Awesome Hermes Plugins › Hooks & Observability

0ldev/refusal

⭐ 34 Python added to this list on 2026-11-02 repository created 2026-10-23

refusal is a Hermes Agent plugin that catches model refusals, policy citations, and soft deflections before they become dead-end replies, then automatically retries the turn on a fallback model. It targets situations where a primary model declines legitimate work, such as security research, penetration-testing help, or malware analysis, and gives the agent a chance to continue on the next model in a configured fallback chain instead of stopping the conversation. The plugin hooks into Hermes through public APIs only: the llm_execution and llm_request middlewares plus four observer hooks (on_session_finalize, on_session_reset, on_session_delete, post_compaction). It never patches Hermes core. Detection is local, rule-based, and dependency-free: no network calls, and even tens of thousands of blank lines classify in a fraction of a second. Tool-calling responses are passed through untouched; only text responses are checked against refusal, deflection, and partial-answer patterns, while fenced code blocks are ignored to avoid false positives on code samples that merely discuss refusals. Refusal detection covers direct refusals, policy invocations, softer phrasings, and deflections to an educational alternative. It also recognizes provider-signalled refusals that carry no text, such as a stop_reason of refusal or a content_filter finish reason. When a refusal is detected, the plugin sets the finish reason accordingly and retries on the next fallback provider. Configuration lives under Hermes's plugin settings and includes an optional retry preprompt appended to the system prompt after a refusal is detected, with controls for how long a session stays armed, which models receive the preprompt, and how an escalated refusal is handled. The plugin requires Hermes >= 0.22.3, Python >= 3.10, and at least one fallback provider; without one, a detected refusal simply ends the turn rather than retrying.

https://github.com/0ldev/refusal

hermes-pluginrefusal-detectionfallback-modelmiddlewarehooks

Also in Hooks & Observability

rullerzhou-afk/clawd-on-desk

Clawd on Desk is a pixel desktop pet that reacts to your AI coding agent's activity in real time, offering visual feedback for various states and supporting numerous AI agents including Hermes.

lennney/stop-that-shit

A cross-harness task-boundary guard that ships a Hermes Agent plugin, blocking an AI coding agent from mutating files, dependencies or hashes outside the mode the user asked for.

alexgreensh/token-optimizer

Native Hermes plugin adding context-fill nudges, per-turn token/cost tracking, session rollups, and a shared dashboard, ported from the broader multi-host Token Optimizer project.

8bit64k/cronalytics

Cronalytics is a Hermes Agent plugin providing detailed analytics and observability for scheduled agentic automations, tracking costs, tokens, and performance.

rarf/hermes-quota-plugin

A Hermes Desktop plugin showing live model-provider quota as a status-bar chip and sidebar pane, fed by a local cache that a Python backend refreshes across eight providers.

kaishi00/hermes-conduit-notifier

Hermes plugin that observes lifecycle hooks and pushes approval, clarification, completion and failure notifications to the companion Hermes Conduit iOS app via a self-hostable relay.

Adolanium/hermes-resetwatch

Hermes Desktop plugin showing live remaining-quota bars and reset timers for Claude, Codex, Cursor, and a dozen other AI provider plans on one page inside the app.

briancaffey/hermes-otel

An OpenTelemetry plugin for Hermes Agent that exports LLM tool calls, model invocations, and API requests as OTel spans to various tracing backends like Phoenix, Langfuse, SigNoz, and Jaeger.