{ "batch": "read2-022", "plugins": [ { "pr": 0, "name": "done-bell", "does": "Notifies users when Hermes finishes a turn or needs approval, with a Desktop status chip, chime and CLI notifications.", "useful": 4, "reliable": 3, "polish": 2, "creative": 2, "beyond": 4, "total": 19, "best": "Uses actual turn and approval lifecycle signals rather than asking the model to remember to notify.", "worst": "Setup uses an author-specific Windows path and manual pane copying, while notification failures are silently swallowed.", "standout": false, "blocked": null }, { "pr": 0, "name": "homeassistant", "does": "Connects Hermes to Home Assistant device tools, state-change events and cron notification delivery.", "useful": 4, "reliable": 4, "polish": 4, "creative": 3, "beyond": 5, "total": 25, "best": "Combines an event-driven gateway platform with scoped credentials, entity policies and tested device-control tools.", "worst": "Entity restrictions are opt-in, so the default token-backed tools permit broad device control and rely on a service-domain blocklist rather than per-action human approval.", "standout": false, "blocked": null }, { "pr": 152421, "name": "sheets", "does": "Adds a local Desktop spreadsheet with typed model-filled columns, per-row research agents and reviewable chat proposals.", "useful": 5, "reliable": 4, "polish": 5, "creative": 4, "beyond": 5, "total": 28, "best": "Integrates native Hermes sessions, source traces, typed cells and reversible human-approved changes into a substantial Desktop workflow.", "worst": "Research sessions receive ordinary Hermes tools: potentially mutating tool calls are flagged after invocation rather than prevented by a plugin-enforced read-only allowlist.", "standout": true, "blocked": null }, { "pr": 152256, "name": "nvidia-switchyard", "does": "Routes Hermes model requests between local and cloud tiers through verification-gated routing, with provider presets and a Desktop monitor.", "useful": 5, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "total": 27, "best": "Changes Hermes model execution itself, with local-first presets, routing visibility, consent checks and extensive routing and recovery tests.", "worst": "Subscription routing still handles provider tokens through private runtime APIs under a documented reviewer exception, and verification can send the task to cloud even when the local model answers.", "standout": true, "blocked": null }, { "pr": 149685, "name": "hermes-teleoperator", "does": "Moves one Hermes conversation’s shell, file, skill and allowlisted MCP execution between SSH machines using a standalone Rust receiver.", "useful": 4, "reliable": 3, "polish": 4, "creative": 5, "beyond": 5, "total": 26, "best": "Makes machines usable without installing another agent while binding approvals to destinations and refusing silent local fallback.", "worst": "Its execution middleware bypasses other plugins’ pre_tool_call checks and agent guardrails, relying on private approval APIs to reconstruct only part of the normal guard path.", "standout": false, "blocked": null }, { "pr": 149272, "name": "no-cap", "does": "Checks final-answer claims against recorded tool evidence and adds receipts, contradiction flags, optional verification nudges and a Desktop honesty board.", "useful": 5, "reliable": 4, "polish": 5, "creative": 4, "beyond": 5, "total": 28, "best": "Builds deterministic receipts from actual tool hooks and catches skipped-test success claims without paying for another model.", "worst": "Receipts depend on parsed tool output rather than independent correctness; unsupported claims are hidden by default and an agent with shell access can fabricate evidence.", "standout": true, "blocked": null }, { "pr": 153821, "name": "learnloop", "does": "Adds structured tutoring and practice with graded questions, phase gates, answer cards and a persistent knowledge graph in Hermes.", "useful": 4, "reliable": 4, "polish": 5, "creative": 3, "beyond": 5, "total": 26, "best": "Enforces question and practice state in code and turns Hermes Desktop’s composer, transcript and sidebar into a coherent learning interface.", "worst": "Teaching quality and graph accuracy still depend on the teacher model; the separate graph extraction and auditing pipeline has not been ported.", "standout": false, "blocked": null }, { "pr": 153071, "name": "cron-budget", "does": "Counts model requests across profiles and holds scheduled chat-completion calls to preserve a configurable quota reserve for human chat.", "useful": 4, "reliable": 3, "polish": 3, "creative": 3, "beyond": 5, "total": 23, "best": "Enforces a real execution gate with observe-first setup, one notice per window and no model call for held jobs.", "worst": "Anthropic/Codex API modes and delegated cron children are not blocked, while quota decisions and call recording are separate transactions that can race.", "standout": false, "blocked": null }, { "pr": 151397, "name": "submission-desk", "does": "Tracks local application readiness by binding quoted requirements, portal fields, file hashes, machine checks and human reviews in a persistent ledger.", "useful": 3, "reliable": 5, "polish": 5, "creative": 4, "beyond": 4, "total": 25, "best": "Carefully separates agent proposals from human decisions and invalidates stale reviews when exact file bytes, rules or mappings change.", "worst": "Its careful workflow is narrow: document-format recognition exceeds its actual page/word checking support, and users must finish the portal work themselves.", "standout": false, "blocked": null }, { "pr": 152871, "name": "hermes-maps", "does": "Adds a shared Desktop map that Hermes can populate and navigate, with persistent pins, straight-line routes and user selections sent back to chat.", "useful": 4, "reliable": 4, "polish": 5, "creative": 3, "beyond": 5, "total": 26, "best": "Provides a real two-way spatial interface with durable scene state, conservative place matching and honest queued-versus-visible delivery results.", "worst": "Routes are straight-line sketches rather than navigable roads, and place lookup depends on a throttled public service with no uptime guarantee.", "standout": false, "blocked": null }, { "pr": 149220, "name": "hermes-notify", "does": "Connects Hermes release workspaces, progress, replies and exact one-use approvals to a phone through Notify and an operator-hosted Worker.", "useful": 3, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "total": 25, "best": "Uses Hermes’s native approval transport with exact action matching, bounded waits and durable replay prevention instead of treating ordinary phone replies as permission.", "worst": "The complete live product path remains unfinished in verification: Worker deployment, App sign-in, deep links and live progress are explicitly unverified.", "standout": false, "blocked": null }, { "pr": 152908, "name": "contract-necromancer", "does": "Locks user-approved acceptance tests and runs them at Hermes verification, blocking ordinary test edits and stamping final replies with failures or recheckable receipts.", "useful": 5, "reliable": 4, "polish": 5, "creative": 4, "beyond": 5, "total": 28, "best": "Moves completion checking outside the model and binds it to approved test hashes, with minimal child environments and extensive tampering and integration tests.", "worst": "The verification gate has a 25-second default budget and does not continue shell-only edits through pre_verify; mutable external test dependencies and same-user evasion remain outside its protection.", "standout": false, "blocked": null } ] }