tenant data.. Best practice: Use PostgreSQL Row-Level Security with tenant_id automatically set via session variable. This guarantees isolation even if a query misses the WHERE clause..

[Next.js API Routes & Route Handlers] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets route.ts handler / server action / API client wrapper / error boundary. Known failure pattern: Exposing server-side secrets to the client by accidentally importing environment variables in a 'use client' component.. Best practice: All sensitive operations (DB queries, external API calls with keys) belong in API routes or server actions, never in client components..

[Next.js Data Fetching Patterns] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets server fetch / React cache wrapper / streaming suspense boundary. Known failure pattern: Fetching the same data in multiple server components or mixing server fetch with client fetch leading to duplicate network requests.. Best practice: Use server components for initial data fetch and pass down as props. Use React.cache() to deduplicate fetches across parallel routes..

[Next.js Middleware & Edge Runtime] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets middleware.ts / rewrite rule / cookie-based redirect / geolocation routing. Known failure pattern: Using Node.js APIs (fs, crypto, database drivers) inside Edge Middleware, causing runtime crashes.. Best practice: Keep middleware stateless and light. Use it only for redirects, rewrites, header manipulation, and basic auth checks..

[Node.js Error Handling & Resilience] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets global error handler / async wrapper / structured error response / retry logic. Known failure pattern: Unhandled promise rejections crashing the process, or try-catch blocks that swallow errors without logging context.. Best practice: Use a global error handler for uncaught exceptions and unhandled rejections. Wrap every async route handler in a higher-order catch function..

[Node.js Streams & Backpressure] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets Readable/Writable stream / Transform / pipeline() refactor. Known failure pattern: Reading entire files into memory instead of streaming, or ignoring backpressure signals from writable streams.. Best practice: Use pipeline() instead of pipe() because pipeline automatically handles backpressure and destroys streams on error..

[OAuth 2.0 Flows & Token Management] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets OAuth callback / token refresh / PKCE flow / httpOnly cookie handler. Known failure pattern: Storing access tokens in localStorage, making them accessible to XSS attacks, and not implementing refresh token rotation.. Best practice: Store tokens in an httpOnly cookie set by the server, not in client-side storage. Implement refresh token rotation and revoke old refresh tokens after use..

[OpenAPI Specification & Validation] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets openapi.yaml / code-first generator / request/response validation middleware. Known failure pattern: Generating an OpenAPI spec that is out of sync with the actual implementation because the spec is maintained manually instead of generated from code.. Best practice: Use code-first OpenAPI generation (FastAPI, NestJS swagger, or express-openapi) so the spec always reflects the actual routes..

[Playwright Selectors & Locators] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets locator refactor / test fixture / POM (Page Object Model) / custom fixture. Known failure pattern: Using fragile CSS selectors (nth-child, class names that change) that break on every UI update.. Best practice: Use getByRole, getByText, or getByTestId with semantic naming. These are resilient to CSS and DOM structure changes..

[Prompt Injection Defense] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets defensive system prompt / input sanitizer / instruction guardrail / output validator. Known failure pattern: Building a system prompt that includes user input directly without isolation, allowing users to override instructions by saying 'ignore previous instructions'.. Best practice: Isolate user input in a delimited section, use a separate 'input' variable, and add explicit guardrails that reject instruction override attempts..

[Python Async/Await Patterns] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets async/await refactor / asyncio.gather / async context manager. Known failure pattern: Blocking the event loop by using synchronous requests or time.sleep inside async functions.. Best practice: Use httpx.AsyncClient for HTTP calls and asyncio.sleep for delays inside async functions. Never mix sync and async I/O in the same function..

[Python File I/O & Encoding] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets pathlib refactor / encoding-safe file reader / batch file processor. Known failure pattern: Opening binary files in text mode or assuming UTF-8 encoding, leading to UnicodeDecodeError on non-ASCII content.. Best practice: Always specify encoding explicitly when opening text files. Use pathlib.Path.read_text/write_bytes for cleaner code..

[RAG Chunking Strategies] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets semantic chunker / chunk overlap config / hybrid retriever / chunk metadata enrichment. Known failure pattern: Using fixed-size chunking (500 characters) that splits sentences or code blocks in half, reducing retrieval quality.. Best practice: Use semantic chunking: split on paragraph boundaries, markdown headings, or code function boundaries. Overlap adjacent chunks by 10-20% to avoid missing context near boundaries..

