hyoka

How to write comprehensive architectural proposals that drive alignment before code is written

How the eval engine works: generate → grade → review → report

Record final outcomes to history.md, not intermediate requests or reversed decisions

Tone enforcement patterns for external-facing community responses

Team initialization flow (Phase 1 proposal + Phase 2 creation)

Core conventions and patterns for this codebase

Expert guidance for authoring and maintaining .prompt.md evaluation files for hyoka. Covers frontmatter formats, file structure, filtering, and best practices.

Team-wide charter and history optimization through skill extraction

Authoritative best practices for generating Rust code against the Azure SDK for Rust — covers official `azure_*` crate selection (and banned legacy crates), `azure_identity` token credentials, `Result`/`?` error handling with `ErrorKind` matching, `Pager<T>` pagination via `TryStreamExt`, `Poller<T>` long-running operations, and `#[tokio::main]` runtime setup. Use this skill whenever you generate or modify Rust code (`.rs`, `Cargo.toml`) that talks to any Azure service — even single-file snippets, quick examples, or prompts that only mention "Azure" plus Rust/cargo without naming a specific service. Apply it before writing the first line so crate names, auth, and pagination are correct on the first pass.

Authoritative checks for reviewing Java code that targets the Azure SDK for Java — verifies modern `com.azure:*` artifacts (rejecting legacy `com.microsoft.azure:*`), `com.azure.identity` token-based auth, correct client construction, async/Reactor usage, and pom.xml/build.gradle hygiene. Use this skill whenever you are reviewing, grading, or critiquing Java code that touches Azure services — even short snippets, single-file samples, or prompts that only mention "Azure" and Java/Maven/Gradle without naming the SDK. Apply it before scoring or commenting so legacy-SDK and anti-pattern violations are caught on the first pass.

Authoritative rules for writing markdown headings (H1/H2/H3) with correct hierarchy, exactly one H1 per document, no skipped levels, and blank lines around each heading. Use this skill whenever you are about to write or edit a markdown (.md) file that contains any heading — even single-heading documents, even one-shot tasks, and even when the user only says things like "add a title", "level-1 heading", "section", or "# Hello". Apply it before writing the file so the heading structure is correct on the first pass.

Authoritative rules for formatting markdown bullet lists and numbered lists, including marker choice, spacing, blank lines around lists, and nested-item indentation. Use this skill whenever you are about to write or edit a markdown (.md) file that contains any list — including short three-item lists, single-bullet lists, and tasks where the user says things like "a bullet list", "a list of items", "three items", or "numbered steps". Apply it before writing the file so the list formatting is correct on the first pass.

Stable id-based voting for multi-model review panels

Verify site/dist/ is up to date before merging site changes (go:embed reads directly from committed bundle).

Validating in-repo prompt/config/criteria example files against the live schemas, including paths the `hyoka validate` CLI does not cover.

Pattern for building flexible N-way comparison UIs over data with arbitrary attributes — groups defined by composable filters, configurable visualizations, localStorage persistence.

Automate browser interactions, test web pages and work with Playwright tests via the playwright-cli command.

How to introspect what tools/skills/MCP servers actually loaded into a Copilot SDK session, and how to compare against the declared config to detect silent load failures.

How to author SKILL.md files that conform to the agentskills.io spec and actually trigger when relevant. Use when creating, reviewing, or fixing skills under skills/, .squad/skills/, or anywhere a SKILL.md is being authored — especially when a skill "exists but never gets used" (almost always a description problem).

{what this skill teaches agents}

Determines which LLM model to use for each agent spawn based on 4-layer hierarchy

Identifies Azure SDK packages in generated code and checks whether they are the latest available versions. Use during code review to catch outdated dependencies.

Determines which LLM model to use for each agent spawn based on 4-layer hierarchy

Context hygiene — compress, prune, archive .squad/ state

User-level personal agents that travel across projects via Ghost Protocol

Cobra commands in cmd/ package, flag conventions

Return errors up, no log-and-return, %w wrapping

slog usage, no third-party logging

Table-driven tests, -race flag, stub patterns

Cross-platform path handling and command patterns

Reads generated Azure SDK code files and adds inline review comments without changing any actual code. Use during code review to annotate quality issues, best practices, and suggestions.

Sets up build environments for generated Azure SDK code samples and attempts to compile/build without modifying generated files. Use during review to verify code compiles correctly.