Auxiliary Artifacts
Three smaller artifact families complement the main set (agents, skills, rules, hooks, commands). Checks supply pass/fail criteria that agents load during review. Prompts are standalone, agent-agnostic instruction templates. GitHub Agents are simplified agent definitions tailored for GitHub Copilot and Codex.
Each family is small enough to share one reference page. Use the tables below to locate an entry, then open the canonical file under the listed path.
Checks
Review criteria definitions that agents (primarily hatch3r-reviewer) load to evaluate diffs against pass/fail criteria. Each check focuses on one concern and tags criteria as [CRITICAL] (blocking) or [RECOMMENDED] (advisory).
| Check | Purpose | Invoked by |
|---|---|---|
| accessibility | WCAG 2.1 AA criteria across semantic HTML, ARIA, keyboard navigation, contrast, screen reader support, and media. | Reviewer agents on UI/template diffs |
| code-quality | Standards compliance, complexity limits (50-line functions, 400-line files, cyclomatic <=10), maintainability, and error handling. | Reviewer agents on any source diff |
| performance | Bundle size, render path (LCP/CLS), memory cleanup, network/DB query patterns, runtime hot paths, and code splitting. | Reviewer agents on perf-sensitive diffs |
| security | Input validation, authn/authz, secrets handling, dependency safety, data exposure, cryptography, and error responses. | Reviewer agents on security-sensitive diffs |
| testing | Coverage requirements, determinism, isolation, integration tests, and regression-prevention rules. | Reviewer agents on any diff with code changes |
Canonical location: .agents/checks/{id}.md. Authoring guide: checks/README.md in the framework repo.
Prompts
Standalone prompt templates that produce a structured output without agent state. Suitable for one-shot invocations from any tool that accepts a prompt — Claude Code, Cursor chat, Copilot Chat, or a CLI/API call.
| Prompt | Purpose | Invoked by |
|---|---|---|
| hatch3r-bug-triage | Classifies a bug by severity (P0-P3), maps the affected area, and produces an investigation plan with reproduction steps, root-cause hypotheses, and a recommended fix. | Engineer pasting a bug report; CI workflow that posts triage on new issues |
| hatch3r-code-review | Reviews a diff across 8 dimensions (correctness, security, code quality, type safety, performance, testing, accessibility, error handling) and emits APPROVE / REQUEST CHANGES / COMMENT with a severity-tagged finding table. | Engineer requesting an ad-hoc review; PR-comment automation |
| hatch3r-pr-description | Generates a Conventional Commits-style PR title plus a body with Summary, Changes table, Test Plan, and Risk Assessment from the staged diff. | Engineer running it locally before opening a PR |
Canonical location: .agents/prompts/{id}.md. Each prompt ships with few-shot examples in the body.
GitHub Agents
Simplified agent definitions designed for GitHub Copilot / Codex, which do not load the full hatch3r agent runtime. They omit pipeline integration and constrain themselves to a single role with explicit boundaries.
| Agent | Purpose | Invoked by |
|---|---|---|
| hatch3r-docs-agent | Technical writer that reads source and updates specs, ADRs, and process docs under docs/ while preserving stable IDs. | Copilot/Codex when assigned a documentation task |
| hatch3r-lint-agent | Code-quality engineer that fixes ESLint, Prettier, TypeScript strict-mode, and naming issues without changing logic; verifies via lint then typecheck then test. | Copilot/Codex on a lint-fix or formatting task |
| hatch3r-security-agent | Security analyst that audits database rules, API endpoints, and data flows; writes rules tests for both allow and deny cases. | Copilot/Codex on a security audit or rules-test task |
| hatch3r-test-agent | QA engineer that writes deterministic unit, integration, and E2E tests with explicit edge cases and regression coverage. | Copilot/Codex on a test-authoring task |
Canonical location: .agents/github-agents/{id}.md. Each agent declares its file-structure assumptions, allowed commands, and "Always / Ask first / Never" boundary list.
Customization
Customization for these families follows the same pattern as the main artifacts: override behavior per-project via .hatch3r/checks/{id}.customize.yaml, .hatch3r/prompts/{id}.customize.yaml, or .hatch3r/github-agents/{id}.customize.yaml. See Customization.