{
 "batch": "read1-128",
 "plugins": [
  {
   "pr": 0,
   "name": "done-bell",
   "does": "Fires a local OS notification and a Desktop status chip when a Hermes turn finishes, needs approval, or goes idle.",
   "useful": 3,
   "reliable": 4,
   "polish": 3,
   "creative": 3,
   "beyond": 4,
   "total": 21,
   "best": "Real hook-driven logic with 9 passing tests and no secrets or network access at all.",
   "worst": "Install docs hardcode a developer's personal Windows desktop path, hurting polish.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 0,
   "name": "homeassistant",
   "does": "Official gateway plugin that streams Home Assistant device state into Hermes and lets it call HA services, with a domain blocklist and regex validation.",
   "useful": 5,
   "reliable": 5,
   "polish": 5,
   "creative": 3,
   "beyond": 5,
   "total": 28,
   "best": "168 real tests including adversarial service-policy regression tests, and the HA token is pulled through Hermes's own secret_scope, never a file.",
   "worst": "Device-control actions rely only on a hardcoded domain blocklist, not an explicit per-call human approval step; the README admits this is HA's limitation, not a gap it fully closes.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152526,
   "name": "openbao",
   "does": "Registers a Hermes secret-source provider that resolves bao:// references against an OpenBao/Vault KV HTTP API so other plugins' secrets can come from Vault.",
   "useful": 3,
   "reliable": 4,
   "polish": 3,
   "creative": 3,
   "beyond": 4,
   "total": 21,
   "best": "Bootstrap token is read only via Hermes's own secret-source environment API, marked protected, and the plugin explicitly never touches .env or auth.json.",
   "worst": "Only 3 test functions; KV v1 fallback, namespace headers, HTTP error codes and the response-size cap are all implemented but untested.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152360,
   "name": "proofread-draft",
   "does": "Adds a Desktop composer button that sends the current draft to an LLM for proofreading and shows a diff the user must explicitly click Apply to accept.",
   "useful": 4,
   "reliable": 4,
   "polish": 3,
   "creative": 2,
   "beyond": 3,
   "total": 19,
   "best": "The rewritten draft is never auto-applied; the user must click Apply, and the 6000-char limit is enforced in both the backend and the UI.",
   "worst": "Functionally it is close to a prompt-based grammar check; the main plugin-only value is the Desktop toolbar button and REST route, not the proofreading logic itself.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152201,
   "name": "followthrough",
   "does": "Four tools that let the model capture, list, complete and snooze user commitments in a local SQLite store tied to the Hermes profile.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 2,
   "beyond": 3,
   "total": 19,
   "best": "Prefers Hermes's own profile-scoped plugin_db storage API and a DEMO.md that explicitly lists 'claims to avoid' so contributors don't overstate it.",
   "worst": "It is a fairly thin CRUD tool; nothing here (hooks, hardware, model/platform access) goes much beyond what a well-designed skill with persistent storage could approximate.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 152068,
   "name": "component-workbench",
   "does": "A review/governance workflow that stores immutable candidate file snapshots, runs static syntax-only checks, freezes a candidate against a committed hash, and records human review verdicts in a tamper-evident SQLite ledger, with no execution or promotion ability at all.",
   "useful": 3,
   "reliable": 4,
   "polish": 3,
   "creative": 4,
   "beyond": 4,
   "total": 22,
   "best": "Tests (outside the plugin's own folder) cover path-traversal rejection, DB tamper detection, and an explicit 'no execution or promotion' guarantee; secrets like .env/auth.json are explicitly blocked from being stored.",
   "worst": "Its test suite lives at the parent repo root rather than inside the plugin's own tests/ folder, an organizational inconsistency, and the README itself flags the plugin as experimental/unaudited.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 150021,
   "name": "hermes-egm",
   "does": "A single egm_review tool that runs fresh, tool-free LLM review passes over a caller-supplied text snapshot and validates that every cited piece of evidence is a literal quote from the stated source line.",
   "useful": 3,
   "reliable": 5,
   "polish": 5,
   "creative": 3,
   "beyond": 4,
   "total": 24,
