{"batch": "read2-c/read2-012", "plugins": [
{"pr": 0, "name": "done-bell", "does": "Rings an OS notification/chime and a status chip when a Hermes turn finishes, with no API keys.", "useful": 3, "reliable": 3, "polish": 3, "creative": 1, "beyond": 3, "total": 21, "best": "Tiny, safe, no-secrets scope with a clean, honest README.", "worst": "Only one test file for the whole plugin, and the idea is common (presence/notification plugins exist widely).", "standout": false, "blocked": null},
{"pr": 0, "name": "homeassistant", "does": "Official Nous plugin providing a Home Assistant gateway platform and four tools (list/get/call entities and services) over HA's WebSocket/REST APIs.", "useful": 5, "reliable": 5, "polish": 5, "creative": 2, "beyond": 5, "total": 29, "best": "Official, test-covered (6 test files), replaces built-in core functionality with a clean plugin boundary and clear token-based auth via env/secret config.", "worst": "Low novelty: it reproduces functionality Hermes core already had, rather than inventing something new.", "standout": true, "blocked": null},
{"pr": 150810, "name": "tone-forge", "does": "Generates reasoned guitar-amp tone recipes (chain, EQ, pedal/mic settings) for AmpliTube 5 and drives assigned MIDI CCs, with optional reference-track analysis.", "useful": 3, "reliable": 4, "polish": 5, "creative": 4, "beyond": 4, "total": 28, "best": "162 passing tests, explicit SECURITY.md disclosing its only subprocess uses (fixed argv, no shell), and an honest 'what it does not do' section.", "worst": "Windows-only and niche to AmpliTube 5 users with MIDI-mapped gear, limiting day-one install appeal.", "standout": true, "blocked": null},
{"pr": 153832, "name": "hermes-screen", "does": "Lets Hermes browse a Jellyfin library by real watch state and cast/control playback on a Chromecast or Roku TV, plus YouTube-to-TV and a movie-night vote.", "useful": 4, "reliable": 3, "polish": 3, "creative": 3, "beyond": 5, "total": 26, "best": "Real device control (Chromecast token-gated LAN capability URLs, Jellyfin watch-state logic) that only a plugin, not a prompt, could do, with 23 test files covering cast/scene/youtube paths.", "worst": "Its own README labels it 'work in progress for the November 2026 hackathon', so parts of the scope are unfinished at the pinned commit.", "standout": false, "blocked": null},
{"pr": 152085, "name": "hermes-voice-lab", "does": "A WASM-compiled, locally-synthesized articulatory voice model in Hermes Desktop that Hermes can drive to speak or sing, with a visual portrait UI.", "useful": 2, "reliable": 3, "polish": 5, "creative": 5, "beyond": 4, "total": 27, "best": "Genuinely novel idea (a playable articulatory voice instrument driven by the agent, no cloud TTS, no network calls) with real demo videos.", "worst": "Niche appeal for day-one installs since it's a toy/performance instrument rather than a task the average Hermes user needs; only 6 test files for a complex DSP/UI engine.", "standout": true, "blocked": null},
{"pr": 153640, "name": "patient-research", "does": "Could not be determined; the catalog PR names a repository that no longer exists.", "useful": 0, "reliable": 0, "polish": 0, "creative": 0, "beyond": 0, "total": 0, "best": "N/A", "worst": "N/A", "standout": false, "blocked": "GitHub repository solo386/patient-research returns 404 Not Found, so the pinned commit and code cannot be read."},
{"pr": 153809, "name": "living-spaces", "does": "An open canvas inside Hermes Desktop where the agent builds and persists interactive dashboards, games and documents via ordinary chat, backed by a sandboxed revision/record store.", "useful": 4, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "total": 29, "best": "Extensive test coverage (43 test files touching sandboxing, canvas actions, lifecycle atomicity, and file-ref security) for a genuinely new Desktop capability.", "worst": "The README leans heavily on self-reported 'proven' claims (functional-check tallies like '9/9') rather than independently verifiable benchmarks, so some claims exceed what the code alone demonstrates.", "standout": true, "blocked": null},
{"pr": 151941, "name": "data-lens", "does": "Intercepts the default read_file call on large local CSV/JSON/JSONL files to build a typed structural map and adds bounded, exact query/grouping tools instead of returning a truncated raw page.", "useful": 4, "reliable": 5, "polish": 4, "creative": 3, "beyond": 4, "total": 28, "best": "Explicit secret/credential filename denylist (blocks reading files like 'secrets.json', 'client_secret.json', 'token.json') and a scrupulously honest README about scope limits.", "worst": "Solves a fairly narrow pain point (oversized file reads) that a well-designed prompt/skill could partially mitigate with chunked reads.", "standout": false, "blocked": null},
{"pr": 154048, "name": "triage", "does": "Replaces the full skills list in Hermes's system prompt with a fast local ranker that surfaces and loads only the skills relevant to the current message, shown as chips/pills in Desktop.", "useful": 4, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "total": 29, "best": "Solves a real scaling problem (context cost of thousands of skills) with a no-model local ranker (0.2s for 2,500 skills) and 20 test files covering auto-attach precision.", "worst": "Value depends heavily on having a large skill catalog already; for a user with a handful of skills the benefit is marginal.", "standout": true, "blocked": null},
{"pr": 150800, "name": "hypnos", "does": "Watches daytime approvals/denials/corrections and command outcomes, then trains a small personal LoRA overnight to absorb procedural habits and shrink the prompt.", "useful": 3, "reliable": 3, "polish": 5, "creative": 5, "beyond": 5, "total": 29, "best": "A genuinely novel mechanism (nightly LoRA consolidation that shrinks prompt tokens while keeping task accuracy, with a reported 24,636->5,256 token drop) backed by a sandboxed exam/replace cycle.", "worst": "Requires GPU training infrastructure and real usage history to pay off, so it is not a day-one install for most users and only 13 test files cover a complex training/safety pipeline.", "standout": true, "blocked": null},
{"pr": 153912, "name": "myfolks", "does": "Deterministic, encrypted-at-rest (AES-256-GCM) exact-match storage and recall of facts about people in the user's life, with no model call needed for save/recall.", "useful": 4, "reliable": 5, "polish": 4, "creative": 2, "beyond": 3, "total": 25, "best": "Reads secrets via agent.secret_scope.get_secret rather than files, encrypts storage at rest, and ships 34 test files plus a dedicated evidence pack of safety checks.", "worst": "The underlying idea (a structured people/contacts memory store) is a fairly direct application of a known pattern rather than something new for agents.", "standout": false, "blocked": null},
{"pr": 151964, "name": "takeover", "does": "Detects when a web page needs a human (login, 2FA, CAPTCHA, payment, OAuth consent) and hands the agent's live browser view to the user via a one-tap banner in WebUI/Desktop/dashboard, then hands control back.", "useful": 4, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "total": 29, "best": "Explicitly keeps payments and credential entry as the human's job (regex-classified nudge logic routes payment/credential/CAPTCHA forms to request_takeover rather than letting the agent act), matching the catalog's money/approval rule.", "worst": "Only 3 top-level test files (plus e2e/ui scripts) for a plugin that bridges live browser control across three surfaces, leaving reliability under-verified relative to its scope.", "standout": true, "blocked": null}
]}