Skills

A collection of agent skills for Claude Code, Codex, Gemini, Cursor, and more.

Available Skills

iOS

#### app-intents

Reviews and writes Swift App Intents code for Siri, Shortcuts, Spotlight, widgets, Control Center, and Apple Intelligence.

• Covers AppIntent, AppEntity, IndexedEntity, TransientAppEntity, FileEntity, and all query types (EntityQuery, EntityStringQuery, EnumerableEntityQuery, EntityPropertyQuery, IntentValueQuery)
• Handles the full snippet-intent lifecycle: ShowsSnippetView, ShowsSnippetIntent, SnippetIntent, Button(intent:), multi-step interactive confirmation
• Apple Intelligence integration via @AppIntent(schema:), @AppEntity(schema:), @AppEnum(schema:) across .photos.*, .journal.*, .mail.*, .browser.*, .visualIntelligence.*
• Widget and Control Center wiring: WidgetConfigurationIntent, ControlConfigurationIntent, App Group shared storage, WidgetCenter.reloadAllTimelines()
• Catches ~35 common mistakes LLMs make, from @Model as AppEntity to missing @Property, stale shortcut parameters, and mutation inside SnippetIntent.perform()
• Supports iOS 16+ / macOS 13+ with iOS 19+ additions (supportedModes, continueInForeground, requestChoice, snippet reload())

#### swift-format-style

Reviews and writes modern Swift FormatStyle code with .formatted() APIs.

• Covers Date, Number, Measurement, and custom FormatStyle types
• Generates correct .formatted() calls with proper modifiers
• Reviews existing FormatStyle code for best practices
• Replaces legacy DateFormatter/NumberFormatter patterns
• Supports iOS 15+ / macOS 12+ APIs

#### background-execution

Reviews and writes Swift background-execution code that runs while the app is backgrounded or suspended.

• Covers BGTaskScheduler: BGAppRefreshTask, BGProcessingTask, and BGContinuedProcessingTask (iOS 26), with registration timing, Info.plist identifiers, expiration, and setTaskCompleted
• beginBackgroundTask / endBackgroundTask assertions and ProcessInfo.performExpiringActivity for finishing in-flight work
• Background URLSession downloads and uploads, including the full handleEventsForBackgroundURLSession relaunch flow
• Silent and VoIP push, the UIBackgroundModes matrix, background audio, and the SwiftUI .backgroundTask modifier
• macOS schedulers (NSBackgroundActivityScheduler, beginActivity) and the system constraints that decide whether background work runs at all

#### widgets

Reviews and writes SwiftUI WidgetKit code across every widget surface.

• Covers all widget kinds: Home Screen, Lock Screen / accessory, and watch widgets, plus Controls (Control Center, Lock Screen, Action button), Live Activities, and the Dynamic Island
• Timeline model: TimelineProvider / AppIntentTimelineProvider, reload policies and the daily reload budget, WidgetCenter reloads, and push-updated widgets (WidgetPushHandler)
• Renders correctly in every appearance: widgetRenderingMode, widgetAccentable(), widgetAccentedRenderingMode, and the luminance-to-alpha trick for the tinted/accented Home & Lock Screen
• Interactivity and Controls: Button(intent:) / Toggle(intent:), ControlWidgetToggle (SetValueIntent) vs ControlWidgetButton, the app-process boundary, and the App Group store
• Live Activities: ActivityAttributes / ContentState, ActivityConfiguration, DynamicIsland, push (token / channel / push-to-start), and supplementalActivityFamilies for Apple Watch and CarPlay
• Spans WWDC 2020-2026 with the SiriKit .intentdefinition to App Intents and ClockKit to WidgetKit migration map, ~40 anti-patterns, and a worked end-to-end example; supports iOS 14+ / iPadOS / macOS / watchOS / visionOS

#### observability

Reviews and writes Swift observability code - logging, metrics, diagnostics, and production telemetry on Apple platforms.

