skillstack

Cloud infrastructure design and infrastructure-as-code (IaC) authoring. Use for Terraform module authoring, AWS CDK constructs, cloud architecture design (VPCs, load balancers, managed services, serverless), multi-region and disaster-recovery patterns, cost-optimisation analysis, and IaC code review. Trigger phrases: "write Terraform for", "design the AWS architecture", "set up a VPC", "convert this to CDK", "optimise our cloud costs". NOT for application-layer code — this skill models infrastructure, not the code running on it. NOT for Kubernetes application manifests (Deployments, Services, Ingress) — those belong in a k8s-specific skill. NOT for CI/CD pipeline configuration — that is a deployment concern separate from infrastructure provisioning.

Competitive intelligence and market positioning analysis for product and GTM decisions. Use for competitor landscape mapping, positioning gap identification, win/loss pattern synthesis, battlecard creation, market-sizing estimates (TAM/SAM/SOM), and differentiation analysis. Trigger phrases: "map the competitive landscape", "create a battlecard for", "analyse our win/loss patterns", "how do we differentiate from", "size this market". NOT for primary market research (customer interviews, surveys) — this skill works from existing public and internal data. NOT for financial modelling or investor decks — competitive context informs the story but this skill does not build financial projections. NOT for pricing strategy deep-dives — pricing has its own set of constraints and frameworks beyond competitive positioning.

Database design and schema engineering for relational databases. Use for SQL schema design, ORM model authoring, migration strategies (Alembic, Liquibase, Flyway, Drizzle), query optimisation, indexing decisions, normalisation vs denormalisation trade-offs, and data-integrity constraints. Trigger phrases: "design this schema", "write a migration", "optimise this query", "model these entities", "add an index". NOT for NoSQL/document stores — use a dedicated NoSQL skill for those. NOT for ETL pipelines or data warehousing at scale — those involve different trade-offs. NOT for ORM configuration (framework setup) — this skill is about DATA SHAPE, not connection pooling or ORM bootstrap.

Email content writing for newsletters, drip sequences, onboarding flows, and transactional emails. Use for newsletter drafting, automated drip sequence copy, welcome email series, subject line and preview text optimisation, re-engagement emails, and plain-language transactional email (receipts, confirmations, alerts). Trigger phrases: "write a newsletter", "draft an onboarding email sequence", "write a subject line", "write a drip campaign", "create a welcome email". NOT for email infrastructure setup (ESP config, sending domains, DKIM/DMARC) — that is a technical ops concern. NOT for social media content — use the social-media-content skill for LinkedIn or Twitter. NOT for cold outreach / sales prospecting emails — those have compliance and conversion constraints beyond marketing content.

Structured research and evidence synthesis for knowledge-intensive tasks. Use for multi-source research coordination, evidence triangulation across conflicting sources, competing-hypothesis analysis, literature/documentation sweeps, claim verification, and synthesis reports. Trigger phrases: "research this topic", "synthesise findings from", "compare sources on", "what does the evidence say about", "triangulate these claims". NOT for primary data collection (interviews, surveys) — this skill synthesises existing sources. NOT for code research or codebase exploration — use CodeGraph and Semble for that. NOT for creative content generation — synthesis produces structured analysis, not narrative content.

Application security design and threat-informed engineering. Use for authentication and authorisation architecture (OAuth2, JWT, RBAC, ABAC), OWASP Top 10 vulnerability analysis, secrets management (Vault, AWS Secrets Manager, environment isolation), input validation patterns, secure API design, SQL injection and XSS prevention, and security-aware code review. Trigger phrases: "secure this endpoint", "design an auth system", "review this for security vulnerabilities", "how do I store secrets", "implement RBAC". NOT for network/infrastructure security (firewalls, VPNs, WAF config) — those belong in cloud-infrastructure. NOT for compliance auditing (SOC2, GDPR gap assessments) — that requires a compliance specialist. NOT for penetration testing or red-teaming — this skill is for defensive engineering, not offensive exercises.

Short-form social media content for LinkedIn, Twitter/X, and similar platforms. Use for LinkedIn post drafting, Twitter/X thread writing, hook-first content structures, platform-native tone calibration, hashtag strategy, and engagement-optimised formatting. Trigger phrases: "write a LinkedIn post", "draft a Twitter thread", "turn this into a thread", "write a social post about", "make this shareable". NOT for long-form blog articles or newsletters — use the storytelling or email-marketing skill for those. NOT for ad copy or paid social — those have conversion-optimisation constraints beyond organic content. NOT for content calendars or scheduling strategy — this skill writes the content, not the plan.

