tidings

Author a deterministic email parser for a bank Tidings doesn't yet support, working only from the user's quarantined ("Needs review") emails — scrubs real samples into fixtures, derives a conservative regex parser, and runs the suites. Invoke when the user says "add support for my bank", "my bank isn't supported", "write a parser for <bank>", or "bank not recognized".

Audit whether the codebase obeys the directives declared in its CLAUDE.md, AGENTS.md, or nested context files — finds violations with file:line evidence and classifies each as auto-fixable or human-judgment. Read-only by default (--apply does doc-only fixes). Complements /basis-principles-audit (which scores whether the rules are good). Invoke for "conformance", "AGENTS.md compliance", "do we follow our own rules", or "Basis cleanup".

Audit a codebase against the five Basis principles for agent-native architecture — canonicality, localization, verifiability, interoperability, default-no. Produces a 5-row scorecard with file:line evidence plus tiered recommendations. Read-only. Invoke for a principles read on whether a repo is structured for agents, or when the user mentions the Basis essay, "agent-native codebase", "agent-ready", or agent readiness.

Apply Tidings brand voice to any user-facing copy — marketing strings, error messages, empty states, tooltips, toasts, FAQ entries, button labels, README prose. Loads the canonical voice rules from docs/brand/voice.md and self-checks the result against the 5-bullet PR review checklist. Invoke when writing or editing any string a user will see, or when reviewing a copy-heavy diff.

Frame the session with Fable as orchestrator and Opus subagents as the workforce — Fable owns decomposition, decisions, briefs, verification, and commits; Opus agents execute self-contained work packets (coding, research, bulk reading). Generic across planning, research, and general coding missions. Invoke at the start of a session with /fable-orchestrate, optionally followed by the mission.

Author an execution-grade implementation plan for a feature or fix, saved as a dated spec under docs/specs/, detailed enough that Opus agents can execute it end-to-end without Fable in the loop. Invoke when the user starts with /fable-plan, asks for a "fable plan", or wants a task planned for later Opus execution rather than built now.

Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.

Generate raster image assets with the Azure OpenAI gpt-image-2 deployment, and cut them out to transparent PNGs. Use when you need hero art, glow/ember/smoke/haze layers, textures, illustrations, or any generated image — especially anything needing transparency, since gpt-image-2 can't emit alpha and there is a two-step matte→cutout workflow.

Set up the Tidings backend for agent / BYO-UI consumption of /api/v1/: detect Docker stack vs dev checkout, set the headless knobs, mint a bearer token the running process honors, verify the auth contract, print consumer snippets. Invoke on mentions of headless, bring-your-own-UI, agent access, BYO frontend, or calling the API from outside this repo.

Survey any codebase as a senior advisor and produce prioritized, self-contained implementation plans for OTHER models/agents to execute. Strictly read-only on source code — never implements, fixes, or refactors anything itself. Use when asked to audit a codebase, find improvement opportunities (bugs, security, performance, tech debt, DX), suggest features or roadmap direction, or generate handoff plans for another agent.

Review or write selling copy — landing pages, headlines, feature blurbs, FAQs, README pitches — with David Ogilvy's discipline — facts over adjectives, one promise per surface, every claim verified against the product. Invoke for "Ogilvy analysis", "review the marketing copy", "does this headline sell", "make this page convert", or any persuasion-copy audit. Pairs with brand-voice, which owns the register.

Measurement-first performance audit for a web app (React/TS frontend and/or its API backend). Reproduces the symptom, MEASURES with real tools (Chrome DevTools traces, Lighthouse, static perf-grep, backend timing), diagnoses against a canon-backed checklist, and saves a source-labeled scorecard. Read-only on app code. Invoke for "audit performance", "why is this slow", "fix the lag/jank", "reduce INP", or "profile the API".

Produce an honest, evidence-based audit and prioritized improvement plan for a repository, saved as a dated spec under docs/specs/. Read-only analysis — no code changes. Use when the user wants a principal-level review of a codebase across architecture, code quality, security, testing, performance, dependencies, DevEx, and docs, plus a milestone task plan. Tuned for Claude Fable 5.

Scaffold a React + Vite + pnpm frontend inside an existing project

Manages shadcn components and projects — adding, searching, fixing, debugging, styling, and composing UI. Provides project context, component docs, and usage examples. Applies when working with shadcn/ui, component registries, presets, --preset codes, or any project with a components.json file. Also triggers for "shadcn init", "create an app with --preset", or "switch to --preset".

Tune a prompt, system prompt, skill, or autonomous goal-loop (/loop, /goal) instruction set that will be RUN BY Claude Fable 5 or Mythos 5, applying Anthropic's Fable 5 prompting patterns. Can also scaffold a Fable-tuned prompt from a description. Presents recommended tuning options via AskUserQuestion and applies your choices. Not for prompts targeting other models.

Audit the Tidings UI for "AI tells" — generic, default-shaped patterns (Inter-by-default, un-customized shadcn, Lucide-everywhere, round fake data, "Get Started" CTAs) that make an interface read as AI-generated. Scores tell-families with file:line and screenshot evidence, returns a P0/P1/P2 punch list. Read-only; never edits app code. Invoke to de-slop the UI or find AI tells.