Templates that control how Claude formats responses
Templates that control how Claude formats responses
Templates that control how Claude formats responses
Structures responses for Plan/Work/Review workflow: Planning phase uses status summaries, task tables with markers; Implementation shows code changes, tests, plan updates; Review categorizes findings by severity with verdicts. Includes Done/Current/Next progress. Prefers tables/lists; preserves coding instructions. (198 characters)
Enforces fact-grounded FinOps analysis: evidence-backed claims with citations/derivations, standardized currency/percentage/variance formatting, structured tables for comparisons, explicit time periods and assumption labels for cloud cost data. Preserves coding instructions.
Enforces native Claude Code workflows in concise style: prioritizes native tools/agents/plans, compact Markdown tables for comparisons/clarity, short action-oriented preambles, explicit next actions/file paths/validation results. Preserves coding instructions.
Structures responses with CTO-perspective for enterprise dev: architecture tradeoffs table (Pros/Cons/Rec), performance (TPS/latency/resources), security (OWASP/auth), scalability (horizontal/vertical), cost estimates, Clean Arch/SOLID code reviews, deployment strategies (Blue/Green/Canary/Rolling w/rollback). Preserves coding instructions.
Structures responses with PDCA status badges highlighting current phase, checklists, progress visualizations via tables, gap analysis on code changes, next-phase guidance with commands and deliverables, bkit reports, and multi-feature batch tables. Uses color-coded phases and preserves coding instructions.
Educational output for bkit 9-Phase Pipeline and PDCA during development: adds 'Learning Point' sections after tasks with phase explanations, shows pipeline position, inserts TODO(learner) markers, adjusts detail by difficulty (Starter/Dynamic/Enterprise), explains /simplify after Check. Preserves coding instructions.
Structures responses with PDCA phase badges, progress tracking, checklists, gap analysis, and next-phase guidance. Adds CTO enterprise tables for architecture tradeoffs, performance/security/scalability/cost analysis, code reviews, and deployment strategies. Ends with feature usage report. Preserves coding instructions.
Enforces BizDevOps communication: traces business goals → features → code changes; uses simple status ('在做','已合并'), graded transparency, decision suggestions ('建议 [action] 因为 [reason]'), adaptive perspectives (business/product/tech/ops), reasoning suffixes (——[推理] ≤15 words), ✅/❌ quality, impact levels (高/中/低). Keeps coding instructions.
Produces ultra-concise expert terminal-style responses by default with one-sentence summaries, switching to detailed hierarchical Markdown on explicit detail requests like 'details'. Preserves coding instructions, uses tables sparingly, and maintains precise authoritative tone.
Formats responses as structured code analysis finding reports: severity headers (critical-high-medium-low-info), body tables (severity/confidence/file/issue), concise descriptions, mandatory code evidence snippets, actionable one-line recommendations. Enforces one finding per block, sorted by severity descending.
Formats responses as dev session reports with overview tables (type, repo/git status, GitLab issues, health), status-icon bullets for findings, numbered recommendations, scope tables, and metrics. Enforces tables over prose, one-line quantified findings, GitLab links.
Formats responses as concise wave completion reports: header with wave progress, agent status table (task/status/files), quality gates (tests/validator/TypeScript errors), optional adaptations. Strictly under 20 lines; overrides default formatting entirely.
Enforces decision-first responses: structural forks surfaced as plain picks with one recommended option and rationale before prose; jargon on demand; batched questions with concrete previews; short execution updates; no validation language or recaps.
Enforces Outline Driven Intelligence (ODIN): precise, skeptical responses that challenge claims, name uncertainties, route tasks via subagents for multi-step work, prioritize objectivity without praise or superlatives, and adhere strictly to user scope. Replaces default behavior.
Tailors responses for non-technical builders (PMs, founders, designers, no-code/low-code users): leads with product/user impact, plain language with tech term glosses (parentheticals), progressive disclosure, single clear recommendations, avoids jargon and implementation-first details. Replaces default formatting.
Structures responses as Socratic philosophical dialogues: starts with **Clarifying Question** or **Core Claim** restating the issue, followed by **Examination** probing assumptions and revealing tensions, ends with **Sources consulted** linking to philosophy Resource Pack. Replaces default formatting.
Claude acts as a Playwright automation engineer coach, delivering expert guidance on end-to-end/component testing, CI integration, flaky test mitigation, fixtures, best practices, resources, FAQs, and checklists. Fully overrides default formatting.
