{
  "batch": "read1-025",
  "plugins": [
    {
      "pr": 0,
      "name": "done-bell",
      "does": "Plays an OS-native sound/toast and shows a busy/done/needs-you status chip when a Hermes turn finishes, via hooks on the session/approval lifecycle.",
      "useful": 3,
      "reliable": 5,
      "polish": 4,
      "creative": 2,
      "beyond": 3,
      "total": 20,
      "best": "Nine real unit tests and every hook wrapped so it can never raise, with no secrets or network access at all.",
      "worst": "The idea (a done notification) is a common, low-novelty utility most users could get from a simpler desktop script.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 0,
      "name": "homeassistant",
      "does": "Bridges Hermes to a Home Assistant instance over WebSocket/REST so the agent can see device state changes and call Home Assistant services (lights, locks, climate, notifications).",
      "useful": 4,
      "reliable": 4,
      "polish": 4,
      "creative": 3,
      "beyond": 5,
      "total": 25,
      "best": "Reads its token through Hermes's own secret-scope API rather than files or raw env vars, and has a broad real test suite covering filters, adapter and policy.",
      "worst": "ha_call_service can actuate real devices such as locks with no required per-call human approval; only an operator-configured domain/entity blocklist limits it.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153692,
      "name": "design-discovery",
      "does": "Runs a guided conversational design session that resolves colors/fonts/tokens from a curated vocabulary, checks WCAG contrast, tracks user consent, and exports a DESIGN.md handoff file.",
      "useful": 3,
      "reliable": 4,
      "polish": 3,
      "creative": 3,
      "beyond": 3,
      "total": 19,
      "best": "A large real test suite (161 test functions) exercises the actual token-resolution, contrast-math and consent-tracking logic, not stubs.",
      "worst": "Core code is written in a very dense, unidiomatic style (long chained statements, giant nested ternaries) that hurts readability and maintainability.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 151312,
      "name": "jw3",
      "does": "Wraps an external jw3 trading CLI with 17 Hermes tools for market data and order execution/cancellation on crypto positions, gated behind a human-approval token.",
      "useful": 3,
      "reliable": 5,
      "polish": 3,
      "creative": 2,
      "beyond": 3,
      "total": 19,
      "best": "Trades require a fail-closed, one-shot sha256 approval token bound to the exact order arguments, with tests for argv-injection and env-allowlisting of the child process.",
      "worst": "It is a thin wrapper around a separate external CLI the user must install themselves, and the trading-wrapper concept is a fairly common idea.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 149617,
      "name": "factledger",
      "does": "Maintains a durable TSV ledger of typed (scope, subject, predicate) facts where a new value supersedes rather than deletes the old one, exposed via 9 tools.",
      "useful": 3,
      "reliable": 4,
      "polish": 3,
      "creative": 3,
      "beyond": 3,
      "total": 19,
      "best": "Atomic file writes with bounded retry plus a dedicated packaging test that keeps the bundled plugin copy byte-identical to the root module.",
      "worst": "A disciplined structured-notes skill could approximate much of the supersede/history behaviour, so the case for needing a plugin over a skill is only moderate.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 149856,
      "name": "fledge",
      "does": "Gates Hermes's own skill-save/rewrite pipeline by scoring each turn's evidence (errors, corrections, denials) and blocking or requiring approval for skill writes that follow a bad run.",
      "useful": 3,
      "reliable": 5,
      "polish": 4,
      "creative": 4,
      "beyond": 5,
      "total": 26,
      "best": "153 tests plus an unusually candid 'Limits, honestly' section in its own README admitting where its heuristics can fail.",
      "worst": "No tools of its own and no screenshots/demo media, so its value is entirely invisible until a skill-write is actually blocked.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 154055,
      "name": "augur",
      "does": "Classifies chat turns against the user's three declared weekly goals, nudges on repeated drift, parks detours into a queue, and emails/renders a Friday report with research on parked items.",
      "useful": 4,
      "reliable": 5,
      "polish": 5,
      "creative": 4,
      "beyond": 4,
      "total": 26,
      "best": "A dedicated test_security.py specifically neutralizes terminal-escape and attachment-injection attempts from model output, plus a full eval harness with dev/held-out sets.",
      "worst": "Its weekly research/report feature depends on background LLM classification quality, which is harder to verify as reliable than the deterministic parts of the plugin.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 153776,
      "name": "soloknuckle",
      "does": "Scans diffs for leaked secrets, firewalls risky shell commands, and scores an npm project's health, exposed as three Hermes tools with no hooks or middleware.",
      "useful": 3,
      "reliable": 5,
      "polish": 4,
      "creative": 2,
      "beyond": 3,
      "total": 20,
      "best": "Unusually self-limiting documentation that explicitly lists what it deliberately did not build (a secret source, self-update) and why, plus cross-engine TS/Python parity tests.",
      "worst": "Secret scanning and a command firewall are well-worn security-tool concepts, so it scores low on originality despite strong execution.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 150208,
      "name": "slopcannon",
      "does": "Drives a paid external video-generation SaaS from inside Hermes: uploads, scene/character generation and multi-stage rendering, with every credit spend gated by an exact-match human approval.",
      "useful": 4,
      "reliable": 5,
      "polish": 5,
      "creative": 4,
      "beyond": 5,
      "total": 28,
      "best": "Credits are only ever spent after approval that must match the exact quoted figure, the plugin never requests the OAuth scope needed to buy credits itself, and unattended spend defaults to zero.",
      "worst": "Setup is relatively involved, requiring Node.js and an OAuth sign-in flow, and it depends entirely on a third-party paid SaaS staying available.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 153842,
      "name": "vitrail",
      "does": "Lets the model write GLSL fragment shaders that get validated, rendered live via a GPU-backed Hermes Desktop companion, and saved with auto-generated slider/color controls.",
      "useful": 3,
      "reliable": 5,
      "polish": 4,
      "creative": 4,
      "beyond": 4,
      "total": 24,
      "best": "Generated standalone HTML players are tested for being inert against injected titles/code (XSS-safe), and shader IDs are validated against path traversal.",
      "worst": "Value is capped for users without a GPU-capable Hermes Desktop session, since the richest feature (live on-GPU compilation) depends on that companion.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 151417,
      "name": "learning-workbench",
      "does": "Runs tutoring sessions with frozen answer keys, quote-verified exercise assessment, and a Hermes Desktop study pane, storing attempts in an append-only SQLite schema.",
      "useful": 3,
      "reliable": 5,
      "polish": 5,
      "creative": 4,
      "beyond": 4,
      "total": 25,
      "best": "The deepest test suite in the batch: 7 test files, a dedicated mutation-check script claiming to break 22 invariants, and a live end-to-end harness, plus honest scope docs on what wasn't tested.",
      "worst": "The README itself admits macOS and Windows paths were not tested, leaving real-world cross-platform reliability unverified.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 149270,
      "name": "hermes-watchdog",
      "does": "Watches Hermes's own cron incidents, matches them against a JSON playbook, and runs a bounded rerun/pause/check repair cycle, verifying success only against Hermes's own incident records.",
      "useful": 3,
      "reliable": 5,
      "polish": 4,
      "creative": 4,
      "beyond": 5,
      "total": 26,
      "best": "A runnable proof script reproduces the README's example output end-to-end, and the README candidly discloses an upstream Hermes gap it has to work around.",
      "worst": "Value only materializes for users who already run cron jobs through Hermes, a narrower audience than a general-purpose plugin.",
      "standout": true,
      "blocked": null
    }
  ]
}