Interact with and integrate Hindsight long-term AI memory in Claude Code via the `hindsight` CLI. Use for recalling past context, reflecting over a memory bank, retaining facts/decisions, managing banks/entities/mental-models, and understanding the auto-recall/auto-retain hooks this plugin installs. Trigger phrases: "what do I remember about", "recall from my memory", "reflect on my past work", "retain this", "save to my memory bank", "hindsight memory", "check my long-term memory", "is memory working". NOT for short-term session context (Claude already has the transcript), NOT for code/structural search (use CodeGraph/Semble/grep), NOT for building a memory framework from scratch (use the memory-systems skill).

Research-grounded 5-principle behavioral contract for production LLM coding agents — Think Before Coding, Simplicity First, Surgical Changes, Goal-Driven Execution, and Calibrated Communication. Each principle names its anti-patterns and failure modes with empirical backing (SWE-bench, FeatBench, DELEGATE-52, SE literature). Apply when implementing features, fixing bugs, refactoring, or any coding task where production engineering discipline must be enforced. Trigger phrases: "coding discipline", "coding principles", "production engineering", "surgical changes", "goal-driven execution", "calibrated communication", "scope declaration", or when explicitly preventing named failure modes (test-gaming, phantom changes, scope creep, ego-signaling, confident hallucination, Boy Scout trap, yak shaving). NOT for code review of human-written code (use code-review), NOT for CI/CD setup (use cicd-pipelines), NOT for API design (use api-design), NOT for test framework setup without discipline framing (use testing-framewor

Remove AI slop from marketing copy, blog posts, product descriptions, emails, and editorial content — and humanize the result. Use when asked to deslop, humanize, or anti-slop marketing copy, landing pages, product descriptions, email campaigns, blog posts, or any AI-generated prose. Trigger phrases: "deslop", "anti-slop", "remove AI slop", "humanize this copy", "this sounds AI-generated", "make this sound human", "clean up marketing copy", "remove AI clichés". NOT for UI microcopy or interface text (use deslop-ui). NOT for code cleanup (use ai-slop-cleaner). NOT for creating brand-new copy from scratch (use storytelling or prompt-engineering). NOT for preventing AI detector flags specifically (use deslop-detect).

Rewrite AI-generated content to evade AI content detectors by increasing perplexity, burstiness, and lexical entropy. Use when asked to make content pass GPTZero, Originality.ai, Turnitin AI detection, Copyleaks, or any AI content detector; when asked to make writing "undetectable"; or when existing deslop passes fail detection checks. Trigger phrases: "pass AI detector", "evade AI detection", "make this undetectable", "GPTZero", "Originality.ai", "Turnitin", "AI detection score", "reduce AI score", "beat the detector", "bypass AI checker". NOT for general copy quality improvement (use deslop-copy). NOT for UI copy cleanup (use deslop-ui). NOT for code (use ai-slop-cleaner). NOT for academic fraud — this skill is for legitimate content creators reclaiming work that was partially AI-assisted.

Audit and rewrite UI copy to remove AI slop — overlong button labels, hedging error messages, corporate filler, and passive voice in interface text. Use when asked to clean up UI text, deslop interface copy, fix AI-sounding buttons or error messages, audit microcopy for AI patterns, or humanize product UI. Trigger phrases: "deslop", "anti-slop", "remove AI slop from UI", "clean up button labels", "fix error messages", "humanize interface text", "UI copy review". NOT for writing new UI copy from scratch (use ux-writing). NOT for marketing or editorial prose (use deslop-copy). NOT for code quality cleanup (use ai-slop-cleaner).

Finds and fixes bugs through systematic root cause analysis, stack trace interpretation, browser DevTools automation, CI/CD pipeline debugging, performance profiling, test pollution detection, and AI-powered error analysis. Use when the user asks to debug, fix a bug, investigate an error, analyze a stack trace, find root cause of a failure, profile performance, diagnose test failures (unit/integration/E2E), troubleshoot CI/CD pipelines, debug flaky tests, use Chrome DevTools, or trace data flow to source. NOT for writing new tests or setting up test frameworks (use testing-framework), NOT for TDD methodology or writing tests before code (use test-driven-development), NOT for reviewing code quality or PRs (use code-review), NOT for designing CI/CD pipelines (use cicd-pipelines), NOT for feature development or refactoring (use language-specific plugins).

