π€ AI Summary
If you run Claude Code alongside Codex and Cursor you can end up with duplicated, drifting instructions: you edit CLAUDE.md, switch tools, and the others read AGENTS.md instead. The practical fix is to pick one canonical file and make the others point to it. Three simple patterns work: 1) reference a file inline (e.g., put @AGENTS.md inside CLAUDE.md β Claude Code will read the referenced file inline), 2) use a symbolic link (ln -s AGENTS.md CLAUDE.md or ln -sf AGENTS.md CLAUDE.md to force), or 3) use a pointer stub (have CLAUDE.md say βREAD AGENTS.md FIRST!!!β and keep the canonical content in AGENTS.md). Symlinks make both filenames identical to tools; references let you include shared content while retaining per-agent notes.
This is a small operational change but significant for prompt engineering and reproducibility: a single source of truth prevents subtle divergent behavior across agents, reduces maintenance overhead, and makes updates atomic. Choose based on needs β symlinks for exact sameness, references or pointers for shared base content plus agent-specific addenda β and apply a familiar software-engineering principle (one source of truth) to your agent configs to avoid repeated, error-prone edits.
Loading comments...
login to comment
loading comments...
no comments yet