budget

Use whenever you are iterating on the look or layout of a screen — tuning a CSS rule, building a new modal, redesigning a table, hunting a mobile-only regression. Walks an edit / reload / screenshot / inspect loop that uses the Read tool to view PNGs inline at every viewport (desktop, mobile, mobile-landscape, tablet). The harness drives the app through reusable flows (sign in, add sheet, add account, swipe a row) so each iteration only changes the bit that's being designed. Manual playbook — not part of the `maintenance` umbrella.

Use when the user wants to clear out the open Dependabot PRs — "do the dependabot PRs", "update the deps", "merge dependabot". Consolidates every open Dependabot bump into ONE branch + PR, resolves the peer-dependency fallout the bumps trigger (Vite / ESLint / TypeScript / React majors drag transitive tooling with them), gets the full CI chain green, then closes the superseded Dependabot PRs. Manual playbook — not part of the `maintenance` umbrella.

Use when src/data/achievements/catalog.ts is stale relative to a newly-shipped user-facing feature — or when a fresh achievement needs to be added, a trigger predicate rewritten, or the AchievementsModal chrome touched. The achievements modal (src/components/AchievementsModal.tsx) is a fullscreen four-tier (Beginner → Intermediate → Pro → Expert) guided tour where every feature is also an unlockable trophy. This skill describes how to add a new achievement, how to slot it into the right tier, how to phrase it (in English AND Swedish), and how to wire its trigger so the unlock fires when the user does the thing.

Use when about to open a PR (or just after pushing one) to decide whether the change needs a `.changes/unreleased/<unix-ts>-<slug>.md` fragment. Resolves the latest `v*` tag, walks commits and existing fragments since, classifies the current change against the rules in AGENTS.md, and either writes a new fragment, folds the substance into an existing fragment, or labels the PR `no-changelog`.

Use when docs/ may be stale relative to src/ layout, the persisted-data shape, or the Makefile target table — or when the in-app privacy policy at src/components/PrivacyPage.tsx may be stale relative to the storage/encryption/Dropbox claims it restates. Discovers commits that touched those sources of truth since the skill last ran and merges the current shape back into docs/architecture.md, docs/getting-started.md, and the privacy page.

Use when the maintainer (or an agent acting on their behalf) wants to cut a new release of the Budget app. Walks the pre-flight checklist, dispatches the workflow, verifies the deploy, and links to the rollback recipe. Manual playbook — not part of the `maintenance` umbrella.

Use when the user asks for high-value performance optimizations — order-of-magnitude wins in algorithms that matter, not micro-tuning. Surveys the hot paths (reducers, per-render passes, per-keystroke handlers, validators, synthesis pipelines, search indexes), presents candidates with complexity and trigger context for the user to gate the work, and is honest when the codebase has already been tuned and no high-value wins remain.

Use when the user wants comments cleaned up, or whenever you're about to touch a file whose comments narrate history instead of explaining current invariants. Removes or rewrites changelog-style comments ("previously", "the legacy behaviour", "see the plan", "used to re-scan", "K calls × O(R) before this fix", PR-number references, wiring narratives) while preserving comments that explain WHY current code is the way it is — especially "don't try X, it doesn't work" anti-pattern guards.

Use when the user wants to update categories or types for the budget. Covers adding, renaming, recoloring, re-glyphing, removing, or moving preset categories (PRESET_CATEGORIES) and preset entry types (PRESET_ENTRY_TYPES) in src/data/presets.ts — and the cascade of changes that has to land with them (CategoryIcon union, icon map, glyph allowlists, picker palettes, English + Swedish i18n, validator-friendly id rules, and changeset/PR conventions). Also covers adding a brand-new glyph from lucide-react so a fresh icon can back a type or category.

Use to work through the refactor backlog in docs/refactoring-roadmap.md or to extend it with newly-discovered code smells. Picks the highest-leverage pending item, re-verifies its severity against the current tree (line counts and the smell shape drift over time), and either lands the fix, skips it with a written reason, or extends the roadmap when exploration mode finds something new. Grounded in the roadmap — stops when the queue is empty rather than refactoring for its own sake.

Use when the home-screen / launcher / browser-tab icon looks wrong on a real device — too small, off-center, transparent, clipped by iOS rounding, or eaten by an Android mask. Walks an edit / regenerate / inspect loop that uses the Read tool to look at the rasterised PNGs after every change, scored against Apple Human Interface Guidelines and the W3C maskable-icon spec. Also the right skill when adding a new icon size to the generator or restyling the existing artwork. Manual playbook — not part of the `maintenance` umbrella.

Use whenever the user pastes captured log output in a message — production diagnostics from the in-app Logs tab, a console transcript, a CI snippet, anything resembling timestamped or scoped log lines. Walks the trace from last-known-good to the failure, traces each suspicious line back to its source by greping the logged string, forms and verifies a root-cause hypothesis, and — critically — ends by evaluating whether the logs themselves were sufficient and adds the missing diagnostics in the same change when they were not.

Use to bring the whole repository back into sync without first diagnosing which artifact is stale. Dispatches every update-* skill in the registry below, in order, and re-runs sync-oss-spec at the end to catch residual conformance violations.

Use when this repo may have drifted out of conformance with OSS_SPEC.md. Runs the bash mirror of the oss-spec validator (no Rust toolchain required), walks the violations, and fixes each one until the validator reports zero violations.

Use when README.md may be stale relative to package.json scripts, Makefile targets, .nvmrc, or the user-visible UI. Discovers commits that touched those sources of truth since the skill last ran and merges the current values back into README.md.