{ "batch": "read1-018", "plugins": [ { "pr": 0, "name": "done-bell", "does": "Rings an OS notification and chime, plus a Desktop status chip and pane, when a Hermes turn finishes or waits for approval.", "useful": 3, "reliable": 3, "polish": 2, "creative": 2, "beyond": 3, "total": 16, "best": "Hook-driven OS and Desktop alerts with a mute tool and /bell command, nine tests, and hooks that never raise.", "worst": "Thin README with a hardcoded personal Windows path, no screenshots, a placeholder catalog sha, and a config_schema the code never reads.", "standout": false, "blocked": null }, { "pr": 0, "name": "homeassistant", "does": "Official Home Assistant platform adapter (WebSocket state_changed events to the agent) plus four REST tools to list, read and call HA services.", "useful": 4, "reliable": 5, "polish": 4, "creative": 3, "beyond": 5, "total": 26, "best": "About 176 tests, timeouts on every network call, secrets via Hermes's scoped secret store, and allow/deny policy checks on the real service payload.", "worst": "ha_call_service can drive locks and alarms with no approval step of its own, and the allow/deny lists are opt-in; no Desktop UI.", "standout": false, "blocked": null }, { "pr": 151538, "name": "follow-through", "does": "llm_execution middleware that catches a final reply promising future work and turns it into a follow_through tool call telling the agent to do the work now.", "useful": 4, "reliable": 3, "polish": 4, "creative": 4, "beyond": 4, "total": 23, "best": "Rewriting the model's final response at the middleware layer fixes a common agent failure no prompt can enforce, and it fails open and caps nudges per turn.", "worst": "Never checks whether work is actually running, and a bare \"I'll\" / \"started\" regex branch makes the trigger much broader than a promise; only 9 tests.", "standout": false, "blocked": null }, { "pr": 150545, "name": "word-miner", "does": "Local SQLite spaced-repetition vocabulary deck: the agent mines words from pasted text, and you study them in chat (/deck) or a Desktop flashcard pane.", "useful": 3, "reliable": 3, "polish": 4, "creative": 2, "beyond": 3, "total": 18, "best": "A real SRS engine with an event log, about 34 tests, no network or secrets, and a banner, screenshots and demo video.", "worst": "Mastery depends on sub-2s response times the chat flow can't hit, the Desktop pane can't answer graduation cards, CSV import has no tool, and the license is TBD.", "standout": false, "blocked": null }, { "pr": 149363, "name": "hermes-openready", "does": "Wraps the external Open-Ready CLI as five release-gate tools and adds a pre_tool_call hook that blocks terminal gh/curl commands making a repo public without a digest-scoped owner approval.", "useful": 2, "reliable": 4, "polish": 3, "creative": 3, "beyond": 4, "total": 20, "best": "A fail-closed enforcement hook with about 87 tests (188 cases with parametrization), CI, an allowlisted child environment and process-group timeouts.", "worst": "Does nothing without the separately installed Open-Ready CLI, only guards the terminal tool, and claims to read the origin remote, which no code does.", "standout": false, "blocked": null }, { "pr": 152686, "name": "ariadne-thread", "does": "Desktop page that draws cron jobs, webhooks and skills as n8n-style step chains, with in-place editing, conflict checks, history/undo and change requests drafted into a new chat.", "useful": 4, "reliable": 4, "polish": 4, "creative": 4, "beyond": 4, "total": 24, "best": "A full Desktop UI over Hermes's own automation, with atomic writes, confined paths, credential redaction, 94 backend tests and screenshots.", "worst": "Imports Hermes internals (cron.jobs), the conflict check on whole-skill saves is optional, and script editing through the dashboard API is effectively code execution.", "standout": false, "blocked": null }, { "pr": 152780, "name": "verifiable-math-kernel", "does": "SymPy identity checks, integer factorization, elliptic-curve invariants and a regex Lean sorry scanner as tools, plus pre/post tool hooks.", "useful": 2, "reliable": 2, "polish": 2, "creative": 2, "beyond": 3, "total": 14, "best": "SymPy and factorization really run locally with no keys and no network.", "worst": "Claims more than the code does (no Tate or 2-descent, torsion via an unchecked Lutz-Nagell test, '100% deterministic' primality), its parse timeout can't fire, and the README promotes a token.", "standout": false, "blocked": null }, { "pr": 150456, "name": "lazy-plugin", "does": "Desktop page and lazyplugin_create tool where your agent drafts a spec, four fixed templates generate a new plugin, and the page validates, contract-checks and test-dispatches each tool.", "useful": 3, "reliable": 2, "polish": 4, "creative": 5, "beyond": 4, "total": 22, "best": "A plugin that builds, validates and grades Hermes plugins, with about 106 tests, secret-store keys in generated code and a designed Desktop page.", "worst": "The README says templates can't carry injected code, but the raw title/description goes into a generated docstring that create() imports in-process, so a description with triple quotes can run code.", "standout": false, "blocked": null }, { "pr": 149251, "name": "multi-chat", "does": "Web Dashboard tab with 1 to 16 xterm panes, each attached to Hermes's own /api/pty real TUI.", "useful": 3, "reliable": 3, "polish": 4, "creative": 3, "beyond": 4, "total": 21, "best": "Candid README, about 72 tests across Python, vitest and Playwright, reproducible bundle with readable source in the repo, and screenshots.", "worst": "Locked to Hermes 0.22.6 because it reads private PTY_REGISTRY internals; the PR is still open with the maintainer's change request not addressed; no Desktop plugin.", "standout": false, "blocked": null }, { "pr": 153660, "name": "repager", "does": "Captures a public page's hero section in Chromium, has you approve its exact copy and CTA, lets ctx.llm pick among four fixed layouts, verifies in a browser and exports HTML/CSS.", "useful": 2, "reliable": 4, "polish": 3, "creative": 4, "beyond": 3, "total": 19, "best": "Evidence-bound design: the model can't write copy or HTML, records are signed, 232 tests, and a sandboxed browser subprocess with SSRF blocking and timeouts.", "worst": "Narrow audience with heavy setup (Node, Playwright, Chromium) and strict model-route rules; the README is dense jargon with no screenshots, and approval is an agent-callable tool.", "standout": false, "blocked": null }, { "pr": 153026, "name": "imagination-studio", "does": "Standard-library HTTP bridge with ten tools that sends image, video, voice, song and sound jobs to a locally run Imagination Studio server and returns the files as MEDIA.", "useful": 3, "reliable": 4, "polish": 4, "creative": 3, "beyond": 3, "total": 20, "best": "Well-hardened bridge (no redirects, key only to private hosts, path confinement, size caps and download deadlines), key from Hermes get_secret, 28 tests, nine screenshots and a GIF.", "worst": "All generation happens in a separate GPU-bound app, so the plugin itself is a thin client, and tool calls can block for up to 600s.", "standout": false, "blocked": null }, { "pr": 150696, "name": "privacy-xray", "does": "Records metadata for every main and side-task model request (vendor, HQ country, counts of personal data types), traces a canary person, audits config with rollback-safe fixes, and can mask PII via llm_request middleware.", "useful": 4, "reliable": 4, "polish": 4, "creative": 5, "beyond": 5, "total": 27, "best": "Uses hooks and middleware to show and fix where user data goes, without storing content, with 79 tests, gateway refusals and a sandboxed dashboard report.", "worst": "No Desktop pane, and the tool's find with export=true writes report files despite being presented as read-only.", "standout": true, "blocked": null } ] }