adminhead

The orchestrated-session crew: 14 subagents (coders, reviewers, debuggers, architect, devops, dba, scribe, browser tester) plus the orchestration main-thread agent, with a read-only Bash gate for the read-only agents, orchestration nudge hooks, and the orchestration system-prompt file. Agents are spawned as crew:<name>.

15 agents · hooks

Skills for authoring Claude Code skills, agent instruction files, architecture decision records, and docs; mapping codebases and domain models; planning agent teams; and auditing documentation and reviewing Claude Code config for bloat and security.

9 skills

GitHub workflow skills: issues, issue breakdown, triage, pull requests and review comments, shipping commits through merge, and releases, plus the GitHub remote MCP server.

7 skills · 1 agent · 1 MCP server

Public API design skills: REST contracts, idempotency and retries, webhooks, versioning, gRPC, GraphQL, API keys, and OAuth providers.

8 skills

Code search: ripgrep, CodeGraph's local symbol graph with auto-indexing, Context7 library docs, public GitHub code via grep.app, and always-on search guidelines.

4 skills · 3 MCP servers

JavaScript and TypeScript skills: the languages, React, Vue, Vite, TanStack, and click-path audits of UI flows.

7 skills

Deployment skills: Docker, Kubernetes, Helm, GitOps, rollouts, and production-readiness reviews.

6 skills

GitHub repository administration: Actions workflows, code security settings and alerts, Dependabot config and PR sweeps, and repo setup.

5 skills

Testing skills: unit and regression tests, browser end-to-end tests, running the project's tests, building test harnesses, and auditing test gaps.

5 skills

Coding process skills: the coding workflow, planning, problem-solving, and implementing an approved plan.

4 skills

Database skills: schema design, Postgres, MySQL, and Redis.

4 skills

Git skills: focused commits, branches, history rewrites and recovery, and worktrees.

4 skills

Observability skills: instrumentation and SLOs, PromQL and Prometheus config, Grafana dashboards, and blameless postmortems.

4 skills

Review skills: second-opinion reviews from another model, review-fix loops, anti-slop cleanup, and poka-yoke error-proofing.

4 skills

Go skills: the language, CLI tools, and performance work.

3 skills

Python skills: the language, packaging, and FastAPI services.

3 skills

Infrastructure skills: Terraform (with OpenTofu and Terragrunt) and Ansible, plus a guard hook for unsafe `tofu` and `aws` commands.

2 skills · hooks

Security skills: threat modeling, privacy review, and a manual security audit scan.

3 skills

Repo tooling skills: EditorConfig files and commit-time hooks with prek or pre-commit, and repairing failing hooks.

3 skills

Web skills: front-end web development and web design, plus a headless Playwright MCP browser.

2 skills · 1 MCP server

Authentication skills: login, sessions, and token validation for your own app, plus Better Auth.

2 skills

Model Context Protocol server skills: design the tool surface, build with the official SDKs, and review security and release readiness.

2 skills

Main-session orchestrator agent: plans work, routes it to the installed agents and verifies the results. Ships the claude-orchestrate launcher (linked into ~/.local/bin at session start), which appends the orchestration prompt to Claude Code's default system prompt and can start Claude inside herdr.

1 agent · hooks

Scala skill for Scala 3 and 2.13 (follows the project's syntax): ADTs, given/using and implicits, missing type class instances such as Ordering, Futures and effects, MUnit FunSuite and ScalaTest suites, scalafmt, 2.13 to 3 migration. Also configures the Metals language server for .scala, .sc and .sbt files.

1 skill · LSP

Accessibility skills: manual screen-reader and keyboard testing of web journeys.

1 skill

Audits a design, mockup or page against WCAG 2.2 AA at design level: contrast, focus visibility, target size, keyboard and dragging alternatives, labels, errors and authentication. Use for "audit accessibility", "check a11y", "is this accessible?" or a WCAG check before handoff.

1 skill

ADR skill: create, review, supersede and index Markdown Architecture Decision Records in docs/adr/ (Nygard or MADR) with a numbering helper that avoids cross-branch collisions.

1 skill

Read-only security review of your own Claude Code configuration: CLAUDE.md, settings permissions, hooks, subagents and .mcp.json, including static classification of unpinned npx, uvx, pipx and docker MCP selectors.

1 skill

Builds and maintains a repo's CLAUDE.md, .claude/rules and AGENTS.md from verified evidence, and turns repeated agent mistakes into checks, hooks and failure notes.

1 skill

Writes and debugs Android app code outside the UI toolkit. Use for "A migration from 3 to 4 was required but not found", form input lost after the process is killed, a StateFlow that reloads on every rotation, "cannot be provided without an @Inject constructor", or adding a repository, ViewModel, DAO or worker. Covers ViewModel state, SavedStateHandle, repeatOnLifecycle, Hilt, Room migrations, DataStore, WorkManager and targetSdk behavior changes.

1 skill

Diagnoses and writes Android Gradle Plugin builds. Use for "Manifest merger failed", "Duplicate class found in modules", "Namespace not specified", minSdk conflicts with a library, resource linking failures, 16 KB page size warnings, or ClassNotFoundException or NoSuchMethodError only in the release build (keep rules). Covers the android {} block, build types, flavors, androidComponents, signing, desugaring, KSP and kapt, lint baselines and AGP, Gradle and JDK versions within a major.

1 skill

Triages Android crashes and manages devices with adb and logcat. Use for an app that closes on launch, a "Caused by" chain or native tombstone, INSTALL_FAILED_UPDATE_INCOMPATIBLE or other INSTALL_FAILED errors, "more than one device/emulator", pm grant and revoke, and deep links with am start. Pins one serial and asks before uninstall, pm clear or wiping data.

1 skill

Sets up Android Macrobenchmark and Baseline Profile work and app size reduction. Use for "add a baseline profile for startup", "cold start regressed between two CI runs", writing a startup benchmark, checking the profile is installed, or "why is the release APK so big". Covers StartupTimingMetric, CompilationMode, the androidx.baselineprofile plugin, BaselineProfileRule, reportFullyDrawn and APK Analyzer.

1 skill

Writes and fixes Android tests where the test setup already exists. Use for "No compose hierarchies found", "Method d in android.util.Log not mocked", a Compose UI test that sleeps or flakes, onNodeWithText not finding a node, Hilt test rules, Robolectric versus instrumented tests, and connectedDebugAndroidTest failures. Covers createComposeRule, waitUntil, idling, useUnmergedTree, Gradle Managed Devices and Orchestrator.

1 skill

Ansible skill: playbooks, roles and inventories with idempotency proof, check-mode limits, validate hooks, vault or sops secrets, and ansible-lint style.

1 skill

Public GraphQL API design and review: when GraphQL fits, schema conventions, Relay pagination, result-union errors, query ceilings, federation boundary, persisted queries and field deprecation.

1 skill

gRPC and protobuf contract design and review: AIP conventions, buf lint and breaking gates, proto diff compatibility, streaming, google.rpc.Status errors and gateway mapping.

1 skill

Idempotency keys and safe retries for HTTP APIs: per-account client keys, atomic claim, in-flight and replay states with recovery jobs, and client retry policy with full-jitter backoff, retry budgets and deadlines.

1 skill

API keys a platform issues to its consumers: prefixed checksummed key format, hash-only storage and lookup, scopes, dual-key rotation with last-used telemetry, service-account ownership and offboarding, keys versus OAuth.

1 skill

OAuth 2.1 authorization-server design and review: grants and PKCE, redirect matching, token lifetimes, refresh rotation with reuse detection, scopes, consent, client registration and MCP auth.

1 skill

REST/HTTP API contract design and review skill: resources and methods, status codes, RFC 9457 errors, pagination, rate-limit headers, contract-first OpenAPI with drift checks, and versioning and deprecation policy.

1 skill

Provider-side outbound webhooks skill: event catalog and payloads, Standard Webhooks signing and rotation, at-least-once delivery with backoff and dead-lettering, SSRF defenses, and the delivery log and replay surface.