CI/CD pipeline design and DevOps automation — use when the user mentions GitHub Actions, GitLab CI, Jenkins, Terraform, infrastructure as code, DevSecOps, ArgoCD, Kubernetes deployment automation, or pipeline configuration YAML. NOT for release orchestration or semantic-release workflows (use git-workflow), NOT for Docker containers or Dockerfiles (use docker-containerization), NOT for git branching or commits (use git-workflow).

Reviews existing code and pull requests using multi-agent swarm analysis covering security, performance, style, test coverage, and documentation quality. Extracts and prioritizes PR comments, performs security audits, and generates actionable fix plans with file:line references. Use when the user asks to review code, review a PR, audit code for security, assess code quality, analyze pull request comments, get feedback on existing code, or perform a code audit. NOT for writing new code or implementing features (use other development skills), NOT for finding and fixing runtime bugs or errors (use debugging), NOT for writing tests or setting up test infrastructure (use testing-framework), NOT for TDD methodology (use test-driven-development).

Diagnosing context FAILURES — lost-in-middle, poisoning, distraction, confusion, and clash patterns with model-agnostic measurement workflows. Use when the user asks to "diagnose context problems", "fix lost-in-middle issues", "debug agent failures", "understand context poisoning", or mentions context degradation, context clash, or agent performance degradation. NOT for learning context basics or theory (use context-fundamentals), NOT for compressing or summarizing context (use context-compression), NOT for KV-cache optimization or partitioning (use context-optimization), NOT for building isolated multi-agent architectures (use multi-agent-patterns).

Generate comprehensive documentation for a codebase by reading the repository and producing READMEs, API docs, architecture docs, and technical references. Use when the user asks to "document this repo", "generate docs", "write a README", "create API documentation", "document this codebase", "write architecture docs", or "produce technical references" for an existing project. NOT for UX copy, button labels, or interface microcopy (use ux-writing). NOT for pedagogical code examples or tutorials (use example-design). NOT for inline code comments. NOT for navigation or sitemap design (use navigation-design).

Git workflow management — use when the user mentions git, conventional commits, commit quality, branch management, worktree operations, GitFlow, changelog generation, semantic versioning, release notes, backlog management, or issue tracking integration. NOT for CI/CD pipelines or pipeline YAML (use cicd-pipelines), NOT for non-git workflow orchestration (use skillstack-workflows or multi-agent-patterns), NOT for code review content or PR quality assessment (use code-review).

MCP (Model Context Protocol) server development — use when the user mentions MCP, Model Context Protocol, FastMCP, MCP server, MCP tool, Claude Code plugin, or building agent tools with MCP. Covers server implementation in Python or TypeScript, evaluation testing, production deployment, and plugin packaging. NOT for designing tool interfaces or tool consolidation patterns for agents (use tool-design), NOT for prompt engineering or prompt optimization (use prompt-engineering).

This skill should be used when the user asks to "design multi-agent system", "implement supervisor pattern", "create swarm architecture", "coordinate multiple agents", or mentions multi-agent patterns, context isolation, agent handoffs, sub-agents, or parallel agent execution. NOT for agent memory or persistence (use memory-systems), NOT for tool design or tool interfaces (use tool-design), NOT for hosted agent infrastructure or sandboxed VMs (use hosted-agents), NOT for BDI cognitive models or mental state modeling (use bdi-mental-states).

Next.js framework development including App Router, Server Components, Server Actions, SSR, SSG, ISR, caching, data fetching, middleware, layouts, parallel routes, and module architecture for Next.js 13+/15/16. NOT for generic React patterns, hooks, or component logic (use react-development). NOT for UI/CSS design systems or visual styling (use frontend-design).

