Agent instructions
Hooks report the lifecycle, but only the model can decide to search Slaunt Context before working or log a decision afterwards. The agent instructions tell it when to. slaunt-install appends them to the files Claude Code and Codex read at startup.
npx slaunt-installIt only appends, between start and end markers, so rerunning it is safe and your existing text is untouched. It refuses symlinks, non-UTF-8 and oversized files rather than risk corrupting them.
Where it writes
| Agent | Default (user scope) | Override |
|---|---|---|
| Codex | ~/.codex/AGENTS.override.md if it exists, else ~/.codex/AGENTS.md | CODEX_HOME |
| Claude Code | ~/.claude/CLAUDE.md | CLAUDE_CONFIG_DIR |
With --scope project it writes to the Git root instead, which is useful if you want the instructions committed for your team. Codex uses AGENTS.override.md or AGENTS.md. Claude Code uses CLAUDE.md, then .claude/CLAUDE.md, else creates CLAUDE.md.
| Flag | |
|---|---|
--scope user|project | Default user |
--only codex|claude|both | Default both |
--instructions <path> | Use your own Markdown instead |
--codex-path, --claude-path | Write to a specific file |
--project-dir <path> | Repo root when not running inside it |
--dry-run, --paths | Preview without writing |
For Cursor and other clients, paste the raw text from /api/agent-instructions into the client's rules, e.g. .cursor/rules/slaunt.mdc.