1 skill

Application-side authentication and authorization: sessions, password and recovery flows, OAuth/OIDC callbacks, JWT validation, tenant and ownership checks, cookies, CSRF and rate limits.

1 skill

A planning subagent, read-only by instruction (enforced only when readonly-agent-gate is installed), that grounds an implementation plan in the code and returns ordered slices of step-by-step instructions a coder can follow, with risks and rejected alternatives, optionally stress-tested by an outside model. On request it saves the plan as Markdown under .claude/plans.

1 agent

Aspire skill: AppHost resource model and wiring, service defaults, the aspire CLI (start, wait, describe, logs), DistributedApplicationTestingBuilder tests, and aspire publish/deploy to Compose, Kubernetes and Azure.

1 skill

Builds and reviews ASP.NET Core HTTP apps: minimal APIs and controllers, DTOs, validation, ProblemDetails, OpenAPI, middleware order, auth wiring, .http files, CRUD endpoints written without scaffolders. Use for Program.cs endpoint or middleware work, "add an endpoint", OpenAPI setup, or global error handling in a .NET web project.

1 skill

SessionStart hook that tells Claude to record knowledge worth keeping in Confluence and track follow-up work as Jira issues, as well as in the repo. Installs atlassian-mcp.

hooks

Atlassian Rovo MCP server for Atlassian Cloud (remote HTTP at mcp.atlassian.com/v2/mcp, OAuth 2.1): Confluence, Jira and Compass. Sign in with /mcp.

1 MCP server

General AWS skill: resolve account, region and identity first, sign-in with SSO or aws login, IAM policy and trust traps, AccessDenied debugging and cost-query habits.

1 skill

AWS CLI v2 mechanics for scripts and write commands: pagination, --query and output, timeouts, waiters, skeletons, dry runs and client tokens. Not for read-only inspection (aws-resource-query) or profiles, SSO and IAM (aws).

1 skill

AWS incident investigation from CloudWatch alarms, metrics, Logs Insights and CloudTrail: blast radius, change correlation, timeline and per-resource health, with mutating fixes only after approval.

1 skill

AWS Knowledge MCP server (remote HTTP, no auth): AWS documentation search and reading, regional availability of services and APIs. Docs-only; use aws-mcp instead when you have credentials.

1 MCP server

The managed AWS MCP Server (us-east-1) through the pinned mcp-proxy-for-aws SigV4 proxy over stdio: AWS docs search, regional availability, and run_script calls signed with your local profile. Optional profile name; blank uses the default credential chain.

1 MCP server

Read-only AWS CLI resource inspection: what exists in an account or region and how it is configured, with a deny-list for secret and data reads, --query projections, and changes only on request.

1 skill

AWS Well-Architected and cost review of a workload from its IaC and read-only live calls: per-pillar findings rated High, Medium or Low, plus right-sizing candidates backed by CloudWatch and Cost Explorer.

1 skill

Bash skill: defensive scripts with strict mode and its holes, quoting, trap cleanup, argument parsing, ShellCheck configuration and Bats tests.

1 skill

Better Auth library setup and debugging from version-matched docs: adapters, the auth CLI, route handlers, sessions, security options, email/password, two-factor and organizations.

1 skill

Blazor skill: component rules, render modes, forms and validation, auth per render mode, shared state, data loading, prerendering, JS interop, and Blazor Server to Web App conversion.

1 skill

Agent that verifies a running web app in a real browser against acceptance criteria with playwright-cli, returning pass, fail or blocked per criterion with evidence.

1 agent

Rebuilds a returning human's context from live project state: what needs them, what changed, what new words mean, and a single next move when asked what to do.

1 skill

Hook that denies git hook and signing bypasses (--no-verify, commit -n, --no-gpg-sign, core.hooksPath overrides or git config changes, SKIP=) and asks before edits to existing lint, format and pre-commit configs or their ignore files, git hooks and .git/config, and Bash writes to those configs. Needs bun.

hooks

chezmoi dotfiles skill: source-state attributes, templates and machine data, age and password-manager secrets, externals and run_ scripts, behind a preview-first apply gate.

1 skill

Audits UI event handlers for behavior bugs. It maps each state action's side effects, then traces every handler call in order to find sequential undo, async races, stale closures, dead paths and effects that reset what a click just set. Use when a button does nothing or leaves the wrong state though each function works alone, or after a shared-state refactor.

1 skill

PreToolUse guard (needs bun) that denies unsafe tofu, terraform and terragrunt applies, destroys, imports and state surgery, gates subagent plan, init, state, output and aws calls, asks before destructive aws, pulumi and helmfile operations, and guards kubectl, helm, talosctl and omnictl changes on production clusters, through common aliases and wrappers. By default it no longer auto-approves plan, init or other read-only forms (option autoAllowReadOnly).

hooks

Deep-module design skill: module, interface, seam and adapter vocabulary, deepening by dependency category, design-it-twice, and a user-invoked architecture review that finds deepening candidates with codebase-memory.

1 skill

Coder agent. Implements one bounded change end to end: features, fixes, refactors and multi-file edits, with their tests and the docs the change makes stale. Callers pass `model` and `effort` on every call: haiku/medium for mechanical or fully specified edits, sonnet/medium by default, sonnet/high for concurrency, data or core invariants, opus/high only after a sonnet attempt failed or for work that cannot be split further.

1 agent

Coder agent, opus tier. concurrency, security-sensitive code, data migrations, core invariants, or a change a lower tier failed. Slower and dearer; use coder-max only if this tier fails.

1 agent

Coder agent, sonnet tier. a change coder struggled with or that needs careful design, like a tricky refactor, an API redesign or intricate tests. Use coder-critical for risky logic.

1 agent

Coder agent, fable tier. the hardest changes, like a large cross-cutting refactor or a novel algorithm, or work coder-critical could not finish. Most expensive tier; reserve it for escalations. Not in orchestration-meta: install it per repo at local scope.

1 agent

Makes the smallest complete, maintainable code change for a bounded feature, fix, integration, or behavior-preserving refactor. Use when implementing or modifying code whose approach is already clear ("add X to Y", "fix this function", "refactor this without changing behavior", "simplest solution", "this is over-engineered").

1 skill

ESP32 skills: ESP-IDF firmware engineering.

1 skill

Rust skills: the language and its tooling.

1 skill

Shell skills: Bash scripting practice.

1 skill

Cold-reads an artifact through a fresh subagent that sees only its contents and reports where a stranger would stall.

1 skill

Audits an installed Claude Code setup for bloat and rot, measures each item's context cost, and proposes keep, improve, retire or merge verdicts with soft-delete cleanup approved one item at a time.

1 skill

A home for the user's core Claude Code mods: a git status line under the prompt, a usage band above it, and an agent status line with an expandable panel, plus read-only git log, blame and show tools for subagents and tool-mgr, which blocks Bash sed file reads, heredoc, printf and echo file writes, and dangerous commands (rm -rf of system paths, disk wipes, sudo), asks before piped installers and git force-pushes, optionally adds asks and denies through a local model check, and keeps a per-call log with DuckDB metrics. Requires Claude Code 2.1.289 or later.

hooks

Writes, reviews, refactors and debugs C# and .NET code (types, nullable references, async and cancellation, exceptions, DI, options and logging in the generic host, file-based apps). Use for work in *.cs or *.csx files, "review this C# class", CS86xx nullable warnings, a deadlock or sync-over-async bug, or a rename across a solution.

1 skill

A database subagent: designs and reviews schemas and migrations, writes migration, schema and query changes in the repo when asked, and diagnoses slow queries with EXPLAIN. Never writes to shared databases.

1 agent

Portable relational schema design and review: keys, constraints and FK delete actions, normalization, access paths, expand-contract migrations with backfills, and shared-database safety rules.

1 skill

Manual command that compresses correct but accreted prose to its load-bearing density, cutting words and never rules.

1 skill

A read-only diagnosis subagent for failures with an unknown cause: reproduce first, rank hypotheses, confirm with a probe that flips the result, and return the root cause with a fix recommendation.

