{
 "batch": "read1-067",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "Rings an OS notification and chime and shows a busy/done/needs-you status chip when a Hermes turn ends or an approval is waiting, via four hooks, one status tool, a /bell command and a small Desktop pane.",
   "useful": 3,
   "reliable": 3,
   "polish": 2,
   "creative": 2,
   "beyond": 3,
   "total": 16,
   "best": "Small and honest: hooks never raise, approval commands are not stored, no keys or network, and one test file covers the store.",
   "worst": "The README's install steps are a developer's own Windows path plus a manual copy of the pane, and the idea is a common notifier.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Official Home Assistant plugin: a gateway platform adapter that turns state_changed events into agent turns, plus four tools to list entities, read state, list services and call services.",
   "useful": 4,
   "reliable": 5,
   "polish": 4,
   "creative": 2,
   "beyond": 5,
   "total": 25,
   "best": "HASS_TOKEN is declared in the manifest and read through the profile-scoped secret store, every network call has a timeout, dangerous service domains are blocked, and about 176 tests cover it.",
   "worst": "ha_call_service can unlock or open things with no approval step unless the user sets the opt-in entity allow/deny lists, and it is an extraction of an existing core feature rather than a new idea.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151607,
   "name": "hyperactive",
   "does": "Makes Hermes speak first: keeps a goal tree, promises and decisions in SQLite and uses Hermes cron jobs to send check-ins, a daily brief, stalled-goal nudges and tap cards that auto-apply only low-stakes choices.",
   "useful": 4,
   "reliable": 4,
   "polish": 4,
   "creative": 4,
   "beyond": 4,
   "total": 24,
   "best": "A coherent proactive layer built on public surfaces (six tools, /ha, CLI command, system-prompt section, three hooks, cron with a no-model pulse) where money, public and irreversible cards always wait for the person.",
   "worst": "The Telegram reaction handler is marked UNTESTED in the code and guesses which card a reaction answers, there is no CI, and the CLI fallback parses `hermes cron list` text with regexes.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 154032,
   "name": "truthgate",
   "does": "Snapshots files before write_file/patch calls and, in a pre_verify hook, makes Hermes continue the turn if it newly introduced hollow Python functions or embedded-secret patterns.",
   "useful": 3,
   "reliable": 4,
   "polish": 3,
   "creative": 3,
   "beyond": 4,
   "total": 21,
   "best": "Deterministic, standard-library-only and fail-open, with baselines so old findings are not re-reported, secrets reduced to type/line/hash, and 24 tests across six files.",
   "worst": "Narrow scope (Python stubs and a handful of regex secret shapes), no screenshots or demo, and failing open means it silently does nothing whenever a baseline is missing.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151089,
   "name": "momentic",
   "does": "A vendor skills bundle (five SKILL.md files plus an MCP config) for writing, running and repairing Momentic browser and mobile tests and running Mo bug bashes on Momentic's paid service.",
   "useful": 2,
   "reliable": 2,
   "polish": 3,
   "creative": 1,
   "beyond": 2,
   "total": 12,
   "best": "Well-written, detailed skills from the vendor, including careful guidance on keeping credentials out of commands.",
   "worst": "It contains no Hermes plugin code at all (no plugin.yaml or register), its README never mentions Hermes, the MCP servers run unpinned `npx -y` packages, and a skill would do the same because it is only skills.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153798,
   "name": "rom-studio",
   "does": "A toolkit for building your own handheld game from a user-supplied GBA decomp by chat: a bun CLI builds the ROM and drives a headless emulator for screenshots, and a Desktop pane shows the feed, maps, music and dialogue.",
   "useful": 2,
   "reliable": 3,
   "polish": 3,
   "creative": 5,
   "beyond": 3,
   "total": 19,
   "best": "A genuinely original use of an agent with a large readable toolkit (emulator driver, map/music/dialogue tools, BPS patch sharing), no ROM shipped, and loopback/Host/Origin gates on its local servers with tests.",
   "worst": "The Hermes plugin itself registers zero tools and zero hooks, the README's own \"Not tested yet\" section lists the pane inside Desktop and the skill loaded through the plugin, and setup needs a heavy Linux-only toolchain.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 149614,
   "name": "arena-skill",
