{ "batch": "read1-112", "plugins": [ { "pr": 0, "name": "done-bell", "does": "Rings an OS notification/chime and shows a busy/done/needs-you status chip when a Hermes turn finishes or an approval is waiting.", "useful": 3, "reliable": 3, "polish": 2, "creative": 2, "beyond": 3, "total": 16, "best": "Small, clean cross-platform hooks plus a Desktop pane and status chip that solve the real walk-away annoyance without any keys or stored command text.", "worst": "Tiny scope and thin README with no screenshots, a leftover personal dev path, and a Desktop pane that must be copied in by hand.", "standout": false, "blocked": null }, { "pr": 0, "name": "homeassistant", "does": "Official Home Assistant gateway platform (WebSocket state_changed events, notify delivery) plus four REST tools to query and control devices.", "useful": 4, "reliable": 5, "polish": 5, "creative": 2, "beyond": 4, "total": 24, "best": "Hardened and heavily tested (about 176 tests): token from Hermes's secret scope, blocklist of code-executing services, entity allow/deny lists.", "worst": "Not new work: it is Nous's own Home Assistant integration moved out of core, so it adds little creatively.", "standout": false, "blocked": null }, { "pr": 152830, "name": "steward", "does": "Collects each department's weekly update in its own group chat, nudges only the teams that have not replied, builds a founder pack with a voice brief and publishes a staff version only after founder approval.", "useful": 4, "reliable": 4, "polish": 4, "creative": 4, "beyond": 4, "total": 24, "best": "A full multi-chat workflow on Hermes cron, gateway and TTS, with founder-only publishing, media-cache-confined file reads and 114 tests, including security suites.", "worst": "Needs gateway setup across many group chats, has no screenshots or demo, and writes helper scripts into HERMES_HOME/scripts.", "standout": false, "blocked": null }, { "pr": 153740, "name": "hermes-arena", "does": "A pixel-art platform fighter that runs as a Hermes Desktop page, against a CPU, a local second player, or an online opponent over WebRTC.", "useful": 2, "reliable": 4, "polish": 5, "creative": 5, "beyond": 3, "total": 22, "best": "A remarkably complete game drawn entirely in code, with netplay, gamepad support, Shadow DOM isolation and a clean destroy() lifecycle.", "worst": "It has nothing to do with the agent: Hermes Desktop is only the host page, and the 4,300-line plugin.js is a generated build of game/pixel-hermes.html.", "standout": false, "blocked": null }, { "pr": 152319, "name": "homework-coach", "does": "Reads photos of a child's homework with a vision model and builds verified, adaptive daily practice PDFs plus a separate parent answer key that adjust to parent feedback.", "useful": 3, "reliable": 4, "polish": 5, "creative": 4, "beyond": 4, "total": 24, "best": "Answers are recomputed independently and the PDFs are checked before delivery, with 7 tools, a Desktop page whose source ships, a cron blueprint and about 190 tests.", "worst": "A narrow audience (elementary math), and the author admits the 0.2 Desktop panel has not been tested live.", "standout": false, "blocked": null }, { "pr": 153828, "name": "dreaming", "does": "A memory provider that records transcripts and notes and runs an offline dream pass that checks notes against transcripts, rewrites a linked wiki and keeps a capped index injected each session.", "useful": 4, "reliable": 3, "polish": 4, "creative": 4, "beyond": 5, "total": 25, "best": "It replaces Hermes's memory layer with consolidation that runs in the background, with real guard rails (path limits, size caps, backups, lock file) and an FTS5 store.", "worst": "Relies heavily on the LLM dream pass, has no UI, and treats tool output as evidence, so fetched web text could be persisted into memory; its tests sit outside the plugin folder.", "standout": true, "blocked": null }, { "pr": 150965, "name": "grounded-recall", "does": "A pre_llm_call hook that