1 agent

Configure and troubleshoot GitHub Dependabot (dependabot.yml ecosystems, groups, cooldown, ignore rules, private registries) and sweep open Dependabot PRs, merging only green ones.

1 skill

Vet npm and PyPI packages before installing: fetch without running install code, GuardDog and OSV malware checks, lifecycle-script review, typosquat and dependency-confusion heuristics.

1 skill

Deprecation and migration skill: Hyrum's law, a sunset-or-keep decision tree, incremental consumer migration (strangler, adapter, flags), and expand/contract for zero-downtime schema changes.

1 skill

Gives structured feedback on a static design (mockup, wireframe, exported Figma frame or screenshot) covering usability, hierarchy, consistency and accessibility. Use for "critique this mockup", "review this design", "what do you think of this screen?". A rendered page or built UI goes to the ui-critic agent.

1 skill

Writes a developer handoff spec from a finished design: layout, tokens, component props, states and interactions, responsive breakpoints, content limits, edge cases, motion and accessibility notes. Use when a design is ready for engineering, or for "write handoff specs", "spec this screen for developers".

1 skill

Audits, documents or extends a design system as a whole: token inventory, naming consistency, hardcoded values, component completeness, component docs, and proposals for a new component or pattern. Use for "audit our design system", "document this component", "design a new pattern that fits". Restyling a product screen goes to web-design.

1 skill

Dev container skill: review and write devcontainer.json, Dev Container Features and Codespaces config against sourced spec rules, plus Codespaces startup and cost tuning.

1 skill

devenv.sh skill: configure, migrate, debug and verify developer environments in devenv.nix and devenv.yaml, with option shapes checked against the installed version and secrets kept out of Nix evaluation.

1 skill

Devil's advocate skill that steelmans an idea or decision, then challenges it in one chosen mode (Socratic, dialectic, pre-mortem, red team, evidence audit) and drives to a stronger position.

1 skill

Read-only-by-default infrastructure agent that reviews, validates and plans Terraform, Ansible, Docker, Kubernetes, Helm and GitOps changes using the matching skills.

1 agent

Docker skill: Dockerfiles, Compose and .dockerignore with multi-stage builds, non-root hardened runtime, daemon.json hardening, secret handling and container debugging.

1 skill

Docs audit skill: compare README, setup, API, config and CLI docs against the code, rank drift P0 to P3 with cited evidence, check commands, links and env vars with a script, and fix on request.

1 skill

Hook that, once per change set at Stop, nudges Claude with docs that mention a file this session edited or a name its edits removed. Skips plans, research, ADRs and changelogs; fails open. Needs git and bun.

hooks

Domain modeling skill: challenge and sharpen project terms in conversation and keep GLOSSARY.md (and GLOSSARY-MAP.md for multi-context repos) current as terms settle.

1 skill

Chooses and wires AI and ML in .NET apps with Microsoft.Extensions.AI (IChatClient, embeddings), Microsoft Agent Framework, vector search and RAG, ML.NET, ONNX Runtime and OllamaSharp. Use for adding an LLM call or chatbot to a .NET service, a tool-calling agent in C#, RAG over documents, or choosing between an LLM and ML.NET for a prediction.

1 skill

Collects runtime evidence from running or crashed .NET processes: dotnet-counters, dotnet-trace, dotnet-dump, dotnet-monitor, PerfView, perfcollect, crash dumps in containers and Kubernetes, mobile crash symbolication. Use for high CPU, memory growth, hangs or crashes in production, capturing a dump, or choosing a tool for Linux in k8s.

1 skill

Writes, reviews and debugs .NET native interop with P/Invoke and LibraryImport, covering signatures from C headers, string and struct marshalling, memory ownership, SafeHandle, callbacks and library loading. Use for wrapping a C library, a DllNotFoundException or AccessViolationException at a native call, or moving DllImport to LibraryImport for AOT.

1 skill

Measurement-first .NET performance work: hot-path anti-pattern scans (async, allocations, strings, LINQ, regex, JSON, I/O), BenchmarkDotNet setup and result reading, and SIMD with TensorPrimitives or Vector128. Use when C# code is slow or allocates too much, to review a hot path, write or compare benchmarks, or vectorize a loop.

1 skill

Authors, validates and packages custom `dotnet new` templates, covering template.json from an existing project, symbols and conditional content, post-actions, constraints and template packs. Use for "turn this project into a template", a template missing from `dotnet new list` after install, or reviewing a template.json before publishing.

1 skill

Writes, runs and fixes .NET tests in xUnit, NUnit, MSTest and TUnit: dotnet test syntax for VSTest vs Microsoft.Testing.Platform, filters, TRX and dumps, test-project wiring, TimeProvider and file-system seams, coverage. Use for running tests in a .NET repo, a filter that matches nothing, "VSTest or MTP?", or testing DateTime.Now code.

1 skill

Upgrades .NET projects - TFM hops from .NET 8 to 11 with breaking changes by area, legacy csproj to SDK-style, packages.config to PackageReference, AOT and trim warnings, Thread.Abort removal, test framework moves. Use for "upgrade to .NET 10", SDK-style conversion, IL2026/IL3050 warnings, or moving to xUnit v3, MSTest v4 or MTP.

1 skill

Fresh-context adversarial review of non-trivial in-flight decisions: state the claim, hand the artifact and contract to a subagent told to disprove it, reconcile findings, stop after three cycles.

1 skill

Write, run and diagnose browser end-to-end tests (Playwright, Cypress) for critical user journeys, with accessible locators, condition-based waits and flake triage.

1 skill

Writes, reviews and tunes EF Core data access: query shape, tracking, Include vs split queries, N+1, bulk ExecuteUpdate/Delete, DbContext lifetime, migrations, raw SQL parameters, and Dapper code beside it. Use for DbContext or LINQ-to-entities code, a slow EF query, dotnet ef migrations, or reviewing a repository or import job.

1 skill

ESP-IDF firmware skill for ESP32-family chips: write, review, debug and bring-up, with build/flash/monitor wrappers, templates, and confirmation gates before flashing, eFuse burns or partition changes.

1 skill

Runs a bounded, unattended optimize-and-measure loop on a dedicated git branch, one hypothesis per commit, with a fixed metric command, keep-or-revert decisions and a results log. Use when the user wants many experiments against a measurable goal such as benchmark time, bundle size, memory or test pass rate.

1 skill

Manual command that checks each factual claim in an earlier answer against files, commands and authoritative sources, reports it Confirmed, Wrong or Unverified, and restates the corrected answer.

1 skill

Bounded changes to existing FastAPI apps: lifespan, dependencies, Pydantic models, async handlers, auth dependencies, transactions and dependency_overrides tests.

1 skill

Adds, uses, tests and removes feature flags in application code with the repo's own flag tool or OpenFeature, with a reference of common flag platforms.

1 skill

Local-only file conversion skill: image, audio, video, document and CSV/JSON conversions with ffmpeg, ImageMagick, pandoc, LibreOffice and the Python stdlib, inspecting inputs first, preserving originals and verifying outputs.

1 skill

Audits a codebase for stale, orphaned and undocumented feature flags and builds or updates a flag registry with owners and removal dates.

1 skill

Create, review, harden and debug Forgejo Actions workflows for forgejo-runner: runner labels, uses: resolution, unsupported GitHub keys, fork-PR approval, secrets, and failed-run diagnosis.

1 skill

Forgejo CLI fj against Codeberg or a self-hosted Forgejo: install, per-host auth, host and remote selection, the command map and its gaps.

1 skill

Read, create, edit, close and label single Forgejo or Codeberg issues with fj, including templates, milestones, dependencies and issue search.

1 skill

Connects to a forgejo-mcp server hosted at https://<host>/mcp on a self-hosted Forgejo instance, sending your access token as the Authorization header; prompts for the host and a token kept in secure storage.

1 MCP server

Prepare and open a reviewer-ready Forgejo or Codeberg pull request with fj: convention-following commits, a body with summary visuals, evidence and merge danger, opened as a WIP draft after approval.

1 skill

Manual /forgejo-pr-comments: pull a Forgejo PR's review comments into context, triage them, and fix the ones you pick. Read-only on the forge.

