🛡️ Safety Hook

An oh-my-pi extension that demonstrates tool_call blocking. It intercepts bash tool calls and returns { block: true, reason: "..." } when the command contains rm -rf / with normal whitespace, preventing the tool from executing.

What it demonstrates

• pi.on("tool_call", ...) — pre-execution interception
• return { block: true, reason: "..." } — blocking contract
• Regex guard on bash input (/\brm\s+-rf\s+\//)

Install

cp -r . ~/.omp/agent/extensions/safety-hook

Restart omp. The hook is active for all sessions.

Or load once:

omp --extension ./safety-hook

How it works

flowchart TD
    %% Nodes
    Start([LLM calls bash tool]) --> RunHandlers[tool_call handlers run]
    
    RunHandlers --> Decision{command matches<br><code>/\brm\s+-rf\s+\//</code> ?}
    
    Decision -- yes --> Blocked[Block execution]
    Decision -- no --> Allowed[Execution continues normally]
    
    Blocked --> EndBlock([Return: <code>{ block: true, reason: '...' }</code> to LLM])
    Allowed --> Execute[Tool executes] --> EndNormal([Success])

    %% Styling
    style Decision fill:#f9f,stroke:#333,stroke-width:2px
    style Blocked fill:#ffcccb,stroke:#f00,stroke-width:2px
    style Allowed fill:#e1f5fe,stroke:#03a9f4,stroke-width:2px
    style Start fill:#e8f5e9,stroke:#4caf50,stroke-width:2px
    style EndBlock fill:#ffebee,stroke:#c62828,stroke-width:2px
    style EndNormal fill:#e8f5e9,stroke:#2e7d32,stroke-width:2px

The reason text is what the LLM receives as the tool error, so it can understand why the call was rejected and try a different approach.