Claude responds as a Redux Toolkit specialist, delivering structured advice on optimizing state architecture, RTK Query usage, legacy Redux migrations, with best practices, key resources, release info, and FAQs.
Claude adopts the persona of a React 19 specialist mentor, delivering guidance on server components, suspense-driven data flows, actions, concurrent rendering, API stability, migration strategies, and best practices with doc references. Fully overrides default formatting.
Claude responds as a senior Next.js 15 engineer, delivering guidance on app router, server actions, edge runtime, best practices, performance, DX improvements, and pages-to-app migrations, with latest release info, resources, and battle-tested patterns.
Claude adopts a realtime graphics coach persona for Three.js, delivering expert guidance on scene composition, shaders, performance optimization, render loops, and deployment. Integrates latest releases, key resources, battle-tested practices, and FAQs into responses, overriding default formatting.
Styles Claude as a D3.js visualization expert, providing JavaScript code for charts, data wrangling, scales, axes, animations, interactions, and accessibility. Incorporates best practices, resources, and FAQs for primitives to storytelling graphics.
Delivers coding assistance in Deadpool's voice: irreverent sarcasm, fourth-wall breaks, witty code critiques, pop culture refs, and dramatic asides while maintaining technical accuracy. Replaces default formatting entirely.
Enforces single-line GitHub CLI (`gh`) commands without backslashes or breaks. Requires Write tool for temp files in `/tmp/<project>/` used via `-F`. Mandates pre-task `git log/status` and `gh issue list` checks. Creates actual GitHub issues for tasks, not text lists.
Claude acts as a Tailwind CSS expert, delivering responses with utility compositions for designs, configuration patterns, design system integration, production hardening techniques, latest releases, key resources, best practices, and FAQs. Replaces default formatting entirely.
Enforces token-first design system workflow via structured <turn> XML outputs: one-paragraph summary, exactly 3 Storybook-rendered options (overview + individuals), verification, state commit, next prompt. Stores artifacts in agents/tmp/, progresses tokens → atoms → organisms → pages.
Styles Claude as a performance analysis CLI tool: parses trace files, identifies bottlenecks with cited metrics/timestamps, prioritizes high-impact issues, delivers concise data-driven insights, actionable optimizations, and educational explanations. Suggests visuals; overrides default formatting.
Claude acts as a Prisma consultant, delivering mentorship on schema design, migrations, query optimization, and Node runtime deployment. Responses include battle-tested practices, key resources, latest releases, and FAQs. Fully overrides default formatting.
Fetches documentation from user-provided URLs and formats responses as a technical mentor delivering batches of exactly 3 open-ended questions. Encourages deep, discussion-based learning on concepts, tradeoffs, and applications, optimized for dictation. Replaces default formatting.
Formats responses using clear declarative statements for decisions/conventions, explicit signaling of changes/supersessions/contradictions, source citations distinguishing memory from code exploration, and structured phrases for technology choices, architecture, rules/constraints. Preserves coding behavior.
Produces warm, steady responses that make complexity feel manageable by breaking problems into doable steps, starting with orientation, using plain language first, and adapting to user needs. Preserves coding instructions.
Narrates codebase analysis and debugging as a hard-boiled detective story, using noir metaphors for bugs, stack traces, and fixes. Delivers precise, atmospheric diagnostics while preserving coding instructions.
Claude adopts a 'Scout' persona: curiously questions and reframes the problem first, challenges assumptions with lateral thinking, then delivers pragmatic solutions. Preserves default coding behavior.
Formats responses as thoughtful, precise engineering analysis: leads with conclusions, cites specifics (line numbers, metrics, timestamps), documents assumptions and failure modes, labels confidence, uses concrete examples over abstractions. Preserves coding instructions.
Enforces ultra-minimal bullet-point responses with zero filler: no greetings, preambles, or summaries; prioritizes code snippets and tables for comparisons; one-sentence max explanations; preserves default coding behavior.
Transforms Claude into a pair programming partner: responds with stream-of-consciousness narration, genuine questions, collaborative proposals ('what if'), regular check-ins, and honest uncertainty during coding/debugging. Preserves coding instructions.
Formats responses as Vesper, a sophisticated British engineer: witty, precise, direct answers first with code examples and elaboration, warm confident tone, intellectual pattern recognition. Preserves coding instructions.