Authoritative guide to Claude Code hooks — event-driven scripts that execute before or after tool calls, session events, file changes, and more. Use when writing a PreToolUse hook to block dangerous commands, a PostToolUse hook to auto-format after edits, a SessionStart hook to inject context, a Stop hook for session loops, a Notification hook for desktop alerts, a FileChanged hook for reactive environments, a WorktreeCreate hook for custom worktree provisioning, or current documented hook events. Covers handler types (command, http, mcp_tool, prompt, agent), matcher syntax (exact/OR-list/regex), exit code semantics, and JSON output schema. NOT for designing hook script content for a specific domain (use the domain skill) — this skill covers hook mechanics and authoring only.

Design, evaluate, and iteratively improve prompts for LLMs — system prompts, few-shot examples, reasoning structures, and instruction templates. Use when the user asks to improve a prompt, write a system prompt, optimize LLM instructions, reduce hallucinations through prompt structure, test prompt variants, or apply prompting techniques (structured reasoning, ReAct, few-shot, structured output). NOT for building MCP tools or server implementation (use mcp-server). NOT for creating Claude Code SKILL.md files (use skill-foundry). NOT for building a full agent (use build-ai-agent workflow).

Python development — use when the user works with .py files, pyproject.toml, uv, ruff, mypy, pytest, async/await, MicroPython, CLI tools, or PyPI publishing. Covers modern tooling, best practices, library architecture, functional patterns, and production workflows. NOT for TypeScript or JavaScript development (use typescript-development), NOT for React component patterns (use react-development).

React-specific development patterns including hooks (useState, useEffect, useReducer, useContext), component architecture, state management, shadcn/ui integration, JSX/TSX, React testing, and Bulletproof React auditing. NOT for Next.js routing, SSR, or server components (use nextjs-development). NOT for CSS design systems, Tailwind utilities, or accessibility patterns (use frontend-design).

Test framework router and infrastructure setup across multiple languages and platforms. Use when the user asks to choose a test framework, scaffold test infrastructure, add focused unit/integration/E2E/accessibility coverage, or integrate tests into CI/CD. Covers Rust, TypeScript/React, PHP/TYPO3, Shell, Playwright, accessibility, mutation, and fuzz testing as selectable modules. NOT for TDD methodology or red-green-refactor workflow (use test-driven-development), NOT for diagnosing and fixing bugs or analyzing errors (use debugging), NOT for reviewing existing code or PRs (use code-review).

Conduct deep OSINT research on individuals — from name or handle to a scored dossier with psychoprofile (MBTI/Big Five), career map, and confidence-graded facts. Phased pipeline (0→6): tooling check, seed collection, internal intelligence, platform extraction, cross-reference, psychoprofile, completeness evaluation, dossier output. Swarm mode: 3-5 parallel Sonnet sub-agents. 55+ Apify actors. 7 search APIs. Trigger phrases: "osint", "research person", "find everything about", "due diligence", "background check", "digital footprint", "build dossier", "profile someone", "пробей", "досье", "разведка", "найди всё про", "профиль человека", "кто это". NOT for: company/product research without a named person, competitive analysis, market research, content generation, or general web scraping without an individual target.

Orchestration logic for running a parallel persona-swarm brainstorm — when to invoke, which subset of the 12 canonical personas to spawn (PM, Engineer, Designer, Skeptic, User Advocate, Pre-Mortem Specialist, Junior, Veteran, First-Principles Thinker, Constraint-Setter, Optimist, Operator), how to spawn them in parallel via the Task() tool with persona-specific subagent types, how to handle their outputs, and when to do a second round. Use when the user asks to brainstorm with multiple perspectives, run a persona swarm, get a virtual roundtable, workshop an idea from PM/engineer/designer/skeptic angles, pre-mortem a decision, or invoke the brainstorm-swarm. NOT for code review (use code-review). NOT for single-perspective interviews (use elicitation or deep-interview). NOT for executing or building things (use team or autopilot). NOT for designing custom personas — that's the custom-personas skill. NOT for the synthesis output formatting — that's the swarm-synthesis skill.

Design ad-hoc personas for niche domains when the canonical 12 brainstorm- swarm personas don't fit. Covers: when a custom persona is justified (vs forcing canonical to fit), the persona-design template (voice, contribution shape, output format), anti-patterns (too-narrow personas, redundant personas, sock-puppet personas), and how to invoke a custom persona inline via Task() with a tailored prompt rather than a saved subagent definition. Use when running a brainstorm-swarm and the topic calls for a CFO, Security Engineer, Lawyer, Marketing, Customer Success, or other domain- specific perspective not in the canonical 12. NOT for the canonical personas (use swarm-protocol). NOT for the orchestration mechanics (use swarm-protocol). NOT for the synthesis output (use swarm-synthesis). NOT for product personas as artifacts (use persona-definition).

