Skip to content

AGENTS.md

Latest commit

 

History

History
40 lines (28 loc) · 1.7 KB

AGENTS.md

File metadata and controls

40 lines (28 loc) · 1.7 KB
Raw

Agent project rules

This repo is a template for Lego-style agent systems. Read README.md and docs/architecture.md first.

Language

Write comments and documentation in English only, unless the user explicitly asks for another language.

This applies to:

  • Code comments
  • README and docs in this repo
  • New skills, scripts, and config notes you add here
  • Commit messages and PR descriptions for this template

User-facing agent replies may follow the user's language (see workspace/INSTRUCTIONS.md). Skill submodules may keep their upstream language - do not rewrite them in place.

Layout

  • workspace/ - shared agent workspace (INSTRUCTIONS.md, .opencode/opencode.json)
  • config.yaml - Coddy HTTP stack
  • skills_cache/ - git submodules at repo root; mounted as /skills_cache in Docker (not under /workspace)
  • workspace/.opencode/skills/ and workspace/.coddy/skills/ - symlinks ../../../skills_cache/<path>
  • atlassian_data/ - mcp-atlassian persistent cache (gitignored)
  • docker-compose.acpbox.yaml - OpenCode + ACPBox (+ optional openwebui profile)
  • docker-compose.coddy.yaml - Coddy HTTP

When implementing features in a fork

  1. Prefer MCP for external systems (Confluence, Jira, ...).
  2. Prefer CLI for one-off heavy jobs; document in workspace/INSTRUCTIONS.md.
  3. Encode repeatable flows as skills; symlink into workspace/.opencode/skills/<name> and workspace/.coddy/skills/<name>.
  4. Keep the stack small - add a compose service only when MCP or agent requires it.

NeuralDeep

  • LLM - https://api.neuraldeep.ru/v1
  • Skills catalog - https://neuraldeep.ru/skills
  • MCP catalog - https://neuraldeep.ru/mcp
  • CLI catalog - https://neuraldeep.ru/tools