---
title: "Skills"
description: "Every skill in the practice, and when an agent uses it."
---

This page is generated. Do not edit it: change the source it names, and the next build changes this page.

### `herdr`

Source: `skills/herdr/SKILL.md`

Run coding agents as hc workers in Herdr (herdr-app with herdr-core), each in a pane the operator can watch. Use inside Herdr (HC_TERMINAL_ID is set) to start, show, steer, replace or clean up an hc worker, to wait for a worker and read what it reported back, or to edit the hc worker templates (owner, implementer, fixer, reviewer). Also use it when the operator wants to watch a worker build, or a milestone or long build should run where the operator can see it. Not for short pstack fan-outs (swarm, arena, interrogate, how, why), which stay Agent subagents, or for Herdr's own setup (themes, tabs, machines, mods).

### `house-architecture`

Source: `skills/house-architecture/SKILL.md`

The house rules for how code is shaped, in every language. Use when you design or change a feature, module, service or CLI; decide where logic goes; split a large file every feature touches; choose whether to throw or return an error; keep each user's or tenant's data isolated; model domain types; plan a refactor or breaking change; or review code against these rules. Covers vertical slices, functional core, pragmatic DDD, CQRS, module boundaries, naming, performance claims and versioning.

### `house-delivery`

Source: `skills/house-delivery/SKILL.md`

The house rules for delivering software, on any forge or cloud. Use for any task that writes or changes a CI or CD pipeline or build spec (.onedev-buildspec.yml, azure-pipelines.yml, .github/workflows); pins a shared template, action or job image; runs helm upgrade or another release to Kubernetes; changes Terraform, OpenTofu or Bicep; names cloud resources; sets up build agents or runners; gives a deploy job its credentials; sets up a ZAP scan; or adds an SBOM or provenance. Covers immutable pins, a release that ends at the last good state, plan before apply, one artefact promoted by digest, short-lived deploy credentials, a fresh agent per job, scan findings as tracked work, and scheduled shutdown of non-production. Not for a shell script's own form, service health and alerts, or driving odev.

### `house-dotnet`

Source: `skills/house-dotnet/SKILL.md`

The house rules for C# and .NET. Use when you write, review or set up C# code: ASP.NET Core minimal API endpoints, EF Core, Aspire, TUnit tests, analyzers, .editorconfig, NuGet packages and Directory.Packages.props, or a .NET solution; or when a C# change brings in repositories, Moq, xUnit, FluentAssertions, Task.WhenAll fan-out or a bare resilience handler. Covers ErrorOr with TypedResults, Vogen, slices with Scrutor decorators, named query filters, Testcontainers, the analyzer set, central package management and lock files. Not for language-neutral design questions.

### `house-otel`

Source: `skills/house-otel/SKILL.md`

The house rules for running services, in every language. Use for any task that adds or changes a background, nightly or scheduled job; a call to an upstream or LLM API (retries, timeouts, 429s, circuit breakers); health checks; an alert, dashboard or monitor, including one that never fires; spans, metrics, logs or tracing; or a message another service must receive reliably (outbox, workflow engine). Covers failure classes and retry ownership, deadlines, idempotent writes, the durability ladder, the use-case span and histogram, OpenTelemetry setup, /alive and /health.

### `house-python`

Source: `skills/house-python/SKILL.md`

The house rules for Python. Use for any Python task in our repos: writing, changing or reviewing .py code, SQLAlchemy models, FastAPI or FastMCP endpoints, click commands, pytest tests, async or concurrent calls, error handling, or pyproject.toml, uv.lock, ruff and pyright changes such as adding or bumping a dependency. Covers outcome unions with match, frozen dataclasses, Protocol seams with hand fakes, deadlines and task groups, SQLAlchemy 2.0, pytest-socket and anyio, uv with a 7-day cooldown. Not for language-neutral design questions.

### `house-shell`

Source: `skills/house-shell/SKILL.md`

The house rules for Bash and PowerShell scripts. Use whenever a shell script is involved: writing, fixing or reviewing a .sh, .bash, .ps1 or bin/ script (including reviewing one before it runs on a server, or one that carried on after a command failed); install, deploy, backup, clean-up and migration scripts that delete, rename or move things across repos or hosts; Claude Code hooks; CI steps; a .sh and .ps1 pair; shellcheck, shfmt or PSScriptAnalyzer setup. Covers the strict-mode header, exit codes, dry runs by default, twins in step, temp files and traps, killing by PID, stub tests, lint and format.

### `house-typescript`

Source: `skills/house-typescript/SKILL.md`

The house rules for TypeScript, React and Next.js. Use for any TypeScript or JavaScript task in our repos: writing, changing or reviewing .ts or .tsx code, React components, Next.js pages, Server Actions or route handlers, data fetching with TanStack Query, calling our API, vitest or Playwright tests, or package.json, tsconfig and ESLint changes such as adding or bumping a dependency. Covers openapi-fetch and zod at the boundary, one Result and ApiError, branded values, React 19, MSW, the 7-day release age and install-script approval. Not for language-neutral design questions.

### `standards-feedback`

Source: `skills/standards-feedback/SKILL.md`

Turn review findings into changes to the house standards. Use after reviewing a PR or at the end of a slice, when a finding might be a gap in the rules and not only in the code: 'what should change in the standards', 'should this go in the house rules', 'feed this back', 'is this project specific or a standards problem', 'the agent did X again'. For each finding it decides where the fix belongs (the repo, a house skill, or a guard in the project template) and opens the PRs. Works with or without pstack. Not for reviewing the code itself (house-architecture, 'When you review'), or for lessons about how an agent session ran (pstack:reflect).
