Design alerts that fire on the symptoms users feel, using error-budget burn rate, a page-worthiness test, and a runbook link on every rule.
Authenticate and authorise access to a remote MCP server so tools act as the right user with the least necessary privilege.
Let teams provision what they need within guardrails, without waiting for an infrastructure team to act. Use when provisioning requests queue and infrastructure work is reactive.
Sequence work by dependency to find which tasks actually determine the finish date, and manage those. Use when a project has many parallel workstreams and an immovable date.
Apply Python idioms where they cut bookkeeping and clarify intent, and refuse them where the clever form hides control flow.
Manage a code freeze with a scoped declaration, risk-classed changes, a fast exception process, and a planned thaw.
Identify what could derail a project, decide responses in advance, and track them rather than reacting when they happen. Use at project start and at every major checkpoint.
Run translation as a pipeline with catalogs, context, review, and continuous updates rather than a one-off handoff.
Choose the structure whose operations match how the data will actually be used, rather than defaulting to a list or a map.
Produce invoices and receipts that are accurate, immutable, and contain what tax authorities and customers require.
Design SDKs that feel idiomatic per language, handle auth and resilience, and version cleanly. Use when building an official client library for your API.
Organize documentation with the Diataxis four-quadrant model, navigable structure, and searchability. Use when structuring a docs site or when readers cannot find what they need.
Select Java collections by access pattern, use streams where they clarify, and default to immutability and records.
Test an MCP server for protocol conformance, tool selection accuracy, and behaviour under real model calls rather than only unit tests.
Design interfaces that survive translated text growing or shrinking substantially, without truncation or broken layout.
Remove or blunt identifiers so a dataset can be analysed or shared without re-identifying people, and know when it is only pseudonymous.
Understand why cash collected is not revenue earned, and record deferred revenue so financial reporting reflects delivery.
Manage prompts as versioned artefacts with review, rollback, and a record of what changed and why. Use when prompts affect production behaviour and more than one person edits them.
Run a lean startup team of agents (builder, marketer, analyst) through one disciplined weekly loop of ship, promote, and measure.
Push work to the edge so requests never touch your origin, using cache headers, invalidation, and origin shielding.
Cut garbage-collection pauses and overhead by lowering allocation rate, sizing the heap, and reading pause logs before touching a flag.
Tell a model what not to do in a way that works, by converting prohibitions into positive alternatives wherever possible.
Make switch and match statements provably total so adding a case forces every site to handle it, with no default that hides the gap.
Write Kotlin that uses null safety, data classes, and coroutines as designed, with clean Java interop. Use when writing Kotlin or migrating Java code and habits to it.
Design LLM tools with descriptions that steer, granularity that composes, and error returns the model can act on.
Construct test objects through builders and factories with valid defaults so each test states only the fields it cares about.
Read coverage as a map of untested risk that steers the next test, rather than enforcing a percentage target.
Write an Amazon six-pager, a narrative memo read in silence at the start of a meeting so a decision rests on reasoning rather than a presenter's charisma.
Connect device fleets with MQTT topic design, QoS chosen per message class, offline buffering, and safe fleet commands.
Combine result sets with UNION, INTERSECT, and EXCEPT, and know when a join or an exists clause is the better tool.
Audit third-party packages by pinning resolved versions, scanning against advisory databases, and catching malicious lookalikes before install.
Decide between prompting, retrieval, and fine-tuning with eval-first discipline and honest data requirements. Use when someone proposes fine-tuning or a prompt has hit its ceiling.
Route each request to the agent best suited to it, with a fallback and an explicit unknown path, instead of one generalist handling everything.
Cascade OKRs so team and individual goals ladder up to company strategy without sandbagged targets or scoring that rewards easy wins.
Break a project into milestones that each prove something, so progress is verifiable rather than asserted.
Return errors an agent can act on, distinguishing retryable failures from permanent ones and never leaking internals. Use when building MCP tools that will fail in production.
Place transformations before or after loading based on warehouse economics, governance, and reuse. Use when architecting a data platform or deciding where a transformation should…
Get the tone of an email right: warm, professional, and matched to the reader and situation, without the coldness email invites.
Apply normal forms pragmatically and denormalize deliberately, weighing update anomalies against read performance.
Deliver messages inside the product where they are contextual and non-interruptive, using the right surface for each urgency.
Reach battery-life targets through sleep-state budgeting, duty cycling, and measured (not estimated) current draw.
Write benchmarks that do not lie, controlling warmup, isolation, and statistics so the number reflects the code and not the environment.
Record decisions with their context and predictions to enable honest calibration and defeat hindsight bias. Use when making consequential decisions you want to learn from later.
Grow engineers through goal-anchored mentoring, calibrated stretch work, and questions before answers. Use when mentoring individuals or building a team's growth practice.
Separate planning from execution and replan when reality diverges, so a long task does not follow a plan that stopped being true.
Ask for, record, and honour consent so processing has a lawful basis and withdrawal actually stops it.
Apply changes locally before the server confirms, then reconcile or roll back honestly when it does not. Use when latency makes a correct-but-slow interface feel broken.
Drive API development from an OpenAPI spec as the contract, with linting, codegen, and drift detection.
Collect and keep only the personal data a feature actually needs, so exposure, cost, and compliance burden all shrink at once.
Choose the JavaScript/TypeScript bundler, test runner, and linter for a project by its type, not by fashion.
Configure tsconfig.json deliberately: the flags that matter, module and target settings, and build vs typecheck configs.
Find and eliminate N+1 database access where one query per row replaces one query for the set, using eager loading, batching, or a dataloader.
Summarise and cross-tabulate large tables quickly with pivots, keeping the source data clean and the refresh reliable. Use when aggregating data by category, period, or dimension.
Decide what to hand to an agent, what to keep with a human, and what to split, based on reversibility and consequence rather than capability alone.
Grant repository and organisation access through teams and least privilege, and review it before it drifts.
Reason honestly about at-most-once, at-least-once, and effectively-once semantics end to end. Use when specifying messaging behavior or auditing where a pipeline can lose or…
Treat tool output as untrusted data, confirm destructive actions, and keep an agent's reach inside what its user may do.
Deploy an LLM on vLLM with continuous batching, deliberate VRAM planning, and multi-model hosting that never overcommits the card.
Find the code that actually burns CPU time using a sampling profiler and a flame graph instead of guesswork.
Replace a hot algorithm with one of better complexity, a fitter data structure, or memoized results instead of micro-tuning the slow one.