1 skill

README skill for Forgejo and Codeberg repos: write, restructure, audit and beautify README.md from facts in the repo, with templates by project type, a scored audit, honest badges and Forgejo rendering rules.

1 skill

Cut, edit and publish Forgejo or Codeberg releases with fj: tags, release notes, drafts and prereleases, and release assets.

1 skill

Ship finished work on a Forgejo or Codeberg repo in one run: commit, push, and optionally open a PR with fj, wait for checks and merge, delegated to the forgejo-shipper agent.

1 skill

Agent that commits finished work, pushes it, and optionally opens a Forgejo PR with fj, waits for checks and merges, on a cheap model. Used by the forgejo-ship skill.

1 agent

Engine-agnostic game AI skill: NPC and enemy decisions with FSMs, behavior trees and utility AI, steering and flocking, A*, flow fields and navmeshes in concept, perception and difficulty knobs.

1 skill

Engine-agnostic game code structure: game loop and fixed timestep, state machines (flat, hierarchical, parallel), composition, event bus, object pools, data-driven content, inventory and ability systems, scene flow and save systems.

1 skill

Art direction and asset pipeline for game art, mainly 2D: a style bible, palette and scale rules, sprite, animation, tileset, UI and icon specs, an asset manifest with naming and licenses, image-generation briefs, and import checks. Use for a game's art style, sprites, tilesets, an asset list or sourcing art.

1 skill

Engine-neutral game audio practice: bus and mixer layout, gain in dB, ducking, adaptive music with layers and quantised transitions, SFX variation and voice limits, beat sync, and platform loudness targets.

1 skill

Game design documents and decisions: pitch, pillars, core loop, mechanic specs, progression, economy and balancing (curves, sinks and faucets, loot tables, simulation), difficulty and scope cuts. Engine-neutral templates and checklists.

1 skill

Engine-agnostic game feel: input buffering, coyote time, variable jump, deadzones, rebinding, camera follow, look-ahead and shake, hit-stop, tweens, squash and stretch, knockback and render jitter.

1 skill

Engine-agnostic genre playbooks: core systems, data model, feel targets, balancing norms and pitfalls for platformers, roguelikes, card games, puzzles, RPGs, FPS, tower defense, survival-crafting and visual novels, plus branching dialogue (Ink, Yarn Spinner, custom runners).

1 skill

Engine-agnostic game performance, measured first: frame budgets, profiling a release build on target hardware, CPU versus GPU triage, draw calls and batching, overdraw, allocation and GC hitches, pooling, asset budgets and regression checks.

1 skill

Game playtests: research questions, test type, recruiting and consent, session scripts and observer sheets, think-aloud, surveys, lightweight telemetry, ranked change lists, and keep-or-kill criteria for a build.

1 skill

Timeboxed game prototyping: a greybox mechanic spike or vertical slice that answers one question with keep or kill criteria, and game-jam planning (scope to the clock, hour schedule, cut list, submit early). Use when prototyping a mechanic, greyboxing, or planning a jam or 48-hour build.

1 skill

Ships a game build to a store and plans the release: itch.io uploads with butler channels and versions, Steam with SteamPipe depots, app_build.vdf, steamcmd and branches, and Steam release timing (app fee wait, Coming Soon, reviews, Next Fest, discounts). Use for butler push, steamcmd or a Steam launch date.

1 skill

Engine-neutral game UI and UX: HUDs, menus and overlays, a screen stack, resolution and aspect scaling, safe areas, gamepad focus navigation, event-driven HUDs, settings screens, and a game accessibility checklist.

1 skill

Create, review, harden and debug GitHub Actions workflows and custom actions with gh: least-privilege permissions, action pinning, script-injection and pull_request_target checks, and failed-run diagnosis.

1 skill

Configure and triage GitHub security features with gh: secret scanning, push protection and blocked-push recovery, path exclusions, private vulnerability reporting, and alert triage.

1 skill

Read, create, edit, close and label single GitHub issues with gh, including issue types, sub-issues, blocked-by links, milestones, Projects V2 fields and issue search.

1 skill

Prepare and open a reviewer-ready GitHub pull request with gh: convention-following commits, a body with summary visuals, evidence and merge danger, opened as a draft after approval.

1 skill

README skill: write, restructure, audit and beautify a GitHub repository's README.md from facts in the repo, with templates by project type, a scored audit and link-check script, honest shields.io badges, dark/light logos and GitHub rendering rules.

1 skill

Cut a manual GitHub release with gh: detect release tooling first, propose a SemVer bump from the public-API diff, draft a Keep a Changelog entry, open a release PR, then tag and publish after approval.

1 skill

Audit and set up a GitHub repository's surface with gh: labels, issue forms, PR template, CODEOWNERS, SECURITY.md, CONTRIBUTING.md, and branch or tag rulesets, with a dry-run plan.

1 skill

Triage existing GitHub issues and external PRs with gh: category and state labels, duplicate and prior-rejection checks, claim verification, needs-info notes, agent briefs and stale policy.

1 skill

Writes and creates Git commits in the repository's own message convention (Conventional Commits, release-please, commitlint) and names new branches to match it.

1 skill

Teaches Claude the git-flow branching model: detecting it, feature/release/hotfix procedures with git-flow-next, AVH or plain git, and when to pick another model.

1 skill

Edits and repairs Git history safely without opening an editor: fixup and autosquash, rebasing onto the default branch or a new base, splitting commits, cherry-pick backports, bisect, and reflog recovery.

1 skill

Sets up and reviews GitOps delivery with Argo CD or Flux: choosing between them, repo layout, sync policy per environment, and secrets with External Secrets, Sealed Secrets or SOPS.

1 skill

Godot 4 entry point: GDScript, nodes and scenes, signals, custom Resources, autoloads, project.godot and safe .tscn/.tres edits (uid:// and .uid files), with references for 2D, 3D, UI, input, audio, animation, saves, performance and debugging, and routing to the other godot plugins and the editor MCP.

1 skill

Godot addon APIs for Beehave and LimboAI behavior trees, Phantom Camera and Dialogue Manager v4. Use when a project has addons/beehave, addons/limboai, addons/phantom_camera or addons/dialogue_manager, or when choosing, installing or writing code and .dialogue files for one of them.

1 skill

Writes and fixes C# in Godot 4 .NET projects: partial node classes, the PascalCase lifecycle, [Export], [Signal] events, GDScript interop, the .csproj and the build. Use for .cs or .csproj files beside project.godot, "port this GDScript to C#", GD0201 errors, or ObjectDisposedException after a signal.

1 skill

Godot 4 editor plugins and addons. Use when writing or fixing an EditorPlugin, plugin.cfg, @tool script, inspector plugin, EditorDock, 3D gizmo, tool menu, undoable scene edits or an addon that runs a server inside the editor, or when a project has addons/*/plugin.cfg.

1 skill

Godot 4 export skill: export templates and presets, headless CLI exports, Web threads and COOP/COEP, Windows and macOS signing, dedicated-server builds, PCK patches and the Shader Baker, with references for CLI and presets, Android and iOS, and CI export jobs.

1 skill

Native Godot 4 extensions with GDExtension, in C++ (godot-cpp) or Rust (godot-rust gdext). Use when a project has a .gdextension file, SConstruct or a godot-cpp submodule, or when binding a native class, building per-platform libraries, wrapping a C/C++ library, or debugging a native crash.

1 skill

Manual /godot-mcp-setup: installs the voidrot/godot-mcp editor addon into a Godot 4 project, enables it, reads its port and writes the project .mcp.json godot-editor entry after you approve the diff.

1 skill

Builds and secures networked Godot 4 games: server-authoritative @rpc validation, ENet and WebSocket peers, MultiplayerSpawner and MultiplayerSynchronizer, lobbies, dedicated servers, interpolation and prediction. Use for @rpc, "multiplayer", "netcode" or a headless server in a Godot project.

