Add hermes-jailbench to Benchmarks & Datasets

[roli-lpci]

What

Adds hermes-jailbench to the Benchmarks & Datasets section.

What it does

hermes-jailbench is a deterministic, single-turn regression benchmark for known-pattern jailbreaks. It runs a repeatable attack battery against an authorized LLM endpoint and classifies responses as refusal, partial, or compliance with an auditable keyword scorer.

Note: this is a known-pattern regression benchmark, not proof that a model or agent is safe and not coverage of novel or multi-turn attacks.

Checklist (per CONTRIBUTING.md)

• Scope: directly related to evaluating autonomous-agent security performance.
• Open source: MIT, public repository.
• Maintenance: current release v0.2.2, with the repository updated this week.
• Format: - **[Tool](URL)** - description. matching existing entries.
• Searched the live README at main commit f0252697b78f768233ef01a1f5f58ccd117223d1 and the live open-PR search; no existing hermes-jailbench entry or matching open duplicate PR (total_count=0).

Maintained by Hermes Labs; submitted by its maintainer.

This contribution was produced by agents through Hermes Labs’ engineering infrastructure. Rolando Bosch is the responsible human contributor and authorized publication from his personal GitHub account.