contember

Phase 1 of opice authoring: turn a rough testing brief (and/or supplied scenarios, a spec, or just the codebase) into reviewable *skeleton* `*.test.ts` files — metadata-first `browserTest`, pending `step` stubs carrying `intent` + `hint`, and `invariant.todo` acceptances. Grounds the skeletons in real flows using whatever's available — supplied docs, code exploration, and/or a light browse of the running app. Does NOT author executable steps (that's opice-author, phase 2) — it produces the skeleton a human reviews first. Trigger when the user says "/opice-plan", "plan opice scenarios for …", "what should we E2E test here", hands you a feature/app/spec, or gives you rough scenarios to turn into tests.

Phase 2 of opice authoring: fill in a skeleton `*.test.ts` (written by opice-plan) by walking the running app via opice-browser (a stateful Playwright browser), turning each pending `step(name, { intent, hint })` stub into an executable step with real selectors, promoting `invariant.todo` to enforced invariants, and verifying it passes by running `bun test`. Trigger when the user says "/opice-author <file>", "write an opice test for this skeleton", "author this opice scenario", or hands you a skeleton `*.test.ts` (or a `*.scenario.md`) and asks for a test.

Diagnose and fix a failed opice CI run. Given a failed-run URL (or run id), pulls the failure details from the platform, reproduces each failing scenario locally in a real browser, and either fixes the test (selector drift, timing) or flags a genuine app regression — without ever gutting an assertion just to make it pass. Trigger when the user says "/opice-reeval <url>", "look at this failed run <url>", "this opice run failed, fix it", "re-eval the failing CI", or pastes a dashboard run link and asks what broke.

Author (or re-author) many opice tests at once by fanning out one `opice-author` agent per skeleton `*.test.ts`. Use when there are more scenarios than fit comfortably in a single context — e.g. "author all skeletons in tests/browser", "fill in every pending opice test". Caps concurrency, isolates each agent's browser session, and collects a pass/fail table. Trigger on "/opice-batch", "author all the opice scenarios", "fill in these skeletons", or any request to author more than a couple at once.

Drive a running Okena terminal multiplexer from the `okena` CLI — run commands in and read output from the user's terminals, and inspect or manage projects, worktrees, services and windows. Use whenever you need to act inside the user's Okena terminals instead of your own shell.

Development skill for Rust GUI applications using GPUI framework (Zed's UI framework). Use when working on GPUI-based desktop applications - creating views, managing state, handling actions, or reviewing/refactoring existing code. Triggers on Rust GUI development, GPUI views, Entity/Context usage, terminal emulators, code editors, or any desktop app built with GPUI.

How to run and verify the dotaz web server locally — build, launch, drive the auth/RPC surface.

Create a new release by tagging and pushing. Use when the user wants to release, publish, or tag a new version.

Author and consume roj presets — the @roj-ai/* agent SDK, transport, clients. Use when writing or reviewing a preset (createPreset / defineAgent / createOrchestrator), authoring a custom plugin (definePlugin), wiring filesystem / shell / resources / uploads / services / sessionState, building system prompts, debugging path or sandbox issues, shipping a bundle (`roj build` / `upload` / `resource` / `deploy`), or integrating a host app via @roj-ai/client / @roj-ai/client-react / @roj-ai/standalone-server / @roj-ai/debug. Triggers also include working under `~/projects/contember/roj/packages/*` or in repos consuming `@roj-ai/*` (roj-platform, webmaster's `packages/agent`, scorm-preset, similar).

Spawn E2B sandbox for debugging and testing tools/commands in the container environment.

Interactive visual canvas for structured communication between agent and user. Opens a rich annotatable document in the user's browser where the user reviews, comments, answers questions, and submits feedback. Supports planning, architecture reviews, code reviews, discovery interviews, implementation summaries, proposals, decision documents, and explanations.

Query a compiler-accurate code graph of a TypeScript repo — who calls, imports, extends or type-references a symbol, across every package, in milliseconds. Use BEFORE renaming, deleting, changing the signature of, or refactoring any exported symbol; when asked "what uses X", "is X still used", "what breaks if I change X", "what does this package depend on"; when tracing blast radius or mapping an unfamiliar codebase; and instead of grep whenever the question is about code structure rather than text. Also finds import cycles and dead exports. Triggers on "pavouk", "who calls", "find all references", "what would break", "is this dead code", "impact of changing".

Author, run and debug semeno database seeds. Use when the user asks to create a seed/demo scenario, populate the database with test data, reset a scenario, or when a task needs a specific DB state (E2E fixtures, manual QA states). Covers the seed contract (idempotency, reset, dependsOn, scopes) and the semeno CLI.

Kurikulum e-learning course framework. Headless Preact compound components (core) + styled Tailwind template. Multi-locale content via virtual modules, SCORM 1.2 / 2004 / cmi5 / xAPI delivery, opinionated <KurikulumApp> root.

Cut a new pletivo release — bump version, push tag, watch the npm publish workflow, and write GitHub Release notes in the project's house style. Use when the user says "release", "ship", "tag a release", or "cut v0.1.X". Only operates on the `main` branch.

Release and publish packages to npm. Use when the user says "release", "publish", "tag version", "cut a release", "npm publish", or wants to create a new version of the packages. Handles version selection, tagging, pushing, and monitoring the CI publish pipeline.