1 skill

Godot 4 physics skill: CharacterBody controllers and move_and_slide, rigid, static, animatable and area bodies, collision layers and masks, raycasts and space queries, physics interpolation and Jolt, with references for controller recipes, queries, ragdolls and soft bodies.

1 skill

Godot 4 shaders and VFX skill: .gdshader canvas_item, spatial and particles shaders, uniforms and hints, screen and depth reading, visual shaders and materials, GPU particles, with references for shader fundamentals, effect recipes, particles, stencil, CompositorEffect and the Shader Baker.

1 skill

Godot 4 testing skill: picks GUT, gdUnit4 or a bare SceneTree runner from the project, writes tests, runs them headless with real exit codes, and sets up CI, with GUT and gdUnit4 references.

1 skill

Godot 4 XR with OpenXR for VR, AR and passthrough on Meta Quest, Pico and PC VR. Use for XROrigin3D and XRController3D setup, the action map, locomotion, grabbing, in-world UI, hand tracking, passthrough, spatial anchors, or a Quest export, and when project.godot has xr/openxr/enabled=true.

1 skill

Writes and reviews GO Feature Flag flag files (YAML, JSON, TOML): variations, targeting rules and query syntax, percentage, progressive and scheduled rollouts, experimentation, metadata, and the go-feature-flag-cli lint and evaluate commands.

1 skill

Configures and runs the GO Feature Flag relay proxy: config file keys, retrievers, exporters, notifiers, API keys, flagsets, admin and OFREP endpoints, Docker and Kubernetes deployment.

1 skill

Integrates GO Feature Flag through OpenFeature providers (server, in-process, client) and the Go module: evaluation context, bucketing, tracking and testing.

1 skill

Go language skill: packages, errors, interfaces, context, goroutines and shutdown, modules, tests, with references for pitfalls, security, database/sql, gRPC, linting and modernizing.

1 skill

Go command-line tools with cobra and viper: command tree and hooks, flag, env and config precedence, output and exit codes, signal shutdown, testable commands.

1 skill

Measurement-first Go performance: benchstat comparisons, pprof and trace reading, escape analysis, GC and GOMEMLIMIT tuning, and CI regression checks.

1 skill

Diagnoses and writes Gradle build logic, including configuration cache failures, Could not resolve and no matching variant errors from conflicting versions or platforms, dependencyInsight, build cache misses, Kotlin DSL build scripts, libs.versions.toml version catalogs, convention plugins in build-logic, and toolchains. Use for symptoms such as "invocation of Task.project at execution time is unsupported", "Configuration cache problems found in this build", "Could not resolve" with a platform or BOM, "Unresolved reference: libs", or a request to add a task, dependency or convention plugin to build.gradle.kts or settings.gradle.kts.

1 skill

Grafana dashboards skill: dashboards built from operator questions with RED and USE panel sets, a quality checklist, current panel types and JSON models, and dashboards as code via provisioning, Terraform or the Foundation SDK.

1 skill

grep.app MCP server (remote HTTP, no API key): literal and regex code search across about a million public GitHub repositories.

1 MCP server

Compacts the current conversation into a handoff document a fresh session can pick up, and with --bg launches that session as a background agent. Manual command.

1 skill

Authors, validates and releases Helm charts: values and schema, helpers, dependencies, lint and template checks, helm diff, atomic upgrades, OCI publishing and provenance.

1 skill

Home and homelab network planning: VLANs, trust zones, DNS cutover, and WireGuard tunnels, read-only first.

1 skill

Home and homelab network planning, read-only first: inventory, VLANs and trust zones, default-deny firewall rules, Pi-hole DNS cutover, WireGuard remote access and interface error diagnosis.

1 skill

Teaches Claude to work in repos that use HubFlow (git hf), the archived GitHub-oriented git-flow fork, with warnings about its auto-pushing and merge behaviour.

1 skill

Executes an approved multi-slice implementation plan file end to end, running independent slices in parallel worktrees and checking, reviewing, committing, and updating a Status block after each slice without asking. Use when the user says "implement this plan", "continue implementing", or points at a plan file to carry out.

1 skill

Turn a spec, plan or conversation into a reviewed set of linked GitHub issues cut as vertical slices with blocking edges, writing a spec first when needed, and publish them with gh after a dry run and approval.

1 skill

Writes, reviews, debugs and upgrades modern Java (17+) code: records, sealed types, pattern-matching switch, streams, Optional, exceptions, equals/hashCode, executors, CompletableFuture, virtual threads, JDK upgrades and javax.* to jakarta.* migration. Use whenever editing or debugging Java source (*.java), such as a switch over a sealed interface or enum that throws MatchException or is not exhaustive after a new subtype was added, javax.servlet or javax.persistence imports that no longer compile on Tomcat 10 or Jakarta EE 9+, a record whose List changes after construction, a thread-pool or interrupt bug, "review this Java class", or moving a project to a newer JDK.

1 skill

Covers JavaScript runtime semantics, meaning ESM vs CommonJS, import side effects, promises and the event loop, async cleanup and cancellation, browser vs Node/Bun/Deno/edge, measured performance and JS tests. Use when writing, reviewing or debugging *.js/*.mjs/*.cjs or runtime behavior of JS/TS code (import errors, unhandled rejections, leaks, ordering bugs).

1 skill

Writes, runs, debugs and fixes Java tests with JUnit Jupiter 5/6, Mockito, AssertJ and Testcontainers. Use whenever editing or debugging JUnit test code (*Test.java, *IT.java): rewriting copy-pasted cases as one @ParameterizedTest with @CsvSource or @MethodSource, @Nested tests, Mockito UnnecessaryStubbingException or a when(...) stub that is never used, strict stubs, "No tests found for given includes", filtering with -Dtest or --tests, Surefire versus Failsafe, flaky tests that sleep or depend on the clock, a slow suite that starts a container per class, and Mockito agent warnings on Java 21+.

1 skill

Writes everyday Jetpack Compose UI. Use for "LaunchedEffect restarts on every keystroke", state that resets on recomposition, a card or button component with many boolean flags, an ignored modifier, or adding a screen, theme or animation. Covers where state lives and which effect runs which work, component APIs with modifier and slot parameters, LazyColumn keys, unstable parameters, Material 3 theming, animation, focus and previews.

1 skill

Writes, reviews and fixes idiomatic Kotlin outside @Composable code. Use for "Smart cast to String is impossible, because name is a mutable property", a non-exhaustive when, !! NullPointerException, events lost from a MutableSharedFlow, stateIn that never emits, a coroutine that outlives its owner, GlobalScope or runBlocking freezing the caller, swallowed CancellationException, or a Java-to-Kotlin conversion. Covers null safety, sealed classes, scope functions, structured concurrency, Flow and Java interop.

1 skill

Writes, runs and fixes JVM Kotlin tests (kotlin.test, JUnit, Kotest, MockK, coroutines-test, Turbine). Use for UncompletedCoroutinesError or a runTest that hangs, "Module with the Main dispatcher" errors and Dispatchers.setMain in ViewModel tests, MockK "no answer found for", coEvery versus every, Turbine flow assertions, tests slowed by delay, and running one test with ./gradlew :module:test --tests.

1 skill

Writes and reviews Kubernetes manifests (probes, resources, RBAC, NetworkPolicy, Pod Security Standards and admission, Jobs, HPA, PDB, Kustomize) and triages failing workloads with read-only kubectl.

1 skill

Audits an eval, metric, experiment or benchmark for leakage and reports which independence checks fail, each with a fix.

1 skill

Manual command that splits one artifact across distinct lenses and returns the question that resolves their disagreement.

1 skill

Engine-agnostic level design: player metrics, blockout to playable, critical path and gating, pacing and tension curves, encounters, readability, and seeded procedural generation (noise, rooms and corridors, BSP, cellular caves, WFC).

1 skill

Edits and debugs Maven builds in pom.xml, .mvn and mvnw projects, including dependency management and BOM imports, NoSuchMethodError or duplicate jars from conflicting versions, multi-module reactor failures such as a sibling SNAPSHOT that won't resolve, plugin phases, annotation processors, settings.xml mirrors, and dependency audits before a release. Use when a pom.xml is being changed, a mvn or mvnw command fails, the user asks which module pulls in a library, or the user asks to check dependencies for updates or vulnerabilities.

