i-m-senior-developer

This skill should be used when the user asks about "functional clarity", "функциональная ясность", "принципы функциональной ясности", fail-fast architecture, Error Hiding prevention, context-adaptive defaults, error handling patterns, refactoring principles, or asks to apply the Functional Clarity methodology. Also activates when the user says "apply our coding principles", "check against our style guide", "review this against functional clarity", "how should I handle errors", "рефакторинг", "обработка ошибок", or "code review against our standards". Provides a 22-principle methodology for building reliable, simple, and understandable software with emphasis on fail-fast error handling, explicit dependencies, and minimal cognitive load.

This skill should be used when implementing code per an existing architecture or PLAN document. Activates when the user asks to "implement", "code this up", "build the feature", "add the endpoint", "write the component", or in russian «реализуй», «закодь», «добавь endpoint», «сделай компонент». Mandates TDD via the `tdd-master:tdd-master` skill (RED-GREEN-REFACTOR before any production code), enforces minimal changes (FPF A.11), fail-fast error handling, and Functional Clarity principles. Loads stack-specific references (backend-python or frontend-react) on demand. Never designs systems; never reviews others' code.

This skill should be used when reviewing code changes — security issues, system-level defects, FPF/Functional Clarity violations, stack-specific pitfalls. Activates when the user asks to "review", "check this PR", "is this safe", "find bugs", "code review", or in russian «отревью», «проверь код», «есть ли проблемы», «code review». Operates on git-diff. Always loads `references/security.md` (OWASP Top 10 — non-negotiable). Loads stack-specific references (backend-python or frontend-react) on demand. Integrates with `functional-clarity:functional-clarity` for Error Hiding and FPF checks. Outputs a review report with file:line references; never modifies code itself.

This skill should be used when designing system architecture: bounded contexts, module boundaries, data flow, integration contracts, API shape, hand-offs between components. Activates when the user asks to "design", "architect", "split into modules", "where are the boundaries", "design the data flow", or in russian «спроектируй», «архитектура», «разбей на модули», «как разделить», when a feature README is presented before implementation, or when an existing system needs a new component. Loads stack-specific references (backend-python, frontend-react, api-design) on demand based on detected project signals. Produces an architecture document; never writes implementation code.

This skill should be used when the user asks to "integrate FPF", "add FPF to project", "FPF audit", "review decisions with FPF", "check evidence quality", "add decay mechanism", "check cognitive biases in decisions", "add FPF checklists to roles", "NQD check", "alternatives check", "evidence graph review", "bounded context audit", "terminology drift", mentions "First Principles Framework" or "FPF", or wants to improve decision-making quality in a multi-agent system.

This skill should be used when editing FICTION prose to detect AI-generation patterns — synthetic feel, problems with narrative distance, ready-made metaphors, implausible physiology, monotonous rhythm. Activates on "проверь прозу на AI-паттерны", "редактура художественного текста", "это звучит синтетически", "check this fiction for AI patterns", "AI prose detection". Suited for psychological thrillers, cyberpunk, any tense prose. Focuses on the principles by which AI produces synthetic text, not on mechanical "banned-word" search. For technical documents use clarity-validator instead; if unsure whether fiction-editing is wanted, ask the user.

This skill should be used when the user asks to "validate text", "check consistency", "проверь консистентность", "валидация текста", "проверка слоёв", "layer check", "clarity check", "проверь логику текста", "найди противоречия", "check cross-file consistency", "терминология дрифтит", "смешение слоёв", "проверь на синтетику", "AI-паттерны в тексте", "выдуманный флоу", "negation of nothing", "leaked invariant". Also triggered proactively by an orchestrating agent after review rounds, before docs-repo publication, and after batch edits to documentation. Catches 12 common AI-authoring antipatterns with FPF grounding.

This skill should be used when writing or reviewing RUSSIAN-language text — documentation, scripts, user-facing copy, replies — to make it read naturally instead of like a translated AI draft. Activates on "пиши по-русски естественно", "убери кальки", "это звучит как AI", "проверь русский стиль", "слишком формально", "упрости текст для разработчиков", "natural Russian", "no calques", "anti AI-slop". Four authoring rules distilled from real corrections: no foreign calques, no empty antitheses, plain language for engineers, intentional informal tone. NOT for English prose, and NOT for fiction (use ai-prose-detector for fiction).

This skill should be used when the user invokes /plan-reflect, says "reflect on the plan", "what did we learn", "post-mortem the planning", or in russian «отрефлексируй», «что пошло не так», «обнови контекст планнера», «сверь план с фактом». Activates after a session that ran /plan-do or otherwise produced a PLANNER_OUTPUT.md. Compares plan against reality using five evidence sources and updates planner-context.md with four learning types: gap-fill, model-strength signals, user-correction patterns, and cost-calibration deltas. Always emits a "Lessons learned" section, even if empty, as a force-function for honest retrospection.

This skill should be used when the user asks to "plan a task", "build an execution plan", "split this work", "make this faster/cheaper", or in russian «план», «разбей задачу», «распредели», «оптимизируй процесс», «как сделать быстрее/дешевле». Also activates when a feature README is presented before /plan-do, when an architecture session is starting, or when an existing PLAN document is about to be executed. Provides two modes: architecture planning (from a feature README) and execution planning (from an architecture document).

This skill should be used when the user asks to "generate llms.txt", "create llms-full.txt", "update llms.txt", "regenerate llms-full.txt", "create AI context files for this project", "обновить llms.txt", "сгенерировать llms-full.txt", "создать контекст проекта для AI", or mentions "llms.txt", "llms-full.txt", "llmstxt", "llmstxt.org standard", "project context for AI agents", "контекст проекта для AI". Also applicable when user asks about the llmstxt.org format, wants to know how to structure llms.txt, or asks "what should llms-full.txt contain".

This skill should be used when the user asks to "write tests", "add tests", "create test", "implement feature", "fix bug", "TDD", "test-driven", "reproduce bug", "write failing test", "Red-Green-Refactor", or when implementing any new functionality that requires testing. Provides TDD methodology based on Kent Beck and Uncle Bob principles.