{"batch": "read1-093", "plugins": [{"pr": 0, "name": "done-bell", "does": "A Desktop/CLI notifier that chimes and shows a busy/done/needs-you status chip when a Hermes turn ends or needs approval.", "useful": 3, "reliable": 4, "polish": 3, "creative": 2, "beyond": 3, "best": "Tests (96 lines) cover the hooks and the hooks are written to never raise, and it needs no API keys or secrets at all.", "worst": "It is a simple notification utility with modest scope; the idea (turn-end notification) is common and the Desktop pane is minimal.", "standout": false, "blocked": null, "total": 18}, {"pr": 0, "name": "homeassistant", "does": "The official gateway platform adapter and toolset for Home Assistant: forwards state-change events to the agent and lets the model query/control devices over Home Assistant's REST API.", "useful": 5, "reliable": 5, "polish": 5, "creative": 2, "beyond": 5, "best": "Official, mature plugin with six dedicated test files, a long-lived-token model read via Hermes's own env/secret handling, and explicit entity allow/deny-list safety controls.", "worst": "Home-automation integration is a familiar pattern (it was previously built directly into Hermes core), so it scores low on novelty even though execution is excellent.", "standout": false, "blocked": null, "total": 27}, {"pr": 151968, "name": "tickover", "does": "Gives Hermes 'ears': local offline audio tools to verify an action happened by sound, enroll and monitor a machine's healthy sound for drift, teach custom classifiers, and run scheduled audio watch routines with notifications.", "useful": 4, "reliable": 5, "polish": 4, "creative": 5, "beyond": 5, "best": "Bundles real ONNX audio-ML inference plus from-scratch SSRF/DNS-rebinding/RTSP-injection hardening, backed by an ~860-line adversarial security test suite and published benchmark numbers in README.md.", "worst": "README references screenshot images (docs/screenshots/*.png) that were not confirmed to exist in the checked-out tree.", "standout": true, "blocked": null, "total": 28}, {"pr": 153595, "name": "tinfoil-sandbox", "does": "A terminal/execution backend that reroutes Hermes's terminal, code-execution and file tools to run inside a remote, attestation-verified confidential-computing enclave over an SSH tunnel.", "useful": 3, "reliable": 5, "polish": 4, "creative": 4, "beyond": 5, "best": "Reads its API key through Hermes's own secret_scope (not a plain file), explicitly strips admin/tunnel keys from any environment it passes along, and is tested with a stand-in CLI across many real failure scenarios.", "worst": "Niche audience (users who already run a Tinfoil confidential-computing container) limits how many Hermes users would install it day one.", "standout": true, "blocked": null, "total": 26}, {"pr": 149988, "name": "canary-trap", "does": "Injects a decoy credential into every turn's context and blocks/locks any tool call whose arguments leak that decoy (in plain, encoded, reversed or fragment form), acting as a prompt-injection tripwire.", "useful": 3, "reliable": 5, "polish": 4, "creative": 4, "beyond": 4, "best": "Compact, well-tested (11 parametrized disguise-detection cases plus concurrency and manifest-consistency tests), makes no network calls, and candidly lists what it does NOT catch in a 'Limits' section.", "worst": "Only catches exfiltration that touches the planted decoy string; a leak that never references it passes through undetected, as the README itself admits.", "standout": false, "blocked": null, "total": 24}, {"pr": 152366, "name": "ctx_handover", "does": "Moves a live Hermes chat session and its project workspace (including uncommitted git state) between machines a user owns, via direct pull/push or an offline bundle file, with session locking so it isn't continued twice.", "useful": 3, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "best": "Implements a real signed Ed25519 peer-pairing protocol with an explicit, documented environment allowlist for subprocess git calls so provider keys never reach the subprocess, backed by hundreds of unit/integration/e2e test files.", "worst": "No Desktop UI of its own; everything is driven through a CLI slash command, so the polish is entirely in documentation rather than a visual surface.", "standout": false, "blocked": null, "total": 25}, {"pr": 152899, "name": "project-command-center", "does": "A Hermes Desktop dashboard tab that aggregates Projects with live sessions, git status, GitHub issues/PRs and a commit graph, lets a user open a pre-filled session from an issue, and auto-advances