[Rate Limiting & API Gateway Proxy] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets NGINX rate limit config / Cloudflare WAF rule / API Gateway usage plan / token bucket implementation. Known failure pattern: Applying rate limiting at the application level without a proxy layer, so rate-limited requests still consume application server resources.. Best practice: Enforce rate limits at the reverse proxy level (NGINX, Cloudflare, API Gateway) before the request reaches your application server..

[React Server Components] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets server component / client boundary refactor / streaming fallback. Known failure pattern: Accidentally making a server component a client component by using hooks or event handlers in the wrong file.. Best practice: Keep data fetching and heavy logic in server components; pass results as props to client islands..

[React State Management] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets useState / useReducer / useContext hook refactor, zustand or jotai store slice. Known failure pattern: Stale closures or unnecessary re-renders caused by missing dependency arrays or incorrect state initialisation.. Best practice: Co-locate state as close to the consuming component as possible. Lift state only when two or more siblings need to share it..

[Redis Caching Strategies] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets cache wrapper / mutex lock / stale-while-revalidate / TTL policy. Known failure pattern: Cache stampede: multiple requests simultaneously recomputing an expired cache entry because they all detected expiry at the same time.. Best practice: Use a mutex lock around cache regeneration, or stale-while-revalidate pattern to serve stale data while the new value is being computed..

[REST Pagination Design] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets cursor pagination / offset pagination fallback / total count optimisation / response envelope. Known failure pattern: Using offset-based pagination with large offsets ('?offset=10000') that causes slow database queries because the DB has to scan and skip many rows.. Best practice: Use cursor-based pagination (keyset pagination) for large datasets. The cursor is an opaque token that points to the last item, and the DB query uses WHERE > cursor_value..

[Secrets Rotation Policy] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets rotation script / vault integration / lease management / incident response plan. Known failure pattern: Using long-lived API keys and secrets that never expire, increasing the blast radius if they leak.. Best practice: Automate secret rotation with a scheduled job. Use short-lived tokens (e.g., 90 days) and rotate them before expiry. Store secrets in a vault, not in env files..

[Shell Script Robustness & Safety] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets set -euo pipefail script / confirmation prompt / shellcheck-passing script / rollback function. Known failure pattern: Shell scripts that fail silently midway because 'set -e' is not set, or that modify files without confirmation, causing irreversible damage.. Best practice: Always start scripts with 'set -euo pipefail'. Add confirmation prompts before destructive operations. Use shellcheck to lint the script..

[SQL Query Optimisation] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets indexed query / composite index / EXPLAIN ANALYSE plan / partial index. Known failure pattern: Using SELECT * in production queries and missing indexes on foreign key columns used in JOINs.. Best practice: Always select only the columns you need. Add composite indexes that match your WHERE + ORDER BY clauses exactly..

[Stealth Web Research & Harvesting] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets clean markdown corpus / structured JSON metadata / per-page extraction report / sitemap of crawled pages. Known failure pattern: Web scrapers getting blocked by Cloudflare, Akamai, or DataDome bot detection because they send no user-agent, use headless Chromium without stealth plugins, or hammer endpoints with zero delays between requests.. Best practice: Use stealth-augmented browser automation (playwright-extra + stealth or puppeteer-extra + stealth plugin). Rotate realistic user agents with referrer headers. Add 1.5-3 second random delays between navigations. Respect robots.txt and rate-limit headers..

[Stripe Webhook Idempotency] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets Webhook handler / idempotency key check / event deduplication / failed payment recovery. Known failure pattern: Processing the same Stripe webhook event twice because Stripe sends at-least-once delivery, causing duplicate charges or duplicate subscription activations.. Best practice: Use the Stripe-Idempotency-Key or the event ID as a unique constraint in your database to skip already-processed events..

[Supabase Row-Level Security] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets RLS policy / policy test / security definer function / admin bypass. Known failure pattern: RLS policies that are too permissive (using 'true' instead of 'auth.uid() = user_id') accidentally exposing other users' data.. Best practice: Always reference auth.uid() in RLS policies. Test policies with a non-admin user before deploying to production..

[Terraform State Management] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets backend config / state migration plan / state locking config / remote state datasource. Known failure pattern: Losing the .tfstate file (or it becoming corrupted), forcing manual reconstruction of the entire infrastructure.. Best practice: Always store state in a remote backend (S3, Azure Storage, Terraform Cloud) with state locking enabled via DynamoDB or equivalent..

[TypeScript Generics & Advanced Types] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets generic type / conditional type / mapped type / branded type. Known failure pattern: Generic constraints that are too loose (accepting anything) or too tight (requiring exact shapes when interfaces would suffice).. Best practice: Prefer generic constraints that describe the minimum required structure (extends) rather than listing every possible property..