searches the user's notes or vault, secret-manager item titles and recent sessions, then injects a bounded evidence block labelled by freshness.", "useful": 4, "reliable": 4, "polish": 3, "creative": 3, "beyond": 4, "total": 22, "best": "Retrieval is deterministic and bounded (1.5 s cap) and audited every turn, with 77 security-focused tests and a minimal environment for the secret-title CLI.", "worst": "Mostly lexical keyword injection; the demo eval baseline is stacked (plugin-off with no tools), and docs have TODOs, a wrong config key and a missing demo.gif.", "standout": false, "blocked": null }, { "pr": 154015, "name": "sauce-vids", "does": "Records a headless Chromium session of an allow-listed HTTPS site with Playwright and encodes it to MP4 with ffmpeg, with ffprobe metadata.", "useful": 2, "reliable": 2, "polish": 2, "creative": 3, "beyond": 3, "total": 15, "best": "Real browser capture and ffmpeg rendering, with a host allow-list that checks DNS answers are public addresses.", "worst": "Thin code with 3 shallow tests; hover/click actions call Playwright's Locator.first as a method (tools.py:114,120), which looks broken; Chromium runs with --no-sandbox, and there are no demo assets.", "standout": false, "blocked": null }, { "pr": 153796, "name": "thumbs-fix", "does": "A thumbs-down on a Hermes Telegram reply finds the skill behind it, asks what was wrong, has the user's model draft a one-place edit, and patches the skill only after approval.", "useful": 4, "reliable": 5, "polish": 5, "creative": 4, "beyond": 4, "total": 26, "best": "A closed, human-approved loop from feedback to skill repair, with nonce-fenced prompts, stale-skill checks, backups, undo and about 2,500 lines of tests plus CI.", "worst": "Works only in Telegram private chats, cannot repair a skill loaded in an earlier turn, and undo can bypass skill_manage with a direct file write.", "standout": true, "blocked": null }, { "pr": 151135, "name": "wake-me-when", "does": "Watches live streams (YouTube, Twitch, HLS, RTSP, files), transcribes them locally with faster-whisper, and pings the user with quote, link, clip and catch-up when a described condition happens.", "useful": 4, "reliable": 4, "polish": 5, "creative": 5, "beyond": 5, "total": 28, "best": "A cascade of free cue matching, then an LLM judge that cites transcript rows, then debounce, with trimmed subprocess env, a private-address guard and bundled demo streams.", "worst": "Heavy external dependencies (ffmpeg, yt-dlp, streamlink, a self-installed Whisper extra), and watches die when the gateway restarts.", "standout": true, "blocked": null }, { "pr": 153755, "name": "chartrunner", "does": "Read-only tools for the ChartRunner Solana trading app: radar, cup standings and buy verdicts, Merkle memo checks, and review links the user signs in their own wallet.", "useful": 2, "reliable": 3, "polish": 2, "creative": 2, "beyond": 3, "total": 15, "best": "Careful read-only design (GET only, no redirects, 1 MB cap, trades only as review links) with Merkle checks tested against upstream vectors.", "worst": "A thin wrapper around one niche app's API with a 35-line README, no demo, and a CHARTRUNNER_ORIGIN override that can point it at any host.", "standout": false, "blocked": null }, { "pr": 153000, "name": "stoa", "does": "A lightweight CRM run by talking to Hermes: market maps, a YAML-playbook SQLite pipeline, review-before-apply lead intake, Slack button cards, a dashboard tab and a Desktop pane.", "useful": 4, "reliable": 3, "polish": 5, "creative": 4, "beyond": 4, "total": 24, "best": "Button clicks move the pipeline at once with no model involved, keys come from Hermes's secret scope, and the same data shows in Slack, the dashboard and Desktop with 11 screenshots and seeded demos.", "worst": "A very broad surface tested in one file, and gtm_export writes a CSV to any path the agent supplies (desk.py:776), which can overwrite user files.", "standout": false, "blocked": null } ] }