   "best": "47 tests cover concurrency, SQLite persistence and restart, and dozens of malformed/forged-evidence inputs; the docs are unusually self-critical ('Reviewers can miss defects, hallucinate reasoning or agree wrongly') rather than hyping the tool.",
   "worst": "It is a fairly narrow advisory review tool for text snapshots; usefulness is real but aimed at a specific coding/review niche rather than a broad Hermes user base.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 153446,
   "name": "signaldesk",
   "does": "Scans public hiring/launch feeds and local-business listings, scores leads against a user ICP, drafts outreach, and only ever writes a local .eml file, never sending automatically.",
   "useful": 4,
   "reliable": 5,
   "polish": 4,
   "creative": 3,
   "beyond": 5,
   "total": 26,
   "best": "A pre_tool_call hook blocks any send-like tool call unless the matching lead is already approved, and unconditionally blocks sends during any cron/unattended execution \u2014 a host-enforced guardrail no prompt or skill could replicate.",
   "worst": "The core idea (lead-gen scanning and scoring) is a familiar SaaS pattern; the novelty is mostly in the enforcement plumbing rather than the lead-gen concept itself.",
   "standout": true,
   "blocked": null
  },
  {
   "pr": 153432,
   "name": "fortograph",
   "does": "Three tools that upload a local image to a third-party 'Fortograph' web API to selectively encrypt (redact) regions and later decrypt them with a password.",
   "useful": 2,
   "reliable": 2,
   "polish": 3,
   "creative": 3,
   "beyond": 3,
   "total": 16,
   "best": "The password secret is resolved through Hermes's secret_scope first before falling back to a plain env var, matching the catalog's secret-handling expectations.",
   "worst": "Calling fortograph_redact/unredact uploads the user's image to a remote third-party server immediately with no approval gate, and the only verification is a single manual end-to-end script requiring live network access, not an automated test suite.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153322,
   "name": "fusen",
   "does": "A single editable 'sticky note' that can be pinned to many chats at once and is injected into the model's context only when stale, with revision history and an optional approve-before-write mode.",
   "useful": 3,
   "reliable": 5,
   "polish": 4,
   "creative": 4,
   "beyond": 4,
   "total": 24,
   "best": "49 real unit tests cover revision/revert, optimistic-concurrency conflicts, and stale-detection logic; the README states plainly 'Network: none. No telemetry, no remote calls, no API keys,' matching the code (all state is one local JSON file).",
   "worst": "It is a genuinely useful but narrow productivity feature; the submission's own checklist admits a recorded demo and some catalog-entry items are still unchecked.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 149358,
   "name": "bumblebee-guard",
   "does": "Wraps an external local supply-chain scanner bridge to check installed packages/MCP configs for malicious code, surfaces results as Hermes tools, and injects a rate-limited one-line alert only when findings exist.",
   "useful": 3,
   "reliable": 4,
   "polish": 3,
   "creative": 2,
   "beyond": 3,
   "total": 18,
   "best": "The scan subprocess is run with an explicitly minimized environment (only PATH/HOME/LANG/INSTALL_DIR) specifically to avoid leaking provider API keys, and this is directly unit-tested.",
   "worst": "The actual scanning logic is delegated entirely to an external bridge project not bundled in this repo, so this plugin is mostly a thin wrapper and alert-injection layer around someone else's tool.",
   "standout": false,
   "blocked": null
  },
  {
   "pr": 153039,
   "name": "glide",
   "does": "Bundles a large pre-existing forked terminal-UI project (hermes-tern) together with a small original 'Glide' add-on that locally recalls past user corrections and tracks deep-work hours to hold back non-critical notices.",
   "useful": 3,
   "reliable": 4,
   "polish": 4,
   "creative": 3,
   "beyond": 3,
   "total": 20,
   "best": "Browser-automation actions are gated behind an explicit tools.override capability grant the plugin never grants itself, and the genuinely new Glide memory sync opens the session DB read-only with a test that blocks any socket/subprocess use during sync.",
   "worst": "The repo's extensive docs and test counts come overwhelmingly from the inherited hermes-tern frontend (PRs #2-#28 attributed upstream); the hackathon-original Glide slice is comparatively modest, mostly a local corrections-recall hook and an attention heuristic.",
   "standout": false,
   "blocked": null
  }
 ]
}