{ "batch": "read2-021", "plugins": [ { "pr": 0, "name": "done-bell", "does": "Adds turn-completion chimes, OS notifications and an approval/status chip to Hermes Desktop and CLI.", "useful": 4, "reliable": 3, "polish": 3, "creative": 2, "beyond": 4, "total": 20, "best": "Actual lifecycle hooks and a Desktop busy-state watcher let users leave long-running turns unattended.", "worst": "The README gives a developer-specific Windows install path and manual pane copying, while the tests omit the Desktop notification path.", "standout": false, "blocked": null }, { "pr": 0, "name": "homeassistant", "does": "Connects Hermes to Home Assistant state-change events, device-control tools and notification delivery.", "useful": 4, "reliable": 5, "polish": 4, "creative": 3, "beyond": 5, "total": 26, "best": "A real gateway adapter combines bounded network calls, reconnects, scoped secrets and tested entity/service restrictions.", "worst": "Entity restrictions are opt-in and default service access remains broad; operators must configure their own safe scope.", "standout": false, "blocked": null }, { "pr": 153876, "name": "docket", "does": "Turns trades-job descriptions into deterministically priced quotes, owner-approved sends, customer acceptance and invoicing.", "useful": 3, "reliable": 5, "polish": 5, "creative": 4, "beyond": 4, "total": 25, "best": "Decimal pricing, unpriced-item blocking, approval gates and a tested quote lifecycle turn a concrete business workflow into a usable product.", "worst": "Its audience is specialized, and external customer acceptance requires an owner-provided public URL or tunnel rather than working remotely out of the box.", "standout": false, "blocked": null }, { "pr": 150261, "name": "excubitor", "does": "Uses a shared turn ledger to deduplicate gateway messages, detect tool loops, enforce spend thresholds and alert on hung turns.", "useful": 5, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "total": 27, "best": "Eight lifecycle hooks turn replay, spend and no-progress evidence into interventions rather than merely displaying logs.", "worst": "Budget enforcement deliberately fails open on internal errors, uses estimated prices when exact costs are absent and cannot cover internal events.", "standout": false, "blocked": null }, { "pr": 149757, "name": "epoch", "does": "Adds Kanban duration estimates and feeds completed-run wall-clock measurements back into a local calibration ledger.", "useful": 3, "reliable": 4, "polish": 4, "creative": 4, "beyond": 4, "total": 23, "best": "Durable run bindings, retry outboxes and concurrency tests handle cross-profile and out-of-order completion instead of trusting comment markers.", "worst": "The default execution path downloads a pinned npm CLI on first use and requires Node 22, adding a setup dependency before estimates can work.", "standout": false, "blocked": null }, { "pr": 151448, "name": "muster", "does": "Maintains a persistent fleet-health board for subagents with stall, repeated-failure and crash states plus controls for its own launches.", "useful": 4, "reliable": 4, "polish": 4, "creative": 4, "beyond": 5, "total": 26, "best": "Observer hooks and a SQLite ledger preserve fleet state across crashes while lifecycle handles enable honest, scoped Stop and Message actions.", "worst": "Stop and Message only work for Muster-launched workers in the live process; observed workers and handles from a previous process remain outside its control.", "standout": false, "blocked": null }, { "pr": 149426, "name": "pinpoint", "does": "Measures rendered pages in Hermes Desktop, overlays audit pins and gives the agent targeted inspection and before/after finding deltas.", "useful": 4, "reliable": 4, "polish": 5, "creative": 4, "beyond": 5, "total": 27, "best": "The consent-gated browser engine and extensive fixtures close the visual edit-and-check loop with measured selectors, dimensions and contrast.", "worst": "Audits are heuristic and omit several rendering cases, while requests lack session/window correlation and real-model end-to-end behavior is unverified.", "standout": true, "blocked": null }, { "pr": 150160, "name": "helm", "does": "Reworks Hermes Desktop into a cross-bot desk with pending approvals, live task plans, session summaries, themes and robot avatars.", "useful": 5, "reliable": 4, "polish": 5, "creative": 3, "beyond": 5, "total": 27, "best": "A coherent SDK-based interface makes existing agents easier to supervise and keeps approval responses behind explicit user clicks.", "worst": "The redesign depends on app data-slot CSS selectors, and approval visibility is limited by which sessions its gateway can reach.", "standout": false, "blocked": null }, { "pr": 153090, "name": "retouch", "does": "Adds marked-region image editing with per-mark agent threads, version history, human review and pixel checks outside each mask.", "useful": 4, "reliable": 5, "polish": 5, "creative": 4, "beyond": 5, "total": 28, "best": "Mask-constrained compositing, pixel-level rejection and state/concurrency tests enforce a useful editing guarantee that prompting alone cannot provide.", "worst": "Hidden edit sessions retain the user's usual tools, so hostile picture text is resisted by instructions rather than a sandbox; generative edits also depend on a configured provider.", "standout": true, "blocked": null }, { "pr": 151546, "name": "wearos", "does": "Pairs Wear OS devices to Hermes through Tailscale, delivers agent notifications and keys over a private lane, and adds optional local voice services.", "useful": 3, "reliable": 1, "polish": 4, "creative": 4, "beyond": 5, "total": 22, "best": "Gateway-minted pairing windows, device-secret pickup and peer-identity binding implement a substantial new hardware access path.", "worst": "tts_air.py reads a plugin-owned HF token file outside Hermes's secret store and SIGKILLs matching air_runner.py --serve processes without restricting them to this gateway.", "standout": false, "blocked": null }, { "pr": 152691, "name": "grimoire", "does": "Adds guarded local SQL analysis, query-linked numeric checks and a Desktop data explorer with charts and learning views.", "useful": 5, "reliable": 4, "polish": 5, "creative": 4, "beyond": 5, "total": 28, "best": "Locked-down DuckDB execution, saved query receipts and output hooks make repeatable data analysis available across chat and Desktop.", "worst": "Its numeric verifier is heuristic: verify.extract skips bare powers of ten and year-range integers, so the claim that every number is checked is too broad.", "standout": true, "blocked": null }, { "pr": 151968, "name": "tickover", "does": "Adds local sound recognition, before/after action verification and baseline-based machine-drift monitoring with scheduled checks and alerts.", "useful": 3, "reliable": 4, "polish": 4, "creative": 5, "beyond": 4, "total": 24, "best": "Real audio encoders, quality gates, confirmation logic and tested stream-source restrictions provide a novel perception tool for agents acting on hardware.", "worst": "The supplied benchmarks show weak detection of subtle machine anomalies and sensitivity to microphone placement, limiting how much users can trust a drift verdict.", "standout": false, "blocked": null } ] }