{
  "batch": "final-11",
  "ranking": [
    {
      "rank": 1,
      "pr": 149640,
      "name": "hermes-ark",
      "total": 29,
      "why": "Useful 5, Reliable 5, Polish 5, Creative 4, Beyond 5: Ranks first for protecting nearly every user's accumulated Hermes state with automatic checkpoints, credential exclusions, verified restore staging and a finished recovery UI; 174 local tests passed and a real Hermes checkpoint worked, while live S3 recovery was not exercised ([pinned source](https://raw.githubusercontent.com/medomohamed22/hermes-ark/a23e529c1f636aa49d974576cdc5cb15fc8242d3/README.md))."
    },
    {
      "rank": 2,
      "pr": 152458,
      "name": "housemate",
      "total": 29,
      "why": "Useful 5, Reliable 4, Polish 5, Creative 5, Beyond 5: The strongest hardware entry makes old browsers into room speakers with a real platform adapter, persistent timers, wake arbitration and device-bound human approvals; its extensive local suite passed, but live room audio and speech providers remain unverified here ([pinned source](https://raw.githubusercontent.com/RaYYeR220/housemate/09c1a1d0a88bfa8762cee6b3ccf94271b0f172c5/README.md))."
    },
    {
      "rank": 3,
      "pr": 153844,
      "name": "fine-print",
      "total": 29,
      "why": "Useful 5, Reliable 5, Polish 5, Creative 4, Beyond 5: A broadly useful, dependency-free keeper exposes silent context loss through structured detectors, restrained notices and actionable recovery advice; the code distinguishes verified detectors from fixture-only coverage instead of claiming universal visibility ([pinned source](https://raw.githubusercontent.com/Joko-F/hermes-fine-print/7b574f1b5c8dc8eafe3ea9f87ae6b0111853f0ba/README.md))."
    },
    {
      "rank": 4,
      "pr": 153714,
      "name": "tenure",
      "total": 29,
      "why": "Useful 5, Reliable 4, Polish 5, Creative 5, Beyond 5: Per-action revocable trust, authenticated undo and a replayable autonomy ledger substantially improve the user's own agent, with unusually complete tests and UI; heuristic shell classification and extra approval friction after external reads keep it below the simpler safeguards ([pinned source](https://raw.githubusercontent.com/otsoa-dubief/tenure/f4a62855e592b9e170ab0c54d9278aabef05f370/README.md))."
    },
    {
      "rank": 5,
      "pr": 151887,
      "name": "cmi",
      "total": 28,
      "why": "Useful 5, Reliable 4, Polish 5, Creative 4, Beyond 5: Deterministic context selection with stored originals and exact recall addresses recurring context costs across models, making it an everyday keeper; the selected engine ran inside Hermes, but private compressor dependencies and experimental project memory limit confidence ([pinned source](https://raw.githubusercontent.com/jeremiebergel/context-manager-intelligence/399468f0687c051476961e4bb045eb84263be6b1/README.md))."
    },
    {
      "rank": 6,
      "pr": 150800,
      "name": "hypnos",
      "total": 28,
      "why": "Useful 4, Reliable 4, Polish 5, Creative 5, Beyond 5: The most ambitious model-changing entry implements real isolated training, gated promotion, fallback routing and lineage-based forgetting rather than merely describing sleep; separate trainer setup, narrow procedural families and unverified training quality here reduce day-one usefulness ([pinned source](https://raw.githubusercontent.com/0xNoramiya/hypnos/0a809b9085fdeaed63fd909b927afcc41de512de/README.md))."
    },
    {
      "rank": 7,
      "pr": 150596,
      "name": "ariadne",
      "total": 28,
      "why": "Useful 4, Reliable 4, Polish 5, Creative 5, Beyond 5: Real middleware recording and offline replay give Hermes a valuable regression-testing capability with comparison and bisect tooling; my replay matched the recorded tool result without reported network leaks, but correctly surfaced CLI-to-TUI and tool-inventory drift rather than reproducing an identical session ([pinned source](https://raw.githubusercontent.com/miracle352/hermes-ariadne/128c95967a0761ad24d3d224441d1502c43bbb69/README.md))."