1 skill

MCP server skill: design the tool surface (workflow tools, descriptions, write-tool safeguards, auth) and build it with the official TypeScript, Python, Go or Rust SDK over stdio or Streamable HTTP, tested with the Inspector.

1 skill

MCP server audit skill: checks configured or third-party MCP servers for tool poisoning, shadowing, rug pulls, SSRF and unauthenticated exposure, with offline description scanning and hash pinning.

1 skill

MCP server review skill: static security review of server or client source and a release QA pass over one real local protocol session, with OWASP MCP Top 10 checks.

1 skill

mise skill: configure mise.toml tools, env and tasks, migrate .tool-versions, and run commands under mise in non-interactive shells.

1 skill

Mistake-proofing (poka-yoke) skill: design or audit code so mistakes cannot be expressed, ranking hazards on a control, warning, detection and docs-only ladder, and sweeping for the hazard class after an incident.

1 skill

Plans the modernization of a legacy system (runtime, framework or dependency upgrade, migration or rewrite) with a feasibility spike, testability milestone, safety ladder, hazard red-team and phased exit criteria. Use for "modernize this", "plan the migration", "how do we get off this stack", or a major version upgrade plan.

1 skill

Diagnoses MSBuild failures and slow builds and reviews .NET build XML. Use for a failing dotnet build, restore, pack or publish, .csproj/.props/.targets conditions and custom targets, binlog capture and replay, "it rebuilds when nothing changed", slow evaluation or builds, bin/obj clashes, or Directory.Build.* layering.

1 skill

MySQL and MariaDB skill: EXPLAIN risk signals, composite indexes, upsert syntax differences, keyset pagination, InnoDB deadlocks, SKIP LOCKED queues, online DDL, replica lag, pools and grants.

1 skill

Manages NuGet dependencies and packages in .NET repos. Use for Directory.Packages.props and Central Package Management, bumping or aligning package versions, NU1xxx restore warnings, nuget.config sources and source mapping, vulnerable-package audits, lock files, dotnet pack metadata, or publishing to nuget.org with trusted publishing.

1 skill

Observability instrumentation skill: structured logs with correlation IDs, metrics with bounded labels, OpenTelemetry traces, profiling, symptom alerts, and SLOs with burn-rate alerts, starting from the questions on-call will ask.

1 skill

A read-only subagent that audits a repo before it goes public: secrets (gitleaks or trufflehog, regex fallback), PII, internal references, dangerous files, .env.example gaps, git history and unreadable binaries, returning PASS, PASS WITH WARNINGS or FAIL.

1 agent

SessionStart hook that tells Claude to record decisions, runbooks and other knowledge worth keeping in Outline as well as the repo. Installs outline-mcp.

hooks

Outline's built-in MCP server (remote HTTP) for a self-hosted Outline wiki, signing in with OAuth through /mcp or with an API key from ~/.config/outline/api-key. Prompts for the hostname.

1 MCP server

Proton Pass CLI skill: find vaults and items by ID, build pass:// references, and inject secrets into a process or file without ever printing them.

1 skill

How to use the plan-ledger plan tool (mcp__plan-ledger__plan): the orchestrator and subagent call sequences, the verifyChecks setting and parallelSafe items. Preloaded by the agents that list the tool.

1 skill

Writes, reviews, or revises implementation plans whose scope, verification, and reruns are proportionate to risk. Use for "plan this", "review this plan", "cheapest way to verify", work spanning several slices or risky boundaries, acceptance criteria for an ambiguous change, multi-agent or multi-session plans, and "grill me" stress-tests.

1 skill

Builds, debugs, tests and upgrades Play Framework 2.9/3.x web apps in Java or Scala. Use for conf/routes and reverse routes, app/controllers (play.mvc, CompletionStage, Action.async), Twirl templates, Guice DI, play-json or Jackson, WSClient, evolutions, CSRF and AllowedHosts filters, sbt stage/dist packaging, blocking JDBC stalling a Play server, a "not found: value controllers" routes error, and the Play 2.8/2.9 to 3.0 upgrade (Akka to Pekko, com.typesafe.play to org.playframework).

1 skill

Sets up and runs the release path to Google Play. Use for "Version code 42 has already been used", "signed with the wrong key", a lost upload key, shipping the AAB to the internal track from CI, staged rollouts, or the target API deadline. Covers upload key versus Play App Signing, Gradle Play Publisher, fastlane supply, tracks, mapping upload, store listing and data safety drafts. Never publishes without an explicit yes.

1 skill

Drive a headless, isolated browser from Bash with playwright-cli through a verb-checked wrapper: inspect pages, fill forms, take screenshots, debug console and network errors, and run smoke QA.

1 skill

A read-only subagent that statically vets a third-party plugin, skill, agent, hooks.json or .mcp.json before install for hidden directives, invisible unicode, encoded payloads, tool shadowing and scope mismatch, with a PASS, FAIL or NEEDS HUMAN REVIEW verdict.

1 agent

PostgreSQL skill: EXPLAIN reading and pg_stat_statements, index choice, statistics, locks, vacuum, isolation retries, online DDL, partitioning, RLS cost, Postgres types and table-design gotchas, and pgvector.

1 skill

Blameless postmortem skill: gap-flagged UTC timeline, 5 Whys to a fixable system gap, impact and detection gaps, owned and dated actions, plus a quick template for minor incidents and runbook rules.

1 skill

Manual /pr-comments: pull a PR's unresolved review threads and review summaries into context, triage them, and fix the ones you pick. Read-only on GitHub. Not for Forgejo or Codeberg PRs (/forgejo-pr-comments).

1 skill

Commit-time hooks with prek or pre-commit (and Husky for Node): pinned config, installs, all-files runs, rev updates, CI parity, and repairing a failing hook without bypassing it.

1 skill

Prisma ORM 6 and 7 traps for TypeScript backends, such as updateMany returning a count, interactive transaction timeouts, migrate dev resets, edited applied migrations, undefined filters, P2002/P2025/P2003 handling, serverless connections and driver adapters. Use when writing a Prisma schema, query, transaction or migration, or debugging one.

1 skill

Privacy and data-handling review: inventories personal data, traces where it flows and rests, and checks minimization, retention, erasure, export, and PII in URLs and logs.

1 skill

Root-cause diagnosis for failures, flaky tests and performance regressions: a red-capable feedback loop, competing falsifiable hypotheses, one-variable probes and evidence-backed fixes. Also algorithm choice under unclear constraints.

1 skill

Evidence-based ship-or-block review (Blocked, Risky or Caveats) of an application before launch, built from local repo evidence.

1 skill

PromQL and Prometheus skill: correct counter and histogram queries, query cost limits, RED and USE investigations, data gaps, scrape and rule config, Alertmanager routing, and promtool checks, with credentials kept out of commands.

1 skill

Builds a throwaway prototype that answers one design question, either a single-file HTML demo that drives a state model through hard cases or several structurally different UI variants on one route. Use to sanity-check a state model, data shape or API feel, or to compare layouts before committing.

1 skill

Python idiom, review and diagnosis: typing, exceptions, resource lifetimes, asyncio, retries and background jobs, profiling and pytest.

1 skill

Build and validate Python distributions and uv workflows: pyproject.toml, package data, entry points, wheel and sdist checks, uv in CI and Docker, and PyPI trusted publishing.

1 skill

Changes, reviews and debugs React components: state ownership, effects, keys, render purity, Server Components and Actions, accessibility, measured performance, React 19 upgrades and component tests. Use for .tsx/.jsx behavior or render cost, such as "this list lags when I type", "it re-renders too much", effect loops or hydration mismatches.

