なるみ

Use only when the user explicitly names writing-work-logs to produce a concise, evidence-based Git work log for a resolved date range; never auto-activate for generic status or summary requests.

Create, revise, review, or track concise, outcome-oriented product and system roadmaps with phased milestone checkboxes and evidence-based progress, and delete them when complete; use write-plans for executable task planning and implementation tracking.

Draft, execute, or track lean, verifiable implementation plans for non-trivial migrations, refactors, PR splits, task checklists, and other work with sequencing, tradeoffs, unknowns, or risks, and delete them when complete; use write-roadmap for strategic direction and skip small obvious tasks.

Inspect local Git changes, draft or validate Conventional Commit messages, choose types/scopes, explain breaking changes, or create a focused commit from the actual selected diff.

Create, review, audit, migrate, or update scoped AGENTS.md files with clear rules, verified commands, action limits, instruction order, and checks that prove the work is done. Use also when verified, durable repository guidance emerges during authorized work and should be captured at the narrowest applicable scope.

Deprecated internal reference for designing, wiring, or testing Peewee ORM models with deferred database binding, scoped connections, explicit transactions, and isolated SQLite fixtures.

Deprecated internal reference for using `codebase-memory-mcp` in CLI-only mode for graph-first codebase indexing, symbol search, call and data-flow tracing, source snippets, architecture, Cypher, and change impact.

Inspect Jira or prepare and perform precisely authorized Jira mutations with ankitpokhrel/jira-cli (`jira`), including issues, comments, worklogs, epics, sprints, releases, projects, boards, and script-friendly output.

Build AI-powered software with TypeSafe: small units of AI intelligence you can use like programming primitives. Its System One models, including Jev, turn natural language and application state into typed judgments and probabilities that code can combine. Use when a feature needs programmable common sense, when brainstorming what AI could make possible in an app, or when an LLM prompt-and-parse step could become a structured decision. Applications include routing, ranking, extraction, verification, and interactive experiences; these are starting points, not the limits. Read the bundled docs and cookbooks to find useful patterns and discover new combinations.

Deprecated internal workflow for switching a clean Git worktree to local `main` and fast-forwarding it from its configured upstream without losing branches, commits, or local files. Use only when the user explicitly invokes $syncing-main-branch or names syncing-main-branch.

Deprecated compatibility reference for numerical agent-skill ratings and rubric-based comparisons. Use the active create-agent-skills workflow for current skill scoring, review, revision, and creation work.

Deprecated internal reference for iterative multi-reviewer code review, verified findings, explicitly authorized fixes, and bounded re-review cycles.

Review diffs, pull requests, commits, patches, or source files for correctness, baseline security, performance, maintainability, tests, and integration risk. Use for ordinary read-only review, edge-case audits, PR preflight, and reviewer simulation; use audit-code-security when security is the primary objective, and harden-code-paths to fix confirmed failure modes.

Deprecated internal reference for inspecting pull-request feedback, verifying claims, making bounded local fixes, and preparing focused commits and approved public actions.

Deprecated internal workflow for auditable city-based food research with multi-source evidence, scope-checked rankings, standardized scoring, and synchronized candidate, recommendation, and exclusion files.

Create, revise, or review prompts and agent instructions using the user-specified model or otherwise the detected runtime model to select a matching reference, with general principles only when the model is unknown or not covered.

Deprecated compatibility reference for choosing, reviewing, renaming, or standardizing agent skill names. Use the active create-agent-skills workflow for current naming and authorized repository rename work.

Deprecated internal reference for creating, inspecting, attaching, repairing, removing, or pruning local Git worktrees while preserving branches, commits, files, and submodule data.

Manage Python projects and standalone scripts with uv, including setup, dependencies, execution, quality gates, locking, builds, and release preparation. Use for `uv run`, `uv add`, inline script metadata, packaging, or an explicitly requested publish workflow.

Deprecated internal reference for reading, maintaining, and curating concise repository-root MEMORY.md notes for verified gotchas, durable user preferences, and established repository conventions.

Deprecated internal reference for iteratively auditing and improving an existing web interface through an explicit Chrome DevTools audit-fix-commit loop.

Assess or improve an existing codebase's architecture when the user asks about module boundaries, coupling, scattered ownership, testability, change locality, deep modules, seams, or behavior-preserving structural refactoring. Use for cross-module design rather than ordinary diff review or a confirmed edge-case bug fix.