    },
    {
      "rank": 8,
      "pr": 150840,
      "name": "kata",
      "total": 28,
      "why": "Useful 4, Reliable 4, Polish 5, Creative 5, Beyond 5: Mining repeated successful calls into human-approved deterministic tools is a substantial extension of Hermes and a credible long-term efficiency gain; conservative validation and schema-drift checks help, but support thresholds and a timeout that cannot cancel the underlying dispatch narrow its dependable reach ([pinned source](https://raw.githubusercontent.com/opeblow/kata/cc59b08f1bbeeb02524c060faac04edb34de0564/README.md))."
    },
    {
      "rank": 9,
      "pr": 149738,
      "name": "trust-ladder",
      "total": 27,
      "why": "Useful 5, Reliable 4, Polish 4, Creative 4, Beyond 5: Persistent corrections become actual pre-tool blocks or post-write checks, and a real Hermes call learned a useful guard with a clear bad-input error; pattern-based coverage, fail-open errors and a less finished presentation put it behind the richer agent improvements ([pinned source](https://raw.githubusercontent.com/merdoks/trust-ladder/fae4e9df58767e5e3832c8f65936b97868bc98fc/README.md))."
    },
    {
      "rank": 10,
      "pr": 152691,
      "name": "grimoire",
      "total": 27,
      "why": "Useful 4, Reliable 4, Polish 5, Creative 4, Beyond 5: Read-only, locked-down DuckDB execution plus a polished Desktop data workspace makes analysis substantially better than a prompt wrapper; its self-test passed, although matching answer numbers to result values is a heuristic check rather than proof that each claim uses the right evidence ([pinned source](https://raw.githubusercontent.com/bettersandeep/grimoire/4824773040b131f3179e5896d71d1435908758a7/README.md))."
    },
    {
      "rank": 11,
      "pr": 150644,
      "name": "hermes-canvas",
      "total": 27,
      "why": "Useful 4, Reliable 4, Polish 5, Creative 4, Beyond 5: The canvas genuinely binds plans to runtime writes and check exits while delivering user edits during work; real Hermes marked my box done from its check and refused a model-forged status, but shell inference and coding/Desktop specialization limit wider retention ([pinned source](https://raw.githubusercontent.com/0xbyt4/hermes-canvas/d6eeff0d6f7ec653262537bdd7bb001cc6d3c7f8/README.md))."
    },
    {
      "rank": 12,
      "pr": 153955,
      "name": "shadow",
      "total": 27,
      "why": "Useful 4, Reliable 4, Polish 5, Creative 4, Beyond 5: Personal conversation replays, blinded order-swapped judging and reversible model switching offer a useful improvement to model choice; extensive isolation machinery and a polished UI are offset by private-core compatibility exposure, mandatory live setup verification and untested paid-provider behavior here ([pinned source](https://raw.githubusercontent.com/akinduroifedayo/shadow/38173bc9c696df3b1f87b5b9b60ccecd36a260ab/README.md))."
    },
    {
      "rank": 13,
      "pr": 150353,
      "name": "fluent",
      "total": 27,
      "why": "Useful 5, Reliable 4, Polish 5, Creative 3, Beyond 5: Installed-help-based flag guards can prevent frequent shell mistakes without another model call, with careful execution budgets and approval escalation; only confident long-option matches on the local backend are enforced, so its practical coverage is narrower than its broad name suggests ([pinned source](https://raw.githubusercontent.com/Bruno-coder21/fluent/182c4301f2437e784e9022f1f38576e1d68c3f5f/README.md))."
