message-injector
par harukaon
Web & Frontend DevelopmentRésumé
OpenClaw plugin that prepends custom text to every user message before it reaches the agent.
Description
Ce que fait ce skill
A workspace extension for OpenClaw that prepends custom text to every user message before the agent processes it. Uses the `before_agent_start` hook to inject text at the Gateway level, making it impossible for the agent to bypass. Works across all channels.
Unlike system prompt edits that vary per session or client, this injects at the Gateway level so the instruction applies universally regardless of which channel or client is used.
Quand l'utiliser
- Force memory search before every agent reply
- Inject current project context into each conversation turn
- Enforce safety rules like confirming before file deletion
- Add persistent reminders without modifying each prompt manually
- Maintain consistent persona or language instructions across sessions
Exemple de workflow
User asks: 'Delete the old logs folder'
- 1
before_agent_start hook fires on incoming message
- 2
Plugin reads prependText from config: '[RULE: Always verify file paths before deletion. Never run rm -rf without confirmation.]'
- 3
prependContext prepends the rule to the user message
- 4
Agent receives combined message: rule + original user request
- 5
Agent processes message with injected constraint in context
Agent responds by asking for confirmation before proceeding with deletion
Prérequis
- OpenClaw Gateway installed and running
Installation
clawhub install harukaon/message-injectorExécute la commande CLI, ou colle l'instruction dans Claude Code, Cursor, Codex ou l'agent de code que tu utilises.