• Covers unified logging (Logger, levels and persistence rules, privacy redaction), Console.app and the log CLI with scripted simulator/device capture workflows, and OSLogStore support-log export
• Signpost instrumentation (OSSignposter, Instruments, xctrace, mxSignpost field histograms) plus activity tracing and correlation IDs
• MetricKit end to end: MXMetricManager payloads and diagnostics through the iOS 27 Swift-native MetricManager, launch histograms, hang and exit-reason tracking
• Crash, hang, and OOM observability: watchdog codes, jetsam's missing crash reports, the hang tooling matrix, and symbolication (dSYM/UUID contract, atos, MetricKit call stacks)
• Request-level network telemetry (URLSessionTaskMetrics, error taxonomy, non-fatal report design) and production pipelines (App Store Connect Power and Performance API, OpenTelemetry)
• Third-party SDK integration rules (one crash handler, dSYM upload, breadcrumb bridging) and ~33 anti-pattern catches; supports iOS 14+ / macOS

#### essential-developer

iOS engineering the Essential Developer way, across five modes: design, build, test, review, and refactor.

• Classicist (Chicago-school) TDD: no mocking, behavior tested through the public interface, triangulation, watch-it-fail-first
• Modular design and boundaries: horizontal and vertical slicing, the modular monolith, per-module DTOs, and no @testable across module lines
• Dependency injection and dependency rejection, with a single composition root owning adapters, navigation, object lifetimes, and threading
• Test technique that scales: makeSUT, trackForMemoryLeaks, spies over mocks, reusable protocol contract specs, and the full pyramid up to snapshot and end-to-end
• Decorator / Composite / Adapter for cross-cutting concerns (auth and token refresh, analytics, feature flags) instead of edits to existing consumers
• Point a mode at a feature, a module, a diff, or an idea; all five share one references knowledge base with a worked end-to-end example

Workflows

#### agent-gauntlet

Runs a user story through a fixed pipeline of fresh-context subagents with deterministic quality gates - Uncle Bob's "run the gauntlet" agent workflow.

• Five stages: specifier (Gherkin acceptance scenarios + QA procedure), coder (tests + implementation to green), cleaner (complexity + coverage), hardener (mutation testing), optional qa agent
• Gates are numbers, not opinions: 100% tests green, complexity warn 6 / fail 8, coverage floor 70%, zero unjustified surviving mutants
• Fresh context per stage, handoff through files; spec files are throwaway scaffolding that lives for one story cycle
• Doubles as a Claude Code plugin: install it from this marketplace, then /gauntlet <story> drives the whole pipeline
• Swift-first gate scripts (SwiftLint, muter) with documented equivalents for JS/Python/JVM/Rust (Stryker, mutmut, PIT, cargo-mutants)

#### handoff

Ends an agent session on a step file the human controls, instead of on context compaction, and starts the next session from it.

• One numbered step file per handoff in <repo-root>/_agent/handoff/<TICKET>/: Goal, Done, Where we are (Verified vs Believed), Next, Session
• /handoff writes the step, /handoff resume <TICKET> opens only the newest one and continues from it
• _agent/ stays out of git through .git/info/exclude; works from worktrees and in plain folders without git
• Records the Claude Code session id and claude --resume line, so the transcript behind every claim is one command away
• With Orca the fresh session is spawned automatically; without it you get a command to paste

#### worknotes

Gives every working document an agent writes during a session one home, <repo-root>/_agent/docs/, never the repo's real docs. Sibling of handoff, same excluded folder.

• Investigations, root-cause write-ups, plans, drafts, review notes go to _agent/docs/; docs/, README.md and ADRs follow the repo's own rules
• Every document carries the session log path and claude --resume line, so the reasoning behind it is one command away
• Resolves the main checkout from any worktree, so a note written in a throwaway worktree survives it
• Past four documents on a subject: a README.md index naming the canonical one and listing the conclusions that turned out wrong
• Overturned documents get a "Superseded" banner; the body is never rewritten

Other

#### git-codebase-preflight

Audits a repository through git history before reading source files.