a card to Review when the agent opens a PR.", "useful": 4, "reliable": 5, "polish": 5, "creative": 3, "beyond": 5, "best": "Ships a real Desktop UI tab backed by a FastAPI route and a SQLite store, with SSRF-guarded fetches, fixed-argv git/gh subprocess calls with environment allowlisting, and screenshots/demo video that were confirmed to exist in the repo.", "worst": "A minor version mismatch between docs/catalog-entry.yaml (0.4.1) and plugin.yaml (0.4.2) is a small, cosmetic documentation slip.", "standout": true, "blocked": null, "total": 27}, {"pr": 152646, "name": "mcp-pin", "does": "Fingerprints every MCP tool definition the first time it's seen and hides/blocks any tool whose description or parameters later change (a rug-pull/tool-poisoning attack) until an authorized owner reviews and approves the diff.", "useful": 4, "reliable": 5, "polish": 4, "creative": 5, "beyond": 4, "best": "Extraordinarily thorough test suite (482 lines, plus Hermes-integration and opt-in real-CLI end-to-end tests) covering concurrent writers, fail-closed behavior, and approval codes cryptographically bound to the exact reviewed diff.", "worst": "No screenshots or visual assets; the README itself admits editing the pin-store file directly (rather than a definition) is a gap the plugin does not catch.", "standout": true, "blocked": null, "total": 26}, {"pr": 152336, "name": "afriradar", "does": "Gives Hermes Nigerian and Kenyan mutual-fund data (from SEC Nigeria and CMA Kenya public reports) to find, rank, compare and project returns for funds, plus a watchlist with weekly change alerts.", "useful": 3, "reliable": 4, "polish": 4, "creative": 3, "beyond": 3, "best": "Does all financial arithmetic (projections, basis-point moves) in code rather than letting the model estimate it, is covered by a 481-line test file, and needs no API key since it reads public regulator data.", "worst": "Narrow geographic niche (only Nigerian and Kenyan collective-investment data) limits how many Hermes users would find it useful.", "standout": false, "blocked": null, "total": 20}, {"pr": 153711, "name": "relic-gumi", "does": "A longitudinal AI-companion plugin that injects relational context into conversations, runs consented check-ins and proactive outreach over Telegram/Desktop, and builds a psychometric baseline of the user, stored entirely on-disk.", "useful": 3, "reliable": 2, "polish": 5, "creative": 4, "beyond": 4, "best": "Exceptionally thorough documentation (850+ line spec-like README, arXiv paper link, onboarding/porting docs) and ~60 test files covering delivery lifecycle and consent guards.", "worst": "Reads its Telegram bot token and allowed-user-ID from a plaintext .env file in the plugin's own profile directory instead of through Hermes's secret store, which is exactly the kind of practice the catalog reviewer sends plugins back for, even though the README discloses it candidly.", "standout": false, "blocked": null, "total": 22}, {"pr": 152312, "name": "flowlight", "does": "A read-only wallet-watching plugin that tracks public NEAR/EVM/Solana addresses, summarizes inflows/outflows with USD pricing, and renders the flows as an animated visualization in a Desktop pane, dashboard tab and standalone page.", "useful": 3, "reliable": 5, "polish": 5, "creative": 4, "beyond": 5, "best": "Explicitly refuses seed phrases and signing keys in code (tested), every tool handler always returns read_only: true even on error, and it adds a genuine new Desktop UI surface (pane, route, sidebar row, status chip) via the plugin SDK.", "worst": "Narrow audience of users who want to passively watch public on-chain addresses, rather than a broad day-one install.", "standout": true, "blocked": null, "total": 27}, {"pr": 150655, "name": "mindhall", "does": "A shared, signed 'mind' (work, dates, notes, change log) that Hermes and other AI apps (ChatGPT, Claude, Cursor) can all read and write via an included MCP server, injected into every Hermes turn as context.", "useful": 4, "reliable": 4, "polish": 3, "creative": 4, "beyond": 4, "best": "Correctly reads its settings through Hermes's profile-aware secret_scope rather than opening .env files directly, and is explicitly designed to never break a turn if the store can't be read.", "worst": "The optional external MCP server (for connecting ChatGPT/Claude) protects access with a secret-bearing URL rather than real authentication, and has only modest test coverage (engine + settings, 237 lines) relative to its cross-app scope.", "standout": false, "blocked": null, "total": 23}]}