Loading...
Loading...
Found 303 Skills
Create diverse synthetic test inputs for LLM pipeline evaluation using dimension-based tuple generation. Use when bootstrapping an eval dataset, when real user data is sparse, or when stress-testing specific failure hypotheses. Do NOT use when you already have 100+ representative real traces (use stratified sampling instead), or when the task is collecting production logs.
(NS) Create and maintain Cypress E2E tests in an independent tests-e2e/ Node package at repo root (bootstrap when missing). DRY device-aware command architecture (shared/pages/device). Use when writing or refactoring Cypress specs, custom commands, scaffolding E2E on greenfield projects, or implementing E2E tasks — not when planning E2E task markdown (ns-spec-driven references/e2e-test-task-generator.md). Read harness e2e rules when present. Mandatory discovery before writing specs.
Read Cursor / Claude Code / Codex agent logs and call write_knowledge for each durable learning found. Default auto-writes then reports what was cached and expected token savings (analytics-style: rediscovery/generation cost reused on each future read), and opens the HTML report. Dry-run lists the exact write_knowledge payloads (title, content, repo, branch) without writing. Use when the user says "Please bootstrap my knowledge with Dosu", "bootstrap agent knowledge", "/bootstrap-agent-knowledge", "log to dosu knowledge", "mine my sessions into Dosu", "backfill branch notes from my agent logs", "save my agent logs to Dosu", or wants a one-shot pass over local histories.
Scaffold and build a full-stack web app: FastAPI backend (Python, uv, SQLModel, Postgres, Alembic, JWT + Google OAuth, boto3/S3) + React frontend (Vite, TypeScript, shadcn/ui + Tailwind, TanStack Router/Query/Table, Zod, Axios), wired with Docker Compose. Use this skill whenever the user wants to spin up, bootstrap, create, or design a new full-stack webapp; an API-first backend + SPA frontend; an admin/portal/dashboard app; file upload + S3; RBAC / role-based auth with seeded test users; local docker dev; or asks for a 'FastAPI + React' / 'Python + React' project. Runs mockup-first: marketing-design (brand/logo raster) + opendesign (HTML page mockups) before code, then ports the design to Tailwind/shadcn. Covers project structure, local setup, auth/RBAC, S3 uploads, and the gotchas that break these stacks.
(NS) Generate or refresh lean architecture-rules.md for AI agents — stack, layout, patterns, constraints. Use when the user asks for architecture rules, .nextstage-harness/rules/architecture-rules.md, agent conventions, or "document how this repo works for the AI". Do NOT use for business specs (ns-harness-codebase-reverse-spec), brownfield maps (ns-harness-bootstrap-brownfield), or one-off Cursor rules without a full architecture pass.
Initialize Spec-Driven Development context — detects tech stack, conventions, architecture patterns, and bootstraps persistence backends. Triggers on 'sdd-init', 'init sdd', 'setup sdd', 'initialize sdd', 'setup project', 'initialize project context'. Creates/updates docs/specs/architecture.md & ontology.md (Constitution), and populates knowledge-graph.json.
Stand up (or extend) an operator-only in-app scratchpad / internal test-bed — a single gated internal page that reuses the project's real components and library functions so you can judge work-in-progress by eye: eyeball generated visuals (OG/social cards, emails, avatars, empty states), compare variants side by side or with live sliders, check config/env readiness, and exercise real side-effecting flows with fake data — without shipping any of it to users or clicking through the whole product. Boards and config panels get a "Copy as JSON" button so a tuned-by-eye result round-trips straight back into the code or to an agent to apply. Bootstraps into the user's existing stack: detects the framework, auth, and env conventions, writes a fail-closed access gate first, then a route with a section switcher seeded from their own code. Use when a user says "spin up a scratchpad", "internal test bed / dev playground / sandbox page", "an operator/admin preview page", "a page to eyeball my work", "preview my OG cards / emails / component variants", "add a test page for these components", or "an internal page to exercise a flow with fake data". NOT for public/user-facing features, NOT a replacement for tests (pass/fail machine-checkable → write a test), NOT Storybook (this lives inside the real app), and NOT a place to do load-bearing production operations.
Harness patterns for coding agents — memory, permissions, context engineering, delegation, skills, hooks, bootstrap.
Bootstrap or merge a .planning/ setup from existing ADRs, PRDs, SPECs, and docs in a repo.
Deploys the complete SDD architecture with engram persistence and ai-context/ memory layer in the current project. Also runs an opt-in interactive wizard mode (/project-setup --interactive) that bootstraps only a project CLAUDE.md with stop-between-question confirmation. Trigger: /project-setup, /project-setup --interactive, initialize new project, setup SDD, configure claude project, bootstrap claude, init project claude, create project CLAUDE.md.
Compute and report task-correct held-out metrics for a trained medical-imaging model — segmentation (Dice plus a boundary metric such as HD95 or NSD, per structure), classification (AUROC plus AUPRC and sensitivity/specificity with bootstrap CIs at the deployment prevalence), detection (FROC or mAP with a stated IoU criterion), interactive/promptable segmentation (the interaction-count, convergence, and per-case-time axes a static Dice omits), or generative/synthesis image evaluation (similarity plus the downstream-task efficacy similarity alone cannot establish) — plus calibration and subgroup slices. Emits a per-case results table that analyze-stats turns into publication tables, and gates the metric choice against Metrics Reloaded, CLAIM 2024, and Park et al. 2024 (no pixel accuracy for segmentation, no bare accuracy under imbalance, no static Dice for an interactive method, no similarity-only claim for a generative model). Numbers come only from executed code, never hand-typed.
The canonical way to run agent-relay - self-bootstrap the local broker and autonomously spawn, monitor, and coordinate a team of worker agents without human intervention. Covers infrastructure startup, agent spawning, lifecycle monitoring, message-based reading via the relay MCP, and team coordination.