• Analyzes commit frequency, active contributors, and project timeline
• Identifies hot files and recently changed areas
• Detects large files, binary assets, and repo structure
• Provides a prioritized list of files to inspect first
• Works on any Git repository without reading source code

#### ffmpeg

Categorized FFmpeg recipes for video automation pipelines, adapted from the Rendi FFmpeg cheatsheet.

• Format conversion, resize with aspect-ratio padding, frame-accurate trimming
• Audio processing: replace, extract, mix, crossfade, normalize, downsample for speech-to-text (16 kHz mono)
• Advanced editing: playback speed without pitch shift, jump cuts, social-media cropping, drawtext, subtitle burn-in, watermarking, vstack, intro/main/outro assembly
• Asset generation: image-to-video, slideshows with xfade, Ken Burns zoompan, looping GIFs, scene-change thumbnails, tiled storyboards
• Encoding tuning: CRF/preset guidance for libx264, libx265 with hvc1 for Apple AirDrop, libvpx-vp9 for the web, +faststart, hardware acceleration (*_nvenc, *_qsv, VAAPI)
• Bakes in common FFmpeg footguns: -c copy rules, input vs output seeking, setpts=PTS-STARTPTS after trim, setsar=1:1 after pad, -pix_fmt yuv420p for QuickTime

#### rich-html

Renders content as a polished, self-contained HTML deck or document instead of plain Markdown.

• Produces two single-file output profiles: a fullscreen Deck (keyboard / click / swipe navigation, progress bar, counter) and a long-scrolling Document (sections, tables, legends, inline diagrams)
• Mermaid diagrams (flowcharts, sequence, dependency graphs) themed to match the page, with the slide-visibility render gotcha already solved
• Expressive layout out of the box: cards, big-stat numbers, callout notes, pills, flow steps, and tables
• Download buttons that link the best experiment, result JSON, or full log straight from the page
• A cohesive GitHub-dark design system; ideal for recaps of long runs, showcases, pitches, architecture write-ups, reports, and specs

#### simplified-technical-language

Writes, rewrites, and reviews technical text with the ASD-STE100 Simplified Technical English rules - the controlled language standard born in aviation maintenance.

• Three modes: write new procedures/descriptions, rewrite existing text into STE, or review text and report rule violations
• Enforces the core limits: 20-word procedural / 25-word descriptive sentences, 6-sentence paragraphs, one instruction per sentence, 3-noun clusters
• Controls vocabulary: one word one meaning, part-of-speech control, a substitutions table for the frequent offenders ("prior to commencing" becomes "before you start")
• Grammar discipline: approved verb forms only, no -ing verb forms, no "shall/should/may", mandatory active voice and imperatives in procedures
• Structures safety text: WARNING vs CAUTION vs NOTE, command-first warnings, conditions before commands
• Ends every rewrite with a rule-by-rule compliance checklist instead of trusting text that merely looks compliant

Installation

Any Agent (via skills.sh)

npx skills add n0an/skills

To install a single skill:

npx skills add n0an/skills --skill app-intents
npx skills add n0an/skills --skill swift-format-style
npx skills add n0an/skills --skill background-execution
npx skills add n0an/skills --skill widgets
npx skills add n0an/skills --skill observability
npx skills add n0an/skills --skill git-codebase-preflight
npx skills add n0an/skills --skill ffmpeg
npx skills add n0an/skills --skill rich-html
npx skills add n0an/skills --skill simplified-technical-language
npx skills add n0an/skills --skill essential-developer
npx skills add n0an/skills --skill agent-gauntlet
npx skills add n0an/skills --skill handoff
npx skills add n0an/skills --skill worknotes

The agent-gauntlet pipeline, handoff and worknotes also install straight from their own repos:

npx skills add n0an/agent-gauntlet --skill agent-gauntlet
npx skills add n0an/handoff-skill --skill handoff
npx skills add n0an/worknotes-skill --skill worknotes

Claude Code

/plugin marketplace add n0an/skills

The marketplace registers as n0an-skills, so a single plugin can be installed by id:

/plugin install agent-gauntlet@n0an-skills

Author

Anton Novoselov, https://github.com/n0an

License

The project is available under the MIT license. See the LICENSE file for more info.