Skip to content

AGENTS.md

Latest commit

 

History

History
56 lines (49 loc) · 2.56 KB

AGENTS.md

File metadata and controls

56 lines (49 loc) · 2.56 KB
Raw

Agent Instructions

This repository keeps agent guidance in a tool-neutral form. Treat this file as the entry point for all coding agents, including Cursor, Codex, Claude Code, Copilot, and Aider.

Canonical Rule Source

  • Full project rules live in docs/agent-guidelines/.
  • Tool-specific files such as .cursor/rules/*.mdc, CLAUDE.md, and .github/copilot-instructions.md are adapters only.
  • When adding or changing persistent rules, update docs/agent-guidelines/ first, then update tool adapters only if their metadata or pointers need to change.
  • Before working in an area covered by the Rule Index, read the relevant guideline file and follow it for the task.
  • Prefer enforcing critical rules with tests, scripts, CI, or Makefile targets instead of relying only on agent memory.

Must-Follow Rules

  • Keep local HTTP API changes in sync with openapi/local-api.json; run go test ./internal/server after API doc changes.
  • csghub-lite must work on macOS, Linux, and Windows. Use platform-aware path, binary, library, process, and environment handling.
  • Do not commit or paste secrets. Load GitLab and OSS credentials only from local/secrets.env.
  • Follow the repository network split: external sites such as GitHub need source ~/.myshrc; internal GitLab access should be direct.
  • Use only http://localhost:11435 for local preview. Do not keep extra frontend preview servers running.
  • If frontend files change, build web, sync web/dist into internal/server/static, then restart the single backend preview.
  • Store all runtime files, including temporary files and subprocess temp output, under the csghub-lite storage root, defaulting to ~/.csghub-lite.
  • Keep commit, PR, and release notes concise and focused on concrete user-facing changes.
  • When syncing llama.cpp binaries: use official ggml-org/llama.cpp GitHub releases for upstream-published assets; build Ubuntu Linux CUDA packages only via scripts/llama-build/ (see docs/agent-guidelines/llama-cpp.md).

Rule Index

  • docs/agent-guidelines/api-swagger-sync.md
  • docs/agent-guidelines/app-installs.md
  • docs/agent-guidelines/ai-app-oss-mirror.md
  • docs/agent-guidelines/cross-platform.md
  • docs/agent-guidelines/frontend-i18n.md
  • docs/agent-guidelines/go-conventions.md
  • docs/agent-guidelines/llama-cpp.md
  • docs/agent-guidelines/local-preview.md
  • docs/agent-guidelines/model-source-routing.md
  • docs/agent-guidelines/network-and-secrets.md
  • docs/agent-guidelines/python.md
  • docs/agent-guidelines/release-notes.md
  • docs/agent-guidelines/storage.md