stevekinney

Tensorlake SDK — sandboxes for AI agents and applications. Use when the user mentions tensorlake or sandboxes, or asks about Tensorlake APIs/docs/capabilities. Also use when building an application, coding agent, or agentic system that needs a sandbox to run code — e.g., executing LLM-generated or untrusted code, persistence via suspend/resume, snapshots/checkpoints for forking parallel workers, custom images, exposing ports, egress allowlists, PTY/interactive shells, computer-use / desktop automation, browser automation (Chrome CDP, Playwright), local tunnels for non-HTTP protocols, async parallel sandboxes, Harbor evals or RL rollouts, file transfer, SSH access, remote-dev (VS Code Remote-SSH), or OCI base images. Also covers Tensorlake's sandbox-native durable workflow orchestration. Works alongside any LLM provider (OpenAI, Anthropic), agent framework (Claude/OpenAI agents SDK, LangChain), database, or API. When this skill applies, ALWAYS WebFetch https://docs.tensorlake.ai/llms.txt first.

Ship Tribunal to production and triage deployment, health-check, Fly, Neon, GitHub OAuth/App, Tensorlake, proxy, review-engine, or live-review rollout issues. Use when a user asks to deploy Tribunal, prepare production secrets, verify production readiness, enable live reviews, debug broken deploys, inspect production health, or recover/rollback Tribunal services.

Ship Tribunal to production and triage deployment, health-check, Fly, Neon, GitHub OAuth/App, Tensorlake, proxy, review-engine, or live-review rollout issues. Use when a user asks to deploy Tribunal, prepare production secrets, verify production readiness, enable live reviews, debug broken deploys, inspect production health, or recover/rollback Tribunal services.

Address unresolved GitHub pull request review comments by analyzing feedback, implementing fixes, and resolving threads.

Generate or refresh a multi-agent project context document and execution-plan authoring prompt from repository analysis. Use when bootstrapping multi-agent orchestration, updating architecture/ownership/contracts guidance, or preparing a reusable planning prompt. Support explicit output path (`OUTPUT=<path>` or plain-language path request); otherwise detect a sensible default path from the project structure.

Stage intended changes, validate them, and create a focused commit with a clear, ticket-aware message.

Apply frontend domain conventions for components, Svelte runes/attachments, route form UX, and testing patterns.

Create or update a GitHub pull request with ticket context, synchronized branch history, and a concrete test plan.

Use this skill when refreshing Weft public documentation, AGENTS.md, CLAUDE.md, or mirrored skills from recent pull requests.

Use this skill when restoring or protecting Weft coverage with LCOV-backed branch targeting, structural test doubles, coverage allowlists, conformance fixtures, or tests for hard-to-reach protocol, schema, lifecycle, and CLI paths.

Use this skill when a Weft change touches ack-gated flows, pending promises, WebSocket workers, shutdown, disposal, cancellation, retries, reconnects, heartbeats, timers, or any lifecycle path where ordering can leave work hanging.

Use this skill when writing or reviewing Weft comments, JSDoc, documentation, README examples, test prose, or operation descriptions that describe wire responses, server diagnostics, masked errors, public APIs, or runtime recovery.

Use this skill when a Weft change renames, reshapes, normalizes, hashes, serializes, or replays persisted data such as effect-log records, checkpoints, tool calls, tool results, JSONValue content, storage payloads, or compatibility fixtures.

Run the full verification cycle before declaring work done: build, typecheck, lint, test, export review, diff review. Use when the user says "verify", "validate", "check everything passes", "run the full suite", "make sure it's ready", or before committing, opening a PR, or finishing a feature. Also use proactively after completing any non-trivial implementation work.

Security review checklist for weft covering server routes, MCP authentication, workflow trust boundaries, storage, activity isolation, and credential handling. Use when the user says "security review", "check for vulnerabilities", "audit security", "review trust boundaries", or when changes touch server routes, authentication, storage, or any code that handles untrusted input. Also use proactively when implementing features that cross trust boundaries.

Security review checklist for weft covering server routes, MCP authentication, workflow trust boundaries, storage, activity isolation, and credential handling. Use when the user says "security review", "check for vulnerabilities", "audit security", "review trust boundaries", or when changes touch server routes, authentication, storage, or any code that handles untrusted input. Also use proactively when implementing features that cross trust boundaries.

Load a pull request, implement review feedback (human and bots), resolve threads, and loop until CI is green and the PR is merge-ready. Trigger on a PR number, PR questions, or fix-it requests.

Review recently completed tasks, sessions, and PRs for uncaptured work, run the validation scripts, and file tasks for pre-existing issues. Trigger on "audit the tasks" or "audit the backlog".

Remove git worktrees for completed tasks whose pull requests have merged, skipping anything still in flight or with uncommitted changes. Trigger on "clean up worktrees" or "tasks prompt cleanup".

Gate PR creation behind a multi-agent review loop: discover subagents, parallel-review the diff, implement feedback, loop until consensus, then open the PR. Trigger on "open a PR", "create a pull request", or "submit for review".

Clear the merge queue: find every open pull request that is genuinely ready, fan out to merge each, and complete the associated task. Conservative — never force-merges. Trigger on "merge the ready PRs".

Work the next available task end-to-end from inside its worktree: plan it (gated by plan-review) if unplanned, implement it, and open a pull request via committee-review, leaving the task in-review. Trigger on "work the next task" or "tasks prompt next".

Adversarially review a task's drafted plan with Codex before it is associated and implemented. Loops until Codex approves or the round cap is hit; fail-warns when Codex is unavailable.

