Initializes, repairs, and checks a repository Memory Bank with a minimal canonical file set, routed loading, evidence-based templates, retention rules, and additive role-specific extensions. Preserves every existing file and creates only missing files for durable repository work. USE FOR: initialize memory bank, create .memory-bank, missing memory bank, bootstrap project memory, create index.md, create projectbrief.md, create productContext.md, create activeContext.md, create techContext.md, create progress.md, create systemPatterns.md, create promptHistory.md, repair incomplete memory bank, check memory bank health, stale project memory. DO NOT USE FOR: read-only questions, clarifications, transient session notes, replacing an existing complete Memory Bank, VS Code native memory.
Debug and troubleshoot Sampler-based PowerShell module builds and Pester 5 test failures. Covers running builds safely (without freezing VSCode), reading Pester results, diagnosing common mock/assertion failures, fixing tests, and reproducing a CI-only failure at the exact commit CI built. USE FOR: build fails, test fails, Pester errors, troubleshoot build, debug tests, run build, Sampler build, module build, Pester mock issues, VSCode freezes during build, passes locally but fails in CI, CI-only test failure, reproduce a CI failure, clean checkout reproduction, GitVersion 0.0.1 fallback version. DO NOT USE FOR: creating new modules from scratch, Azure deployments, CI/CD pipeline config.
Assembles paginated evidence packages (Anlagen) for authorities, courts, and insurers on Windows: verifies documents against the claims made about them, decides which sheets may be omitted, renders a Markdown cover with a locator index to PDF via pandoc and headless Edge, merges with pypdf, and verifies page ranges, text layer, and the sources' own page numbering. USE FOR: Anlage bauen, Beweispaket, Belege zusammenstellen, evidence package, exhibit bundle, Deckblatt, Fundstellenverzeichnis, locator index, Markdown nach PDF, markdown to PDF on Windows, pandoc Edge headless, print-to-pdf erzeugt keine Datei, Edge headless exit 0 kein PDF, PDFs zusammenfuehren, welche Seiten darf ich weglassen, Seiten aus Kontoauszug entfernen, Schwaerzung, redact bank statement, Seitenzaehlung lueckenlos. DO NOT USE FOR: extracting text from PDFs (use pdf-to-markdown), Markdown to Word (use pandoc-docx-export), slides (use marp-slide-overflow), legal or tax argumentation (use german-legal-research, german-tax-research).
Legal research and statement drafting for German law (Deutsches Recht). Specializes in tenancy/rental law (Mietrecht), property management (Immobilienverwaltung), operating cost disputes (Betriebskostenstreit), landlord-tenant relationships, and general civil law under the BGB. Produces structured Schriftsätze, Aufforderungsschreiben, and Rechtsgutachten in German with proper citation and argumentation. USE FOR: German law, Mietrecht, tenancy disputes, Betriebskosten, Nebenkostenabrechnung, Mietminderung, Kündigung, Schönheitsreparaturen, Kaution, Mängel, Instandhaltung, Verkehrssicherungspflicht, Schriftsatz, Abmahnung, BGB, BetrKV, WEG, Mietvertrag, Zivilrecht, Wohnraummietrecht, Hausverwaltung, Rauchwarnmelder, Dachrinnenreinigung, Gartenpflege. DO NOT USE FOR: Strafrecht, Steuerrecht (use german-tax-research), Ausländerrecht, Familienrecht (unless intersecting tenancy), international law, non-German jurisdictions.
German income tax (Einkommensteuer) case work: reviews an assessment notice, computes the objection deadline, answers a Belegaufforderung point by point, reconciles every figure against the transmitted return, and drafts Einspruch, Begründung, and applications in formal German with exact § citations. USE FOR: Einkommensteuer, Steuererklärung, ELSTER, Steuerbescheid prüfen, Einspruch, Einspruchsbegründung, Belegaufforderung, Einspruchsfrist, Bekanntgabefiktion, Aussetzung der Vollziehung, Verspätungszuschlag, Schätzung, Anlage V, Vermietung und Verpachtung, AfA, § 7i Denkmal, Werbungskosten, Arbeitszimmer, Homeoffice-Pauschale, Fortbildungskosten, Reisekosten, Arbeitsmittel, Sonderausgaben, Kapitalerträge, Lebensmittelpunkt, § 153 AO, German income tax return. DO NOT USE FOR: Miet- or Arbeitsrecht argumentation (use german-legal-research), building the Anlagen PDF (use evidence-package-assembly), reading a PDF or XLSX (use pdf-to-markdown, xlsx-to-markdown), tax law outside Germany.
Builds evaluations for your own Copilot skills, prompts, and agents so changes are measured, not vibed. Starts from the native VS Code tooling (Chat Customizations Evaluations analysis and the Waza eval runner) and falls back to a bundled run-evals.ps1 harness. Covers capability vs regression eval sets, grader types (deterministic / LLM-as-judge / human), pass@k vs pass^k, and eval-driven development. Rule of thumb: start from 20-50 real failures, not synthetic prompts. USE FOR: evaluate agent, evaluate skill, evaluate prompt, eval harness, build evals, Waza, analyze-prompt, Chat Customizations Evaluations, LLM-as-judge, grader, capability eval, regression eval, pass@k, pass^k, eval-driven development, does my skill work, test a prompt, measure agent quality, run-evals. DO NOT USE FOR: authoring the skill itself (use skill-creator), MCP server eval questions (use mcp-builder Phase 4), unit-testing PowerShell code (use pester-patterns), security review of an agent (use agent-security-review).
A disciplined general debugging workflow for PowerShell/DSC code, builds, and tests: reproduce, localize, reduce to a minimal case, fix the root cause, then guard with a regression test. Stop-the-line on a red build; fix causes not symptoms; never swallow an error to make a failure disappear. A language-neutral method with PowerShell tactics (`-ErrorAction Stop`, `try/catch`, `Get-PSCallStack`). USE FOR: debugging, how to debug, systematic debugging, reproduce the bug, isolate the failure, root cause analysis, minimal repro, bisect a regression, flaky test, intermittent failure, error handling strategy, stop the line, regression guard, works on my machine, heisenbug. DO NOT USE FOR: DSC resource failures on a node (use dsc-troubleshooting), WinRM connectivity (use winrm-troubleshooting), Sampler build or Pester failure diagnosis (use sampler-build-debug), whether a long job is stuck (use long-running-job-monitor), security review (use agent-security-review).
Authors, audits, and improves `Skills/**/SKILL.md` files following Anthropic's canonical Agent Skills guidance: progressive disclosure (body ≤ 500 lines, references one level deep), the six-step authoring frame (Name / Trigger / Outcome / Dependencies / Step-by-step / Edge cases), trigger-keyword descriptions, the 1024-char description cap, third-person rule, degrees-of-freedom calibration, evaluation-driven development with the Claude-A / Claude-B loop, and cross-skill overlap audits. USE FOR: create a new skill, write SKILL.md, improve a skill, audit a skill, why is my skill not triggering, skill description optimisation, skill triggering, skill keywords, progressive disclosure, skill body too long, description over 1024 chars, split skill into references, eval a skill, skill auto-selection, skill overlap audit, point dont dump, six step skill framework. DO NOT USE FOR: writing `Instructions/*.instructions.md`, `Agents/*.agent.md`, or prompts; configuring MCP servers.