Skip to content

AGENTS.md

Latest commit

 

History

History
42 lines (29 loc) · 1.85 KB

AGENTS.md

File metadata and controls

42 lines (29 loc) · 1.85 KB

Nethercore AI Plugins (Agent Instructions)

This repo is a set of prompt packs (skills + task agents) for Nethercore development. The on-disk layout is compatible with the Claude Code plugin loader (.claude-plugin/), but the Markdown content should stay assistant-agnostic so it can be reused in other tools.

Repo Layout

  • nethercore/ — Console-agnostic Nethercore workflow (build/test/optimize/publish)
  • zx/ — ZX fantasy console specs + FFI reference

Each plugin pack typically contains:

  • .claude-plugin/plugin.json — plugin metadata
  • skills/<skill>/SKILL.md — concise “entrypoint” instructions
  • skills/<skill>/references/*.md — deeper reference material (tables, long lists)
  • agents/*.md — task-focused agents (e.g. build analysis, replay debugging)

Writing Guidelines

  • Keep SKILL.md files short; move long content into references/.
  • Prefer concrete steps, checklists, and small tables over prose.
  • Avoid naming specific assistants/vendors (e.g. “Claude”, “ChatGPT”) inside skills/ and agents/ bodies.
  • Prefer workspace-relative paths when referencing other repos (example: nethercore/include/zx.rs).
  • Don’t assume network access; instructions should work offline.

SKILL.md Standard

All skills must comply with the SKILL.md open standard:

  1. name — lowercase-hyphenated, matches parent directory name exactly
  2. description — explains WHAT + WHEN to use, max 1024 chars
  3. license — Apache-2.0 for this repo
  4. metadata.author — nethercore-systems
  5. metadata.version — semver string (e.g., "1.0.0")
  6. Body under 500 lines; move long content to references/
  7. Run python3 scripts/lint_repo.py --strict before committing

Validation

  • Run python3 scripts/lint_repo.py before sending changes.
  • CI runs the same script.