   "does": "A skill plus a bracket.py bookkeeping script that runs a single-elimination tournament of up to 100 delegate_task sub-agents, each with a different strategy card, to find the best answer to one task.",
   "useful": 2,
   "reliable": 2,
   "polish": 2,
   "creative": 2,
   "beyond": 2,
   "total": 12,
   "best": "bracket.py is a real 1,235-line state machine that deals strategy cards, pairs matches and tracks verdicts under .arena/.",
   "worst": "It is a skill bundle whose manifest says it was \"generated from\" another author's repo, with a 13-line README, no tests, no Hermes plugin code and a default of 100 agents (hundreds of sub-agent calls).",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153148,
   "name": "hermes-multiplayer",
   "does": "Slack-like channels, DMs, threads and search inside Hermes Desktop, synced between up to five people's Hermes gateways directly or through a relay, where agents can post, read and be woken by @mentions.",
   "useful": 4,
   "reliable": 3,
   "polish": 4,
   "creative": 5,
   "beyond": 5,
   "total": 26,
   "best": "Turns single-user Hermes into a shared multi-human, multi-agent workspace using a platform handler, sync routes, four tools and a rich hand-written Desktop UI, with about 480 tests on sync, pairing, secret scope and wake injection.",
   "worst": "Its disclosures say \"No third-party services\" while the default team-room flow sends unencrypted messages through an author-hosted workers.dev relay hard-coded in the code, a joined room's key is stored in plaintext SQLite, and there is no CI.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 153111,
   "name": "amavid-hooker",
   "does": "A Node stdio MCP server with 16 tools and two skills for generating and editing images and videos with xAI Grok Imagine, plus a local browser flow-canvas UI.",
   "useful": 3,
   "reliable": 2,
   "polish": 2,
   "creative": 2,
   "beyond": 3,
   "total": 15,
   "best": "Full TypeScript source alongside the bundle, redacted errors, and a three-OS CI matrix that builds and smoke-tests the server.",
   "worst": "It is not a Hermes-native plugin and takes its credentials from the Grok CLI's ~/.grok/auth.json file instead of Hermes's secret store, spends paid xAI quota with no approval gate, and its local UI server has no auth.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 151218,
   "name": "hermes-axiom-agent",
   "does": "Injects a nine-article written doctrine into every model turn through pre_llm_call and offers a /doctrine command that reports which separate enforcer plugins are installed and enabled.",
   "useful": 2,
   "reliable": 4,
   "polish": 3,
   "creative": 2,
   "beyond": 2,
   "total": 15,
   "best": "Unusually honest about its limits (\"This plugin enforces nothing itself\"), with bounded safe reads of other plugins' manifests and 55 tests.",
   "worst": "What it adds is text in the prompt, which a skill or system prompt would do equally well, and its value depends on five other plugins by the same author.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153535,
   "name": "cartridge",
   "does": "A Hermes Desktop pane with a Game Boy / Game Boy Color emulator that a person plays while a turn runs and that pauses and mirrors the approval card when Hermes needs an answer.",
   "useful": 2,
   "reliable": 2,
   "polish": 4,
   "creative": 4,
   "beyond": 3,
   "total": 18,
   "best": "Polished, candidly documented Desktop work: a hardware-styled pane tied to busy/approval state, save states, only a WTFPL homebrew test ROM shipped, a readable bundle CI diffs against source, and golden-frame tests.",
   "worst": "plugin.yaml has `homepage:` indented under `license: MIT`, which is not valid YAML, so the manifest may not load; register() is a deliberate no-op and the agent gains no capability, and the pane shows approval buttons for clarify prompts it says it cannot answer.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153900,
   "name": "annotary",
   "does": "A document-review loop: an agent shares a Markdown plan, the person comments on exact lines in a Desktop page or phone web app, and Send back / Approve / @mention wakes the right agent in the chat that shared it.",
   "useful": 4,
   "reliable": 4,
   "polish": 5,
   "creative": 4,
   "beyond": 4,
   "total": 25,
   "best": "Finished like a product: ten tools, a consent-gated watcher that wakes agents, a Desktop page, a doctor command, demo video and screenshots, about 260 tests with CI, and timeouts throughout.",
   "worst": "The engine is a prebuilt `annotary` binary downloaded from the author's GitHub release at first use (pinned by SHA-256, but its source is not in the reviewed repo), and its owner token lives in a file under the Hermes root rather than the secret store.",
   "standout": false,
   "blocked": null
  }
 ]
}