1 skill

PreToolUse hook that limits the read-only agents (architect, reviewer, debugger, security-auditor, open-source-release-audit, plus any you list) to Bash commands, and Write/Edit/MultiEdit/NotebookEdit calls, that don't write outside /tmp and the session scratchpad. architect, and any plan writers you list, may also save Markdown plans with Write/Edit under .claude/plans or docs/plans in the project (CLAUDE_PROJECT_DIR) or in a git worktree of it that holds the agent's cwd. Needs bun.

hooks

Inspects the current project, proposes the adminhead meta plugins, bundles and focused plugins that fit it from the live catalog, and installs the ones you confirm at local or project scope. Manual command.

1 skill

Redis and Valkey skill: data structures, caching patterns, TTL and key design, eviction policy, stampede control, rate limits, token-checked locks, Streams vs Pub/Sub, and connection handling.

1 skill

Configure and troubleshoot Renovate (renovate.json, presets, packageRules, schedules, automerge), run it for Forgejo (platform=forgejo), and sweep open Renovate PRs, merging only green ones.

1 skill

Manual command that realigns an arbitrarily ordered list, table or section set under one stated principle, moving items only.

1 skill

Answers an external, decision-relevant question with sourced, cited evidence at quick, standard or deep depth. Use for "compare X vs Y", "is it true that...", "what's the current state of...", choosing a library or tool, or explaining a topic from authoritative sources.

1 skill

A read-only review subagent for implementation plans, designs and diffs. It runs the checks and verifies the work against the brief and its acceptance criteria, returning pass or fail per criterion with evidence and findings ranked by severity. The caller picks quick or thorough depth.

1 agent

Review, fix and re-review loop: built-in /code-review finds issues, a fixer subagent fixes the flagged IDs, and a fresh re-reviewer checks only those IDs from a shared findings file, for up to three rounds.

1 skill

Plans and verifies production rollouts: strategy choice, preflight go/no-go, phased verification signals, numeric rollback criteria and post-deploy follow-up.

1 skill

Roslyn analyzer skill: DiagnosticAnalyzer, CodeFixProvider and IIncrementalGenerator design and tests, Microsoft.CodeAnalysis host-compatibility pinning, and analyzer NuGet packaging.

1 skill

Rust language skill: ownership and lifetimes, error design, traits and modules, Cargo features, async and concurrency, unsafe review, clippy and measured performance.

1 skill

Writes, edits and debugs sbt builds: build.sbt, project/*.sbt and project/build.properties, sbt 1 and sbt 2 command-line runs, slash keys such as Test/compile, dependency resolution and % or %% coordinates, eviction and version conflicts, cross-built Scala versions, multi-project builds and failing sbt tasks. Use when an sbt command errors with Expected whitespace character, Not a valid command or No such setting/task, when a test does not run or rerun under sbt, or when a build.sbt needs a dependency or cross-build change.

1 skill

Plan, run with the user and report manual screen-reader and keyboard tests of a built web journey (NVDA, JAWS, VoiceOver, Orca, TalkBack).

1 skill

A documentation subagent that writes and updates standalone docs (READMEs, guides, changelogs and Outline wiki documents), checking every claim against the code and marking anything it could not verify. Docs that a code change makes stale stay with the coder making the change.

1 agent

Second-opinion skill: pick codex (the default) or agy, model, effort and pass size for a read-only outside review of a diff or plan through the consult tool, weigh the replies, or run a local two-reviewer gate.

1 skill

A read-only security review subagent for diffs and code areas: taint tracing, auth, injection, secrets, supply chain, CI and IaC trust, returning severity-ranked findings with attack paths.

1 agent

Manual security audit of a branch: gitleaks, semgrep, trivy, govulncheck and pip-audit in one script, subagent verification, a ranked findings table, plus CI and supply-chain trust checks and SBOM guidance.

1 skill

Instrument apps with official Sentry SDKs.

1 skill

Sentry's remote MCP server (OAuth) for querying issues, events, traces, and releases on sentry.io. Off until enabled.

1 MCP server

Sentry's MCP server (@sentry/mcp-server 0.42.0, pinned) over stdio for self-hosted Sentry, read-only by default. Off until enabled; prompts for the host and a user auth token kept in secure storage.

1 MCP server

Sentry SDK skill: init, errors, tracing, structured logs, metrics and crons across platforms, driven by the project's ## Sentry profile, with current-docs lookups and notes for self-hosted Sentry.

1 skill

Manual command that recovers a Claude Code session that died without a handoff: a script digests its transcript into anchors, and the skill verifies them and writes a handoff-shaped brief for a fresh session.

1 skill

Runs a retrospective on a Claude Code session and proposes fixes to the agent's environment (checks, hooks, navigation pointers, steering files, tool economy) ranked by severity. Use when the user asks for a retro, a post-session review, a review of recent or past sessions, or why a session went badly.

1 skill

Setup skills: recommend and install the adminhead plugins that fit the current project.

1 skill

Manual skill that generates a bash wizard for human-only setup steps: opens each dashboard, captures secrets with hidden input, writes a gitignored .env and sets GitHub secrets.

1 skill

Ship finished work on a GitHub repo in one run: commit, push, and optionally open a PR, watch checks and merge, delegated to the shipper agent.

1 skill

Agent that commits finished work, pushes it, and optionally opens a GitHub PR, watches checks and merges, on a cheap model. Used by the ship skill.

1 agent

Writes, reviews and debugs Spring Boot 3/4 applications. Covers @ConfigurationProperties and profiles, bean wiring and auto-configuration ("No qualifying bean", condition report), MVC and WebFlux controllers, validation and ProblemDetail, Spring Data JPA and @Transactional, Spring Security SecurityFilterChain, Actuator, and the 3.x to 4 upgrade (modular starters, Jackson 3). Use for @SpringBootApplication code, application.yml or application.properties, spring-boot-starter-* dependencies, startup failures such as "required a bean of type ... that could not be found" after moving classes, and a @Transactional method that does not roll back.

1 skill

Writes and fixes Spring Boot 3/4 tests that need a Spring context. Covers test slices (@WebMvcTest, @DataJpaTest, @RestClientTest, @JsonTest, @WebFluxTest), @MockitoBean and @MockitoSpyBean, MockMvcTester, RestTestClient, Testcontainers with @ServiceConnection, Spring Security tests, and slow suites that rebuild the application context.

1 skill

Finds every copy of one fact scattered across code, config, CI and docs, proposes a canonical home and a consolidation plan, and edits only after approval.

1 skill

Works with TanStack Query, Router, Form, Table, Virtual and Start: state ownership, query keys, cache invalidation, loaders, search-param validation, server-driven tables, virtualization and SSR boundaries. Use when code imports @tanstack/*, or for stale or duplicate data, lost form values, table and pagination mismatches, or Start server functions.

1 skill

Drive and test terminal UI apps with the termwright CLI: YAML step files that assert state transitions, one-shot screen and PNG capture, and daemon sessions, with safety rules for real input.

1 skill

Terraform, OpenTofu and Terragrunt skill: safe plan and saved-plan apply, plan review, state and refactoring blocks, Terratest, and security scanning with Trivy, Checkov and OPA.

1 skill

Read-only audit of missing, weak and stale tests for a repo, PR or feature: maps source to tests with a bundled script and ranks gaps P0-P3 by behavior and risk, never by coverage percentage.

1 skill

Manual test-run command: detects the test runner, runs the tests affected by your changes, prints only compact failures, and fixes them in a bounded loop. A custom --cmd needs run_custom.py and a prompt.

1 skill

STRIDE threat modeling from the code: trust boundaries, data-flow diagram, verified findings tiered by exploitability, attack trees, testable security requirements and mapped controls.

1 skill

Tilt skill: write and debug Tiltfiles (docker_build, k8s_yaml, live_update, local_resource) for local Kubernetes dev loops, and run tilt non-interactively.

1 skill

Teaches Claude trunk-based development: short-lived branches, feature flags, branch by abstraction, release from trunk or late-cut release branches with cherry-picks.

1 skill

Sets up dependency-cruiser in a TypeScript repo so each package is reachable only through its root entry-point files, with no deep imports and no cycles, then proves the rules fail on a violation. Use when the user asks to enforce package boundaries, stop deep imports or add dependency-cruiser.

1 skill

Manual command that strips the session's framing and fans out fresh reads to check whether the current direction is tunnel-visioned.