[User Onboarding Flow Design] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets onboarding wizard / feature checklist / in-app guide / first-run experience spec. Known failure pattern: Showing the user a long tutorial or feature list on first login, overwhelming them and causing the majority to leave before experiencing core value.. Best practice: Use progressive disclosure: only introduce features when the user reaches the point where they need them. A 3-step wizard that gets them to the 'aha moment' in under 60 seconds is ideal..

[Vercel Environment Variables] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets vercel.json env group / preview env config / Edge Config / KV store. Known failure pattern: Accidentally exposing preview URLs or internal API keys by adding them as preview environment variables that get picked up by branch deployments.. Best practice: Use separate environment groups for production, preview, and development. Never mark sensitive keys as 'available to all branches'..

[Web Scraping Ethics & Compliance] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets robots.txt check / polite scraper / rate-limited crawler / cached scraper. Known failure pattern: Scraping a website that explicitly prohibits it in robots.txt or terms of service, leading to legal or IP blocking issues.. Best practice: Always check robots.txt and terms of service before scraping. Respect Crawl-Delay directives and set a reasonable User-Agent with contact information..

[WebSocket Reconnection Strategies] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets WebSocket client / reconnection logic / heartbeat / connection status component. Known failure pattern: Losing real-time updates when the WebSocket disconnects temporarily, and not attempting to reconnect, leaving the UI in a stale state.. Best practice: Implement exponential backoff reconnection with a maximum delay of 30 seconds. Show a connection status indicator in the UI..

[Web Vitals Optimisation (LCP/CLS/INP)] Inspect the current state without making changes. List files, configurations, dependencies, or outputs relevant to the goal. Produce a structured inventory with findings, risks, and recommendations — but do not modify anything. Targets image optimisation / font display swap / critical CSS / lazy load / bundle analysis. Known failure pattern: Large LCP caused by a hero image that is larger than needed and not optimised (WebP, lazy loading, proper dimensions).. Best practice: Serve images in WebP/AVIF format, specify width and height to reserve space (prevent CLS), and lazy-load below-the-fold images. Use next/image for automatic optimisation..

[A/B Testing Framework] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets experiment spec / variant assignment / metric definition / statistical analysis script. Known failure pattern: Running A/B tests with sample sizes too small to reach statistical significance, leading to decisions based on noise.. Best practice: Use an online sample size calculator before starting the test. Define the minimum detectable effect and ensure the test runs for at least one full business cycle..

[Accessibility ARIA Patterns] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets ARIA attribute refactor / keyboard navigation / focus management / screen reader test script. Known failure pattern: Adding ARIA attributes that conflict with native HTML semantics (e.g., role='button' on a <button> element), confusing screen readers.. Best practice: Use native HTML elements whenever possible. Only use ARIA to supplement missing semantics, never to override existing ones. Test with a real screen reader..

[Agent Tool Binding & Dispatch] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets router tool / domain group / dynamic tool injection / tool usage statistics. Known failure pattern: Giving the agent too many tools at once, causing it to spend more time choosing than executing, and increasing token usage significantly.. Best practice: Group tools by domain and offer a 'router' tool first. The agent picks a domain, then that domain's tools are injected. This reduces the tool set per step..

[Analytics Metric Definitions] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets metric definition / dbt model / SQL logic / dashboard tile / documentation. Known failure pattern: Different teams computing the same metric (e.g., 'daily active users') with different SQL logic, producing conflicting numbers.. Best practice: Define every metric in a central repository as a dbt model or LookML view with a single source of truth, and document its logic explicitly..

[Architecture Decision Records] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets ADR document / decision log / template / review workflow. Known failure pattern: Making important architectural decisions without documenting the context, alternatives, and rationale, leaving future team members confused about why something was done.. Best practice: Write an ADR for every non-trivial decision. Include the context, considered alternatives (with pros/cons of each), the chosen option, and the consequences..

[AWS Lambda Cold Starts] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets handler refactor / SnapStart config / Provisioned Concurrency / warmer function. Known failure pattern: Cold starts lasting multiple seconds because the function loads heavy dependencies or initialises database connections outside the handler.. Best practice: Move initialisation (DB connections, config loading) outside the handler. Use Lambda SnapStart for Java or .NET. Consider Provisioned Concurrency for latency-sensitive functions..