    },
    {
      "rank": 14,
      "pr": 153020,
      "name": "breadcrumbs",
      "total": 27,
      "why": "Useful 4, Reliable 4, Polish 4, Creative 5, Beyond 5: Exact content-addressed recall, progressive zoom and an actual context engine form an original memory architecture, and configured Hermes searches and invalid-id handling worked; append-only retention, lexical retrieval and a terminology-heavy setup make it a less universal keeper than the leaders ([pinned source](https://raw.githubusercontent.com/breadcrumbs-agency/hermes-breadcrumb-plugin/b8975645e5174edbcc00d5dc86a78e536bce938b/README.md))."
    },
    {
      "rank": 15,
      "pr": 150556,
      "name": "hermes-ui-cards",
      "total": 27,
      "why": "Useful 5, Reliable 4, Polish 5, Creative 3, Beyond 5: Seventeen validated interactive card types give Desktop users useful new ways to steer Hermes, with working CLI fallbacks and precise malformed-card errors; familiar UI patterns and explicitly eventual cross-client answer locking place it below the deeper agent changes ([pinned source](https://raw.githubusercontent.com/robbyczgw-cla/hermes-ui-cards/c40df4d41cd5439cf4902995fd4a9e5505483f25/README.md))."
    },
    {
      "rank": 16,
      "pr": 153851,
      "name": "foreman",
      "total": 27,
      "why": "Useful 4, Reliable 4, Polish 5, Creative 4, Beyond 5: Diff-based anti-test-rigging checks, owner-only exceptions and a second-model review create a polished coding safeguard; the disclosed alternate merge routes and heuristic detector coverage leave less comprehensive protection than its green-check framing might imply ([pinned source](https://raw.githubusercontent.com/PedValme/foreman/c15453cb2fc07fa674d6f7b548a553b3ce9e6f5d/plugin/README.md))."
    },
    {
      "rank": 17,
      "pr": 152908,
      "name": "contract-necromancer",
      "total": 26,
      "why": "Useful 4, Reliable 4, Polish 4, Creative 4, Beyond 5: Owner-frozen tests, fresh verify-step execution and content-pinned receipts enforce a real definition of done; the bounded gate, shell-only edits being caught at stamping rather than continuation, and a primarily command-driven product make it narrower than the higher coding finalists ([pinned source](https://raw.githubusercontent.com/vijaygopalbalasa/hermes-necromancer/0ff7bec40aba10091a84f94025d4471984fd7386/README.md))."
    },
    {
      "rank": 18,
      "pr": 150251,
      "name": "gemini-live",
      "total": 26,
      "why": "Useful 4, Reliable 4, Polish 5, Creative 3, Beyond 5: Full-duplex voice with delegation and thread steering is a well-integrated new Hermes surface, with scoped secrets and server-locked ephemeral tokens; a required external voice API, unverified live calls here and conservative read-only path restrictions reduce its edge over the leaders ([pinned source](https://raw.githubusercontent.com/cineglobe/hermes-gemini-live/b6c2f24354aeed1a8b5651b62944d14ed96a472d/README.md))."
    },
    {
      "rank": 19,
      "pr": 153809,
      "name": "living-spaces",
      "total": 26,
      "why": "Useful 4, Reliable 3, Polish 5, Creative 4, Beyond 5: Persistent sandboxed mini-apps, shared records, revisions and native cron give Hermes a flexible Desktop capability; private integrations, plaintext Space-supplied credentials and the documented WebRTC exception to Internet Off warrant a larger reliability discount ([pinned source](https://raw.githubusercontent.com/akinduroifedayo/living-spaces/e6145ccbd028edf55417822020426f4830343eb6/src/living-spaces/README.md))."
    },
    {
      "rank": 20,
      "pr": 152651,
      "name": "workroom",
      "total": 25,
      "why": "Useful 4, Reliable 2, Polish 5, Creative 4, Beyond 5: The extensive document workflow is held back by the pinned snapshot hook's unbounded leading-cd regex: a local probe exceeded four seconds on 30,000 newlines, so later PR-described fixes cannot earn this commit reliability credit ([pinned source](https://raw.githubusercontent.com/Pratiikpy/workroom-for-hermes/c5f6bf334c1163951e203dc7c5777faadfe49795/adapters/snapshotter.py))."
    }
  ]
}