Plan every unplanned incomplete task: fan out a subagent per task to draft a plan, gate each through plan-review, and associate approved plans back to the task. Trigger on "plan the tasks" or "plan the backlog".

Adapts K-12 lesson plans to classroom learning needs using the Universal Design for Learning (UDL) framework. Use when a teacher asks to differentiate a lesson, adapt activities for diverse learners, set up a classroom profile, update learning needs, or says "help me differentiate", "adapt my lesson", "modify this lesson plan", "accommodation suggestions", "set up my classroom", or "I have students with different needs". Requires the Lesson Plan Adapter MCP connector.

Sweep the current branch's pull request: commit uncommitted work, rebase onto origin/main, check CI, triage and fix failures, review all PR comments (both review comments and conversation comments), build a task list, address every item, push, resolve review threads, and reply to conversation comments. Use this skill whenever the user says "address the PR", "sweep the PR", "handle PR feedback", "fix PR comments", "triage the PR", or any variation of wanting to get a pull request into mergeable shape. Also trigger when the user asks to "check on the PR" or "get the PR ready".

Audit a codebase for dead code, unused exports, stale dependencies, legacy shims, backwards-compatibility layers, deprecated markers, commented-out code, and anything else that no longer serves a purpose — then remove it. Use this skill whenever the user asks to "clean up dead code", "audit unused code", "remove legacy code", "strip deprecated code", "find unused exports", "prune dependencies", "remove backwards compatibility", or any variation of wanting to eliminate code that is no longer needed. Also trigger when the user says "audit the codebase", "clean house", or "remove cruft" — especially in pre-MVP or early-stage projects where carrying dead weight is never justified.

Health-check diagnostic that verifies CLI availability, environment variables, database connectivity, GitHub authentication and secrets, and Railway linkage. Reports findings with contextual recommendations. Use this skill when the user says "doctor", "health check", "diagnose", "check status", "is everything working", or any variation of wanting to verify the project's configuration.

Batch-replace the @lesson-adapter/ package scope with a new scope across the entire codebase. Use this skill when the user says "rename", "change scope", "rebrand", "change the package name", "replace @lesson-adapter", or any variation of wanting to change the npm scope used throughout the monorepo.

Interactive onboarding wizard that creates a Neon project, configures environment variables, optionally sets up Railway and GitHub secrets, runs the initial migration, and generates SvelteKit types. Use this skill when the user says "set up", "onboard", "initialize the project", "configure the environment", "run setup", or any variation of wanting to get the project ready for development.

Interactive teardown that walks through removing configured services: GitHub secrets, Railway linkage, environment file, and Neon project. Each phase requires explicit user confirmation. Use this skill when the user says "teardown", "tear down", "remove services", "clean up infrastructure", "delete the project", or any variation of wanting to undo the setup.

Add a dependency to the correct workspace in this Turborepo monorepo, handling the workspace:* protocol for internal packages. Triggers when the user says "add dependency", "install package", "add a package", "bun add", or "/add-dependency". Also triggers when the user asks to add, install, or use a new library or package.

Scaffold a new app or package workspace in this Turborepo monorepo. Triggers when the user says "add a new app", "create a workspace", "new package", "scaffold workspace", "add workspace", "create an app", "create a package", or "/add-workspace". Also triggers when the user asks to add or create something that should live in apps/ or packages/.

Run a comprehensive health check on this Turborepo monorepo. Validates workspace configurations, dependency consistency, TypeScript resolution, linting, and the full turbo pipeline. Triggers when the user says "check health", "validate setup", "validate monorepo", "is the monorepo healthy", "audit the monorepo", or "/validate-monorepo". Also triggers when something seems broken across workspaces or after significant structural changes.

Guidance for publishing npm packages with modern npm auth (trusted publishing/OIDC, granular tokens, 2FA) and release automation. Use when configuring GitHub Actions or other CI publish workflows, updating npm publish docs, or troubleshooting npm token/auth changes.

Guidelines for Silvan CLI and UI UX consistency, output conventions, JSON and verbosity behavior, error recovery, onboarding, and operator workflows. Use when implementing or reviewing CLI output, help or onboarding flows, error handling, JSON schemas, run lifecycle UX, or dashboard features in this repo.

Automate browser interactions, test web pages and work with Playwright tests.

Tasks CLI — inspect or update the local task graph for this project. Use when working with the `tasks` CLI, choosing new work, decomposing documents into tasks, or managing the local task database under tmp/tasks.db.

Use this skill when the user asks to create, debug, or operate Temporal workflows, activities, or workers. Triggers include: "create a Temporal workflow", "write a Temporal activity", "debug stuck workflow", "fix non-determinism error", "workflow replay", "activity timeout", "signal workflow", "query workflow", "worker not starting", "activity keeps retrying", "Temporal heartbeat", "continue-as-new", "child workflow", "saga pattern", "workflow versioning", "inspect workflow history", "check task queue", "describe namespace", "list schedules", "worker deployment", "Temporal Cloud", "temporal CLI", or mentions Temporal SDK development or Temporal cluster operations.

Ship Sandman to production and triage deployment, health-check, Fly.io, Neon, E2B template, demo-token access, sandbox lifecycle, rate-limit, reaper, or live smoke-test issues. Use when a user asks to deploy Sandman, prepare production secrets, verify production readiness, debug a broken Sandman deploy, inspect production health, recover a failed sandbox rollout, or validate cost/concurrency controls.