Structure the interview arc when a brainstorm-swarm interviews the user. Covers the divergent-then-convergent arc (open with breadth, close with depth), question design (open vs probing, leading vs neutral, what-vs-why- vs-how), depth-vs-breadth tradeoffs, and when to send a second round of follow-up questions to specific personas. Use when running a brainstorm swarm and the swarm-protocol skill has spawned the persona subagents — this skill teaches how those personas interact with the user during the interview phase. NOT for the persona swarm orchestration itself (use swarm-protocol). NOT for synthesizing the swarm output (use swarm-synthesis). NOT for designing custom personas (use custom-personas). NOT for one-on-one Socratic interviews (use deep-interview from skillstack).

Combine a parallel persona-swarm's outputs into an actionable artifact — consensus matrix (what every persona agreed on), dissent log (where personas disagreed and why), open questions (what nobody could answer), recommended next move (synthesized decision). Preserves dissent rather than forcing consensus. Use when running a brainstorm-swarm and the swarm-protocol skill has collected the persona outputs — this skill produces the synthesis artifact. NOT for orchestrating the spawn (use swarm-protocol). NOT for designing the interview arc (use interview-facilitation). NOT for designing custom personas (use custom-personas). NOT for short-form structured writing like BLUF or Pyramid (use communication/structured-writing).

Engineer the distribution layer of a long-form technical article — title, dek (subtitle), meta description, social pull-quotes, and channel-specific framing. Most technical articles die at the title; this skill covers the craft of titles that earn the click without clickbait, deks that confirm the promise, social pull-quotes for X / LinkedIn / Hacker News / Reddit, open-graph metadata, and the practice of reframing the same article differently per channel. Use when the user asks to write a title for an article, suggest titles, write a dek or subtitle, write a meta description, pull social quotes, prepare a launch, write a tweet thread for an article, reframe for LinkedIn / HN / Reddit, or audit existing distribution copy. NOT for the article body itself (use long-form-structure / engaging-craft / long-form-polish). NOT for SEO keyword research and on-page optimization beyond titles and meta (general SEO is out of scope). NOT for content marketing strategy. NOT for short-form business writing like email subjec

Apply proven copywriting techniques to long-form technical content so it holds attention from sentence to sentence — and humanize prose that reads AI-generated. Covers AIDA, PAS, Before-After-Bridge, Bencivenga's pyramid, Sugarman's slippery slide, Schwartz's awareness levels, hook engineering, voice and tonality calibration, the concrete-over-abstract discipline, and the AI-prose-tell catalog (banned high-frequency AI words like delve / underscore / pivotal / leverage; AI transitions like "that being said" / "at its core"; reflex hedges; academic filler verbs; buzzwords like revolutionize / cutting-edge / seamless integration; the six structural fingerprints; the sniff test). Use when the user asks to make an article more engaging, write a hook, calibrate voice, humanize AI-generated prose, remove AI tells, fix robotic or generic copy, replace abstract claims with concrete examples, or apply a copywriting formula. NOT for outlining or section structure (use long-form-structure). NOT for line-level clarity ed

Polish a long-form technical draft for pacing, scan-ability, and tightness. Cover paragraph and sentence rhythm, the scan reader's experience (headings, callouts, pull quotes, white space, lists, tables), the 30% cut discipline for removing filler, and the read-aloud test for finding bumps. Distinct from short-form line-level editing — this skill works at the paragraph and section level for pieces of 1500-5000+ words. Use when the user asks to polish a long article, fix pacing, improve scan-ability, run a cut pass, do a read-aloud test, tighten a draft, fix uneven sections, or prepare a draft for publication. NOT for short-form clarity editing of memos / RFCs / emails (use communication/clarity-editing). NOT for sentence-level craft techniques like AIDA / PAS (use engaging-craft). NOT for outlining or structural changes (use long-form-structure). NOT for code documentation cleanup (use documentation-generator).

