Skip to content

Latest commit

 

History

History
76 lines (59 loc) · 3.11 KB

AGENTS.md

File metadata and controls

76 lines (59 loc) · 3.11 KB

Agent Entry Point

This file is the stable entry point for agents working in this repository. It routes agents to the reusable pipeline, the project-specific profile, and the verification gates that keep changes reviewable.

First Reads

Read these in order before making non-trivial changes:

  1. docs/agent-pipeline.md for the reusable agent workflow.
  2. docs/agent-project-profile.md for this plugin's runtime, architecture, risk, and documentation rules.
  3. docs/agent-verification-matrix.md to select tests and manual checks.
  4. Task-specific docs when present:
    • Current plan and open work: ../PLAN.md (repo-wide, single plan file; MountLicense items live in its per-plugin section)
    • Operator-facing behavior: README.md
    • Architecture, API, compatibility, regression, or release docs under docs/

Skill Routing

Use reusable skills for process discipline, then load project facts from docs/agent-project-profile.md.

  • skills/code-optimization/SKILL.md: refactors, performance, reliability, or maintainability improvements.
  • skills/compatibility-update/SKILL.md: Java, Minecraft, server platform, dependency, shading, or optional integration compatibility work.
  • skills/docs-maintenance/SKILL.md: README, API, config, migration, release, or operator documentation updates.
  • skills/hci-design/SKILL.md: command UX, GUI flows, permission feedback, in-game copy, onboarding, or human-facing workflow changes.

Do not put project-specific details into skills. Skills should remain portable; project facts belong in docs/agent-project-profile.md.

Working Rules

  • Preserve existing behavior unless the task explicitly asks for a behavior change.
  • Keep changes small, scoped, and reversible.
  • Prefer existing service, manager, controller, lifecycle, and utility patterns.
  • Do not bypass documented project boundaries in docs/agent-project-profile.md.
  • Do not modify user-visible behavior without checking documentation and regression impact.
  • Do not change config, data schema, permissions, public API, economy, storage, or runtime scheduling without updating the relevant docs and tests.
  • Do not hardcode player-visible messages in Java when the project profile says localization is required.

Verification Commands

Use the verification matrix to select the smallest sufficient gate:

  • Narrow unit check: mvn "-Dtest=ClassNameTest" test
  • Standard check: mvn test
  • Release-grade check: mvn verify
  • Artifact check: mvn clean verify package

Record what was run and what remains unverified using docs/agent-evidence-template.md.

Definition of Done

A change is complete when:

  • The risk class is named and matched with appropriate verification.
  • Automated tests pass for the affected surface, or any gap is explicit.
  • Documentation is synchronized for user-visible, API, config, compatibility, migration, permission, storage, economy, or release-impacting changes.
  • Manual regression scope is identified when runtime behavior cannot be fully covered by unit tests.
  • The final response includes evidence, residual risk, and touched files at a level useful for review.