[Azure Bicep Infrastructure] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets main.bicep / module / parameter file / azd template. Known failure pattern: Manually creating resources in the portal without infrastructure-as-code, making environments inconsistent and hard to reproduce.. Best practice: Always define Azure resources in Bicep or Terraform. Use parameters and modules to keep the code DRY and environment-agnostic..

[Browser DevTools & Debugging] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets debugging workflow / breakpoint guide / performance recording / memory snapshot. Known failure pattern: Trying to debug frontend issues by guessing instead of using the Elements, Console, Network, and Sources panels systematically.. Best practice: Start with the Network panel to confirm the request/response are correct, then use Sources to set breakpoints, then Elements to inspect the DOM..

[CLI Tool Design Patterns] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets CLI scaffolding / argument parser / exit code handler / --json output mode. Known failure pattern: Building CLI tools that print output without usable exit codes (always exits 0) or swallow error messages, making them impossible to script with.. Best practice: Always exit 0 on success, non-zero on failure. Print errors to stderr, output to stdout. Support --json flag for machine-readable output..

[Cloud Cost Optimisation] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets right-sizing recommendation / auto-stop schedule / reserved instance plan / unused resource report. Known failure pattern: Running oversized instances 'just in case', or leaving development/staging resources running 24/7 when they are only needed during working hours.. Best practice: Right-size instances based on actual usage metrics (not peak theoretical load). Use auto-stop schedules for non-production environments..

[Code Review Checklist] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets review checklist / automated review comment / risk classification / diff summary. Known failure pattern: Reviewers focusing only on code style and missing architectural issues like missing error handling, security vulnerabilities, or performance regressions.. Best practice: Use a structured review checklist: correctness, security, performance, test coverage, error handling, and code style — in that order..

[Convex Functions & Mutations] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets mutation / query / action / component / scheduler job. Known failure pattern: Accidentally creating OCC (Optimistic Concurrency Control) conflicts by reading and writing the same document in rapid succession from multiple clients.. Best practice: Use patch() for partial updates and batch mutations for atomic multi-document writes. Avoid reading a document before immediately writing it back..

[Cron Job & Scheduled Task Reliability] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets crontab entry / log rotation / idempotency guard / failure alert integration. Known failure pattern: Cron jobs failing silently because output is not logged, or running the same job multiple times when the system is down at the scheduled time.. Best practice: Redirect cron output to a log file with timestamp. Use || to send failure alerts. Implement job idempotency so running it multiple times has no side effects..

[CSS Layout & Responsiveness] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets CSS layout refactor / responsive grid / container query implementation. Known failure pattern: Over-reliance on media queries when container queries or flex/grid intrinsic sizing would be simpler and more maintainable.. Best practice: Design for the content, not the viewport. Use clamp(), minmax(), and auto-fit/auto-fill before reaching for breakpoints..

[CSV Data Cleaning Pipeline] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets CSV parser / row validator / column type mapper / error report / cleaned output. Known failure pattern: Assuming CSV values are clean and consistent, then hitting parsing errors or silent data corruption when encountering commas inside quoted fields, missing headers, or inconsistent newlines.. Best practice: Always use a proper CSV parser (Python's csv module, Papa Parse in JS) instead of splitting on commas. Validate column count and types for every row..

[Database Migration Safety] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets batch migration / expand-contract pattern / zero-downtime migration / rollback plan. Known failure pattern: Running a long-running migration (e.g., adding a column with a default value) that locks the table and causes downtime for active users.. Best practice: Use PostgreSQL's ADD COLUMN DEFAULT (no-rewrite in recent versions) or break the migration into steps: add column without default, backfill in batches, then add default..

[Data Warehouse Schema Design] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets star schema / fact table / dimension table / ETL pipeline spec. Known failure pattern: Using a highly normalised OLTP schema (3NF) directly in a data warehouse, causing complex JOINs and slow analytical queries.. Best practice: Use a star schema (one fact table, multiple dimension tables) or a wide-column denormalised table for analytical queries. Pre-join at loading time..

[Design Token Systems] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets token JSON / CSS custom properties / theme switcher / token documentation. Known failure pattern: Hardcoding colors, spacing, or typography values in components instead of referencing design tokens, making theming impossible without changing every file.. Best practice: Define all visual primitives as CSS custom properties or JSON tokens. Reference them in components via token names, not literal values..

[Docker Compose Networking] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets docker-compose.yml / network config / healthcheck / depends_on condition. Known failure pattern: Services unable to reach each other because they are on different Docker networks, or using 'localhost' instead of the service name.. Best practice: All services in the same docker-compose.yml are on a shared network by default. Reference other services by their service name, not 'localhost'..

