hoard instructions
The hoard instructions command injects a canonical Hoard guidance block into agent instruction files.
Usage
hoard instructions [options]Options
| Option | Description |
|---|---|
--all | Update all targets (CLAUDE.md, AGENTS.md, OpenClaw SKILL.md) |
--claude | Update project Claude instructions |
--codex | Update project Codex instructions |
--openclaw | Update OpenClaw skill instructions |
--root <path> | Explicit project root for CLAUDE.md and AGENTS.md |
--dry-run | Show planned updates only |
--yes | Apply without confirmation |
What It Writes
Hoard inserts an idempotent marker block:
<!-- HOARD:START -->...<!-- HOARD:END -->If the block already exists, Hoard replaces it in place. If multiple blocks exist, Hoard collapses them into one canonical block.
Root Detection
For project files (CLAUDE.md, AGENTS.md), Hoard resolves root in this order:
--rootvalue (if provided)- Nearest parent containing
.git - Current directory only if it contains a supported project manifest
If no valid root is found, Hoard exits with an error and prints the exact remediation command.
Examples
Preview updates for Codex and Claude
hoard instructions --codex --claude --dry-runApply all targets without prompt
hoard instructions --all --yesApply for a specific repo path
hoard instructions --claude --codex --root /path/to/repoGolden Path Guidance
The injected block is intentionally short:
- Stable fact/preference:
memory_write - Durable note/document:
inbox_put - File-based changes not visible yet:
sync
Advanced lifecycle tools (memory_supersede, memory_retract, conflict/duplicate tools) are linked from the block.
See Also
hoard setup- Configure clients and optionally inject instructions- Claude Code Setup - Client-specific setup
- Codex Setup - Client-specific setup
- OpenClaw Setup - Client-specific setup