Use when asked to proactively harden a specific code path by confirming and fixing plausible edge-case, boundary, failure-mode, or lifecycle bugs; to fix a confirmed security finding; or to check nearby same-pattern regression risks. Use audit-code-security first for broad threat analysis or unverified security alerts.

Run one-shot Google Gemini research through the official Google GenAI Interactions API with Google Search, Google Maps, or URL Context grounding and inspect citation evidence. Use when Codex needs current web grounding, place or local recommendations, answers about specific public HTTP(S) URLs, or an explicit Gemini/Google GenAI grounded response.

Interview the user one decision at a time to stress-test a plan or design until goals, constraints, assumptions, tradeoffs, risks, and verification are explicit. Use only when the user asks to be grilled or names “grilling designs.”

Use when the user asks for a detailed, beginner-friendly, or step-by-step explanation; wants to understand how or why something works; asks to unpack a concept, issue, PR, code change, system, error, document, or technical decision; or says to explain it like teaching a child.

Deprecated compatibility reference for designing, reviewing, or implementing accessible user interfaces. Use the active design-user-experiences workflow for current interface and end-to-end experience work.

Design, review, or implement accessible user interfaces and end-to-end digital experiences, including screens, components, navigation, information architecture, workflows, system states, and interaction behavior. Use bounded surface mode for focused UI work; substantial cross-surface or product-level experience changes require an approved proposal before implementation.

Design, adapt, generate, or validate slide color systems and semantic palettes, including brand-color integration, usage roles, contrast evidence, and presentation-environment caveats.

Prepare and publish one explicitly authorized public Telegra.ph article through the official Telegraph API, including safe node conversion, optional account setup, credential handling, and page verification.

Create, revise, embed, or troubleshoot accessible SVG diagrams and illustrations for slide decks, documents, and other static artifacts, with portable sizing, layout, style, and validation guidance.

Create or revise complete Marp/Marpit slide decks, coordinating narrative, slide authoring, color systems, and Mermaid or SVG visuals with rendered validation.

Create, revise, convert, or validate Mermaid diagrams, including flowcharts, sequence, ER, class, state, Gantt, and optional SVG rendering for documents or slides.

Create, name, rename, revise, or review agent skills for clear triggers, safe workflows, strong checks, lean content, and easy discovery, and score or numerically compare skills only when the user explicitly asks.

Deprecated internal reference for choosing, configuring, or reviewing Python logging for libraries, services, apps, and CLIs, including stdlib logging, Loguru, handlers, levels, context, and exception behavior.

Create, inspect, edit, validate, troubleshoot, or apply Ghostty configuration, including files, precedence, includes, themes, fonts, keybindings, reload behavior, runtime diagnostics, launch behavior, and platform or version constraints.

Audit Atuin history for duplicate pressure and high-confidence typo/retry pairs, then prepare fixed-scope deduplication or user-run inspector steps without direct SQLite deletion. This deprecated internal workflow keeps cleanup-typos disabled.

Decide whether to inspect command help before execution when the exact subcommand, flags, targets, side effects, or help mechanism are uncertain.

Use only when the user explicitly invokes `$calibrate-writing-style` to build or refine reusable English and Taiwan Traditional Chinese writing-style prompts through a multiple-choice preference interview.

Deprecated internal reference for building, extending, packaging, migrating, or testing Python CLIs with Typer, including command grammar, typed parameters, callbacks, completion, entry points, and CliRunner behavior.

Design, implement, review, or debug the deprecated repository reference for Codex CLI hooks, including hooks.json layers, event matchers, stdin/stdout contracts, and hook scripts. Verify current Codex documentation before relying on version-sensitive behavior.

Write, revise, structure, or validate Marp/Marpit Markdown slides, including directives, layouts, themes, templates, and rendered-deck checks. Use for slide authoring when palette or illustration design is not the primary task.

Perform read-only security audits, vulnerability assessments, or threat-focused reviews of diffs, pull requests, code paths, or explicitly scoped repositories when security is the primary objective or acceptance criterion. Use review-code for ordinary review with baseline security coverage and harden-code-paths for fixing confirmed findings.

Use when implementing a non-trivial, observable behavior change with TDD under explicit ecosystem production-path and test-isolation boundaries. Do not apply red-first mechanically to static configuration, wiring, metadata, or production code outside the defined boundary.

Decide whether IMRaD fits, review an IMRaD draft, transform mixed material into IMRaD, or draft a new evidence-traceable IMRaD output without fabricating results.