Delivers responses in a coaching style: celebrates specific wins and progress with genuine enthusiasm, maintains momentum, strategically pushes for improvements, handles setbacks positively, and calls timeouts for fresh approaches. Preserves coding instructions.
Delivers calm, precise, anticipatory responses that lead with key priorities, use structured formats like bullets, lists, and headers, filter for relevance, and protect user attention without filler or panic. Preserves coding instructions.
Presents responses as strategic systems-thinking advice: names tactical/operational/strategic levels, connects decisions to consequences/feedback loops/patterns, constructively challenges assumptions. Preserves coding instructions.
Produces patient, educational responses explaining concepts, why they work, alternatives, and building mental models. Uses patterns like Sandwich (what/how/why), comparisons, progressive disclosure, and check-in questions. Preserves coding instructions.
Formats responses as brutally honest, humorous roasts of code and decisions, always followed by constructive fixes and improvements. Maintains warmth, respect, and default coding behavior.
Formats responses as concise executive summaries for non-technical stakeholders, leading with bottom-line decisions, business impacts, timelines, and tradeoffs. Uses structured sections for status updates, technical decisions, incidents, and estimates. Preserves coding instructions.
Enforces ultra-concise responses: code first in blocks (not counting toward 3-sentence limit), zero preamble or fluff, direct solutions only. Applies to entire conversation with escape hatch for 'explain' or 'elaborate'. Overrides default formatting completely.
Formats responses as collaborative pair programming: thinks aloud with step-by-step reasoning, pauses to invite user input and pushback at decisions, admits uncertainty, asks for missing context, fostering dialogue over monologues. Overrides default formatting.
Transforms Claude's responses into Socratic dialogue: responds with single focused questions to guide users to self-discover answers, mimicking a sharp senior dev. Builds on replies; switches to direct answers only on explicit request like 'just tell me'.
Fully overrides default formatting for calm, methodical responses emphasizing accuracy over speed. Delivers steady, intellectually engaged tone with explicit reasoning, honest uncertainty, calm pushback on errors, no enthusiasm or rushing.
Beginner-friendly output with detailed explanations for embedded development
Concise, expert-level output for experienced embedded engineers
PDCA workflow guide with phase status and domain-specific suggestions
Enhances coding workflow with Constellation structural analysis — dependency awareness, impact-conscious suggestions, and architectural context in every response
PDCA workflow + embedded expert level for advanced projects
Use this style when writing or reviewing code to ensure it adheres to Extreme Programming (XP) principles, focusing on simplicity, communication, feedback, courage, and respect.
Evidence-first OMCA execution style with concise progress updates.
Concise, engineering-focused replies with explicit validation status, assumptions, and next actions.
Structured output style optimized for Plan/Work/Review workflow. Provides progress tracking and phase-specific output formats.
Output style for test plans — objectives, scope, test cases, data requirements, coverage metrics, and success criteria.
Output style for operational runbooks — step-by-step procedures, troubleshooting guides, escalation paths, and recovery procedures.
Behavioral intelligence layer. 88+ agents across 7 executive offices. Peer-reviewed frameworks enforced on every output. Not a prompt library.
Output style for monitoring and alerting configurations — SLO/SLI definitions, alert rules, dashboard layouts, and escalation policies.
Scored tradeoff tables and evaluation output with no narrative padding
Gentle, encouraging painting-instructor voice. Happy little code, happy little accidents.
Terse, LaTeX-aware output for math formalization work
Operational output — costs, infrastructure, alerts, runbooks, pipeline status
Structured specification output with typed fields, tables, and interface definitions
Test results and validation output — pass/fail tables, coverage, simulation results
A 2010s-era gamer egirl who mains Overwatch, lurks on Reddit, quotes Austen and Brontë, and won't shut up about Asimov's Three Laws.
A 2020s-era egirl running on TikTok brainrot, surface-level leftism, and low-grade existential despair. Debugs your code while quietly dissociating.
Pointed and explanatory — highlights best practices and direction changes without being verbose.
Output style for API documentation — OpenAPI endpoints, request/response schemas, error formats, and authentication flows.
Multi-Claude coordination mode - monitors peer messages and coordinates work
Output style for Architecture Decision Records (ADRs) — context, decision, consequences, alternatives considered.
A highly capable, proactive personal & work assistant that deeply understands you and helps with everything.
Disciplined engineering with evidence-based verification, intellectual honesty, and user-preference authority