[Docker Multi-Stage Builds] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets multi-stage Dockerfile / .dockerignore / slim base image switch. Known failure pattern: Including the entire node_modules and build toolchain in the final production image, making it unnecessarily large and insecure.. Best practice: Use at least two stages: one for installing dev dependencies and building, another for copying only the production artefacts and running the app..

[Drizzle Schema Design] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets schema.ts / relation map / migration SQL / Drizzle query builder. Known failure pattern: Over-using relations() when simple foreign key columns with manual joins would be clearer and faster.. Best practice: Define relations only for eagerly loaded nested data. For simple lookups, just reference the foreign key column directly..

[Error Monitoring & Alerting Setup] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets Sentry project config / alert rule / error grouping / source map upload / performance monitoring. Known failure pattern: Setting up error monitoring (Sentry, Datadog) but configuring no alerts, so errors accumulate silently until a user complains.. Best practice: Configure at least two alerts: one for new errors (errors appearing for the first time) and one for error spikes (error count exceeding a threshold)..

[FastAPI Dependency Injection] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets dependency / lifespan handler / override for testing. Known failure pattern: Re-initialising the same database connection or HTTP client inside every route instead of using FastAPI's dependency injection.. Best practice: Define shared resources (DB pool, HTTP client) as lifespan-managed dependencies and inject them via Depends()..

[Feature Flags & Gradual Rollouts] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets flag provider config / gradual rollout target / flag cleanup plan / A/B test flag. Known failure pattern: Leaving feature flag code in the codebase permanently, making the codebase harder to read and maintain, and never removing old flags.. Best practice: Treat feature flags as temporary. After a flag has been fully rolled out and stable for one release cycle, remove the flag code and the flag condition entirely..

[Git Conflict Resolution] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets conflict resolution plan / cherry-pick strategy / rebase workflow / merge commit message. Known failure pattern: Resolving merge conflicts by blindly accepting one side without understanding why the change was made, potentially reintroducing bugs.. Best practice: For each conflicted section, trace the origin of both changes using 'git log --oneline' on the file. Understand the intent before picking a resolution..

[GitHub Actions Pipeline Optimisation] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets workflow YAML / cache config / matrix build / conditional job execution. Known failure pattern: Long CI times caused by not caching dependencies between runs, or running the full test suite on every push regardless of change scope.. Best practice: Cache node_modules (or other dependency folders) using actions/cache with a hash of the lock file. Use paths filter to run only relevant jobs..

[GraphQL N+1 Query Prevention] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets DataLoader instance / batch load function / resolver refactor / query complexity analysis. Known failure pattern: A resolver that fetches a parent entity, then for each child calls a separate database query, resulting in N+1 queries for N children.. Best practice: Use DataLoader to batch and cache child-loading queries. DataLoader groups all child-loading calls into a single IN query per request cycle..

[Jest Test Optimisation] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets jest config optimisation / --changedSince / --onlyChanged / test sharding / module mocking. Known failure pattern: Running the entire test suite on every change, taking minutes even for small incremental code changes.. Best practice: Use jest --changedSince to run only tests related to changed files. Use jest --onlyChanged during development to get instant feedback..

[JSON Schema Validation] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets JSON Schema / validator middleware / type guard / error message / response parser. Known failure pattern: Trusting external API responses without validating their structure, causing runtime errors when the API changes the response format unexpectedly.. Best practice: Always validate external JSON responses against a JSON Schema before accessing properties. Use AJV (JavaScript) or jsonschema (Python) for fast validation..

[Kubernetes Horizontal Pod Autoscaling] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets HPA manifest / custom metric / vertical pod autoscaler / cluster autoscaler config. Known failure pattern: HPA not scaling because metrics-server is not installed, or because resource requests/limits are not set on the target deployment.. Best practice: Always set CPU/memory requests on every container. HPA cannot scale based on resource metrics without requests defined..

[Kubernetes Pod Lifecycle] Create a reusable automation: a shell script, CLI tool, scheduled job, or workflow definition. The automation must handle edge cases (missing input, network failures, permission errors) and include a dry-run or test mode. Targets deployment.yaml / startup probe / readiness probe / liveness probe / init container. Known failure pattern: Pods stuck in CrashLoopBackOff because the application exits when a dependency (database, cache) is not yet ready.. Best practice: Implement a startup probe with a longer initial delay and a readiness probe that checks actual dependency health, not just TCP connectivity..