Feature request: official HOL Guard before_tool_call security example

[kantorcodes]

Proposal

Add a small official example/docs integration showing how to use HOL Guard at fast-agent's existing ToolRunnerHooks.before_tool_call boundary.

ToolRunnerHooks already exposes before_tool_call immediately before tools execute, and the hook is intentionally low-level/mutation-friendly. That gives fast-agent a concrete place to demonstrate a local security policy decision without adding a new generic hook or HOL Guard dependency to core.

Suggested example behavior

• install HOL Guard separately; Guard Cloud is not required for the local path;
• inspect the pending tool call name + structured arguments from the tool-runner context before execution;
• invoke HOL Guard locally with a bounded timeout/payload;
• explicit allow -> preserve the pending call and let it execute exactly once;
• deny -> remove/reject the pending call before execution;
• review-required -> do not execute until explicitly resolved;
• unavailable Guard, timeout, malformed output, or unknown decision -> fail closed for the protected call;
• avoid echoing raw tool arguments/secrets in error diagnostics.

Verification

A focused example/test should prove:

1. allow -> protected tool executes exactly once;
2. deny -> protected tool executes zero times;
3. review/error/timeout/malformed result -> zero executions by default;
4. no HOL Guard hook configured -> existing fast-agent behavior is unchanged.

I maintain HOL Guard and am disclosing that affiliation. I searched current issues/PRs for HOL Guard / hol-guard and found no existing proposal. If this fits the project, I can follow the repository's normal contribution/test conventions after maintainer direction and keep the change limited to the example/docs plus the smallest contract test.