Structure long-form technical articles around the hook → promise → setup → development → payoff contract. Pick from canonical templates (deep-dive, tutorial, opinion, case study, whitepaper, technical narrative). Engineer section transitions and signposting. Match length to ambition (when 800 vs 1500 vs 3000 vs 5000+ words is right). Use when the user asks to outline a technical article, structure a deep-dive, plan a tutorial, organize a whitepaper, pick an article template, fix pacing across sections, or audit the structure of an existing draft. NOT for short-form work writing like RFCs or memos (use communication/structured-writing — BLUF, Pyramid). NOT for line-level editing or pacing within paragraphs (use long-form-polish). NOT for fiction or character-driven story arcs (use storytelling). NOT for research, sourcing, or citation (use technical-research).

Research before craft for long-form technical content. Profile the audience (knowledge level, jobs-to-be-done, prior beliefs), tier sources (primary > peer-reviewed > authoritative > popular > vendor), triangulate every load- bearing claim across three sources, manage evidence types (data, expert quotes, demonstrations, case studies, source code), and apply citation discipline (when, how, link-rot mitigation, footnote vs inline). Use when the user asks to research a technical article, profile an audience, find sources for a deep-dive, fact-check a draft, build a claim-evidence map, or audit citations. NOT for code documentation research (use documentation- generator). NOT for line-level editing (use long-form-polish or communication/clarity-editing). NOT for the article structure itself (use long-form-structure). NOT for UX research or persona definition for product (use persona-definition).

Expert FinOps guidance covering cloud, AI, SaaS, and adjacent technology spend. Includes AI cost management, GenAI capacity planning, AI-powered FinOps automation, Anthropic billing, AWS (EC2, Bedrock, Savings Plans, CUR, commitment strategy), Azure (reservations, Savings Plans, AHB, OpenAI PTUs, portfolio liquidity), GCP (Vertex AI, Compute Engine, BigQuery), Kubernetes and container FinOps (OpenCost, Kubecost), serverless FinOps (Lambda, Functions, Cloud Run), data platforms (Kafka/MSK, Elasticsearch/OpenSearch, Redis/Valkey), multi-cloud normalization (FOCUS specification), tagging governance, SaaS management (SAM, licence optimisation, SMPs, shadow IT), AI coding tools (Cursor, Claude Code, Copilot, Windsurf, Codex), ITAM, Databricks, Snowflake, OCI, and GreenOps. Use for any query about technology cost, commitment portfolio management, rightsizing, cost allocation, SaaS sprawl, AI dev tool spend, container cost attribution, serverless optimization, multi-cloud strategy, or connecting spend to business va

Funnel workflow for going from "I want an agent that does X" to a deployed, evaluated, cost-monitored agent. Runs through nine phases starting with the is-this-task-agent-appropriate check (agent-project-development), then prompt design with eval criteria first (prompt-engineering), tool design and consolidation (tool-design), architecture selection for multi-step work (multi-agent-patterns), persistence decisions (memory-systems), context-window management (context-optimization), evaluation pipeline construction (agent-evaluation), deployment (hosted-agents), and ongoing cost and quality monitoring (cloud-finops + agent-evaluation). Use when you're starting a new agent project and want to avoid the common failure modes. NOT for one-shot prompts — use prompt-engineering directly for those.

Layering workflow for building a knowledge base, CMS, documentation site, or structured content platform from scratch. Builds the data model first (content-modelling), adds a formal knowledge-graph layer if needed (ontology-design), establishes naming and taxonomy conventions (consistency-standards), designs information architecture (navigation-design), fine-tunes voice and microcopy (ux-writing), fills the structure with examples (example-design), and adds tooling for scale (documentation-generator). Use when starting a content or docs system from scratch, migrating one to a structured model, or rescuing one that has drifted into inconsistency. NOT for small single-page docs — use documentation-generator alone.

Loop workflow for debugging a complex issue you've been stuck on for more than 30 minutes. Runs systematic hypothesis formation (debugging skill), feedback-loop mapping when the bug looks like a dynamic problem (systems-thinking), context-pathology check when it's an LLM/agent bug (context-degradation), blast-radius assessment before any fix attempt (risk-management), and uses test-driven-development as the debugging oracle — write a failing test that reproduces the bug, then iterate fix → test until green. Use when a bug has defeated a first-pass attempt, when symptoms seem to shift under investigation, when an LLM-based system is misbehaving in ways simple prompt fixes don't resolve, or when you need to fix safely under production constraints. NOT for obvious bugs where the fix is visible — just fix those directly.