LUC4N3X

Curates task context to reduce hallucination, stale assumptions, and attention dilution. Use when starting unfamiliar work, switching major task areas, after long sessions/compaction, when agent quality drifts, or when a task would otherwise require loading…

Retrieves, classifies, validates, and optionally persists compact durable knowledge so future tasks can reuse verified experience without treating memory as current truth. Use when prior project decisions, recurring failures, user-authorized workflow…

Extracts reusable, verified engineering knowledge after difficult tasks so future work gets cheaper. Use after a non-trivial bug fix, architecture discovery, recurring failure, benchmark, or review that revealed a stable lesson. Routes useful outcomes through…

Converts verified task outcomes into compact reusable experience cards. Use after meaningful bug fixes, failed hypotheses, performance investigations, migrations, design corrections, or repeated workflows when the result contains a lesson worth reusing.…

Maintains an opt-in, repository-local project knowledge layer for stable architecture, invariants, decisions, regressions, commands, design rules, and project-scoped experience. Use when a project is revisited repeatedly and durable context would reduce…

Chooses the smallest compatible process, language, domain, platform, visual, cognitive, and installed-skill stack for a task. Use before complex work and after project discovery. Combines autonomous skill intelligence with task, stack, risk, evidence,…

Runs a bounded teacher/judge/revision loop around model-produced engineering work. Use when an independent evaluator can materially improve correctness, scope discipline, regression safety, evidence honesty, design quality, or general-purpose skill/prompt…

Coordinates Android/Kotlin/Jetpack Compose/Gradle/Media3 work. Use for Android UI, playback, services, device behavior, build/toolchain, R8, Room/data access, testing, edge-to-edge, adaptive layout, localization, performance, or runtime debugging. Prefer…

Apply a compact final prose pass to Levyra pull request descriptions, release notes, and owner-facing writing so it sounds natural without changing facts, evidence, structure, or validation state.

Use for genuinely high-volume Levyra work such as builds, tests, lint, logs, broad searches, dependency output, Git/GitHub or CodeRabbit inspection, CI diagnostics, agent setup, cross-domain exploration, or useful cross-session retrieval. Reduce token waste…

Use only for explicit owner execution cues such as VAI, PROCEDI, INTERVIENI, FAI TU, FALLO TU, or requests to open/create a PR. Keep execution fast and low-noise while preserving full Levyra engineering rigor.

Automatically use for Levyra Android Intent, deep-link, PendingIntent, exported component, receiver, service, provider, URI-grant, FileProvider, caller-verification, or onNewIntent security work. Pair it with levyra-security-review and the affected Android…

Automatically use for Levyra GitHub Actions, CI, F-Droid, Gradle/AGP/Kotlin/KSP compatibility, build performance, configuration/build cache, artifacts, release automation, workflow security, or configuration-sync work.

Implement, debug, or review Levyra Jetpack Compose screens, state projections, navigation, animation, lifecycle, accessibility, performance, RTL, localization, edge-to-edge, adaptive layout, and screenshot-validation changes.

Implement, debug, or review Levyra Room entities, DAOs, migrations, schemas, caches, stores, backups, downloads, favorites, playlists, history, and persistent queue data.

Implement, debug, or review Levyra InnerTube, YouTube extraction, stream resolution, player-config synchronization, token, retry, timeout, cache, and network fallback behavior.