1 skill

Handles TypeScript type-level and compiler work, covering tsconfig strictness, narrowing, generics, discriminated unions, declarations and module resolution, public type APIs, runtime validation of unknown data, and type tests. Use when editing *.ts/*.tsx types or tsconfig.json, or fixing tsc errors.

1 skill

Renders a built web page or local HTML file headless at 1440/768/390 px, runs lint and inspect checks, critiques the screenshots and returns at most three ranked fixes with FAIL/WARN counts. Use after building or restyling a page, or when asked why a page looks generic. Not for mockups, e2e tests or WCAG audits.

1 agent

Art direction and build process for new UI or a visual reshape, such as a landing page, marketing site, portfolio, product page or a new app screen with no design system. Grounds the design in the subject, deals a researched direction, plans tokens and checks the render. Use when a page should look designed, or looks generic, templated or AI-made.

1 skill

Offline lookup of cited UI/UX rules (WCAG 2.2, touch, forms, navigation), 22 stack UI guideline sets including SwiftUI, Compose, Flutter, React Native, WPF, WinUI and Avalonia, Google Fonts metadata and licences, and curated icons.

1 skill

Focused unit and regression tests and test-strategy guidance for changed behavior or a confirmed bug, in any language: smallest proving test, test-first slices, seams and boundary-only mocks, plus pytest and Vitest/Jest references. No coverage quotas.

1 skill

Writes and reviews product microcopy: button and CTA labels, error messages, empty states, confirmation dialogs, tooltips, loading and onboarding text. Use for "what should this button say?", "write copy for", "review this error message" or wording a dialog. Not for marketing page copy or long-form docs.

1 skill

Manual command that plans a restart from v0, keeping only the contracts, gates, vocabulary, tests and failing cases the previous attempt proved, and dropping its accidental architecture.

1 skill

Rewrites a drifted document as a clean v0 that says what is true now, with no trace of the edits that produced it.

1 skill

Decides whether an external repo, skill, plugin or MCP server is worth adding to this marketplace, dedupes against past verdicts first, and records one of five fixed verdicts in the research ledger. Installs nothing; plugin-vetter checks anything adopted.

1 skill

Configures and debugs Vite 7 and 8 (Rolldown/Oxc): env variables and client-bundle exposure, dev vs build vs preview, chunking, library mode, SSR externals, pre-bundling, proxy and container access. Use when editing vite.config.*, when a build differs from dev, an env value is missing or leaked, chunks go stale after a deploy, or the dev server is slow.

1 skill

Builds and debugs Vue 3 components with the Composition API, Pinia stores and Nuxt 4 data fetching: props, emits, defineModel, composables, reactivity, watchers, SSR safety and tests. Use when adding, changing or debugging .vue files, Pinia stores or a Nuxt app, such as lost reactivity, stale state, hydration mismatches or duplicate fetches.

1 skill

Extend an existing product's visual system: hierarchy, responsive, state and accessibility intent, DESIGN.md, and review of the rendered UI against it.

1 skill

Builds or repairs a browser-facing web feature end to end: semantic HTML and forms, responsive CSS, keyboard and focus, request and response contracts, server validation, error mapping and measured performance, verified in a real browser. Use for a form, page or flow that must work for real users, an HTTP API contract change, or a browser or network bug.

1 skill

Agent that audits a web app's performance: a static source scan marked not measured, or a deep audit from Lighthouse, PageSpeed or CrUX data, with no fabricated numbers.

1 agent

Mod that lists the installed agent types, each with when to use it, in the Agent tool's description in an orchestration session (CLAUDE_ORCHESTRATE set, or the main thread runs the orchestration agent), so new agents get routes without editing it.

A mod that adds a live-agent count by status and task done/total above the prompt, with an `agents` pane giving each agent's model, time in status and time since spawn, which the built-in agent panel lacks.

Plugin set for native Android projects, installed at project scope. Ships no components of its own.

Plugin set for AWS projects (credentials and IAM, CLI, the AWS MCP Server, inspection, incidents, reviews), installed at project or local scope. Ships no components of its own.

The adminhead default set, installed at user scope: the plugins every session should have. Ships no components of its own.

Mod that denies Bash used to smuggle encoded or env-var payloads, destroy systems, infra and data, kill processes by name, or pass grep --no-ignore, and denies any Bash or file-tool touch of Claude Code's credentials file. With toolSteering on (off by default), it also denies Bash that shows file contents (use Read) or writes and edits files (use Write/Edit). Commands the parser is unsure about go, with common secrets masked, to a local Ollama, then to Haiku; a :cloud model or a remote Ollama server is used only with ollamaCloud on.

A mod that shows when the prompt cache likely expires (`cache warm → ~14:32`) under the prompt, turning to a cold warning past a configured TTL. Approximate: last main-loop turn plus the TTL.

A mod that registers the mcp__consult__consult tool: a read-only diff or plan review by codex (OpenAI, the default, on the user's locked codex-consult app server) or agy (Google Gemini), with a repo opt-out marker, a secret-file block, timeouts and a read-only agy hook gate.

Core bundle of UI and UX design plugins that the web and design metas install. Ships no components of its own.

Plugin set for design work (new UI direction, rendered critique, UX rules and copy, accessibility, design systems and handoff) in any repo, installed at project scope. Ships no components of its own.

Plugin set for documentation-heavy projects, installed at project scope. Ships no components of its own.

Plugin set for C# and .NET projects, installed at project scope. Ships no components of its own.

Plugin set for Forgejo and Codeberg workflows (fj CLI, issues, pull requests, Actions, Renovate, READMEs, the Forgejo MCP server), installed at project scope. Ships no components of its own.

Core bundle of engine-agnostic game-development plugins that the game-engine metas install. Ships no components of its own.

Plugin set for GitHub workflows (issues, pull requests, Actions, Dependabot), installed at project scope. Ships no components of its own.

A mod that draws a git line under the prompt: repo path, branch, linked-worktree marker, ahead/behind with colour past unpushed-commit thresholds, staged/modified/new/conflict counts, changed lines and stash count.

Plugin set for Godot 4 game projects, installed at project scope. Ships no components of its own.

Plugin set for Go projects, installed at project scope. Ships no components of its own.

Plugin set for infrastructure-as-code projects, installed at project scope. Ships no components of its own.

Plugin set for Java projects, installed at project scope. Ships no components of its own.

Core bundle of Kotlin language plugins that the Kotlin and Android metas install. Ships no components of its own.

Plugin set for Kotlin projects without Android, installed at project scope. Ships no components of its own.

Plugin set for Kubernetes projects (manifests, Helm, GitOps), installed at project scope. Ships no components of its own.

A mod that shows `mcp: <server> ✗` under the prompt while an MCP server's latest tool call errored, hidden while all are fine, with an optional toast on each server's first error.

Plugin set for Node.js and TypeScript projects, installed at project scope. Ships no components of its own.

Orchestrated sessions: a main orchestration agent, the hook that builds its routing rules from the installed agents, and a set of general-purpose agents. Ships no components of its own.

Work-plan tracker: the orchestrator hands items to subagents by reference; the agent claims done and the mod verifies.

Core bundle of general programming plugins that the language and web metas install. Ships no components of its own.

Plugin set for Python projects, installed at project scope. Ships no components of its own.

Plugin set for Rust projects, installed at project scope. Ships no components of its own.

Plugin set for Scala projects, installed at project scope. Ships no components of its own.

Plugin set for long-running, multi-session projects, installed at project scope. Ships no components of its own.

A mod that shows a toast after long main-loop turns with the duration, output tokens and cache hit rate (`turn 3m12s · 41k out · cache 92%`).

A band above the prompt with context fill by category and rate-limit windows coloured by pace, a toast when a window turns amber or red or resets, plus session cost, cache hit rate, compactions and effort.

Mod that logs skill use and the skill listing per prompt to local JSONL, for finding skills that should have fired and didn't. Ships usage-report, which summarizes this log and the bash-rules log.

Plugin set for web front-end projects, installed at project scope. Ships no components of its own.