Skip to content

CLAUDE.md

Latest commit

 

History

History
50 lines (41 loc) · 2.2 KB

CLAUDE.md

File metadata and controls

50 lines (41 loc) · 2.2 KB
Raw

Claude Code Project Guide for AetherSDR

Claude Code: read AGENTS.md for the full project guide. This file exists only because Claude Code auto-loads CLAUDE.md. Everything project-wide — architecture, conventions, the constitution, commit signing, branch protection rules, the must-knows — lives in AGENTS.md so every AI tool reads the same source of truth.

Why this is a thin file

AetherSDR's contribution surface is multi-agent in practice (six+ distinct AI tools touch the codebase). Each tool has its own well-known file at a different path:

Tool File
Claude Code CLAUDE.md (you are here)
OpenAI Codex CLI / spec-kit / Foundry AGENTS.md (the canonical)
GitHub Copilot .github/copilot-instructions.md
Gemini Code Assist GEMINI.md
Aider CONVENTIONS.md

All of those are pointers to AGENTS.md. The duplication-by-pointer pattern keeps each tool reading from its native discovery path without forcing a single canonical file to be copied N times.

Claude-Code-specific notes

These tidbits are about how Claude Code as a tool interacts with this codebase. Everything else is in AGENTS.md.

  • git ship alias — the project has a git ship shell alias documented in AGENTS.md that squashes local commits ahead of origin/main, creates a branch, pushes, opens a PR with auto-squash-merge enabled. Use it for batch shipping rather than one-commit-per-PR.
  • Skills directory — Claude Code's skill loader picks up skills from ~/.claude/skills/. The AetherClaude bot keeps its own copies under ~/build/aetherclaude/skills/. Those are the authoritative source for any agent-orchestration logic.
  • Plans directory — implementation plans live at ~/.claude/plans/. They are conversation-scoped scratch files, not project-canon; AGENTS.md, CONSTITUTION.md, and CONTRIBUTING.md are.
  • Memory directory — Claude Code's auto-memory lives at ~/.claude/projects/AetherSDR/memory/. Index is MEMORY.md. Use it for cross-session continuity; everything in it is operator-scoped (per Principle XIII), not project-canon.

That's everything Claude-Code-specific. Read AGENTS.md for the rest.