{"batch":"read2-017","plugins":[{"pr":0,"name":"done-bell","does":"Rings when a Hermes turn finishes with OS notifications, a chime, status chip, and approval-needed state.","useful":3,"reliable":4,"polish":3,"creative":2,"beyond":3,"total":18,"best":"It is a focused Desktop-and-hook improvement that lets a user leave a long-running session unattended.","worst":"The core idea is narrow and mostly notification plumbing, with modest product depth compared with platform plugins.","standout":false,"blocked":null},{"pr":0,"name":"homeassistant","does":"Connects Hermes to Home Assistant events and REST services, providing filtered gateway delivery and four device-control tools.","useful":5,"reliable":5,"polish":5,"creative":3,"beyond":5,"total":28,"best":"It turns Hermes into a useful, bidirectional smart-home platform with explicit filtering, cooldowns, and service-domain guards.","worst":"It still depends on a long-lived Home Assistant token and remote-instance configuration, so deployment is not frictionless.","standout":false,"blocked":null},{"pr":153595,"name":"tinfoil-sandbox","does":"Adds a terminal backend that runs Hermes terminal, file, and code operations inside a user-selected attested Tinfoil Container over verified SSH.","useful":4,"reliable":4,"polish":3,"creative":4,"beyond":5,"total":25,"best":"It changes where Hermes runs and keeps the existing tool surface while stripping tunnel credentials and retaining dangerous-command approval.","worst":"The small plugin has no bundled tests and requires several separately installed Tinfoil, SSH, container, and key prerequisites.","standout":true,"blocked":null},{"pr":154078,"name":"harness-story","does":"Collects Hermes lifecycle telemetry and renders sessions as inspectable causal narratives with tools, context pressure, code impact, subagents, and a Desktop player.","useful":4,"reliable":4,"polish":5,"creative":4,"beyond":5,"total":27,"best":"The combination of lifecycle hooks, reconstruction, comparison, and an animated Desktop viewer is unusually complete and directly improves agent observability.","worst":"It reads sensitive session state and can send it to a configurable OpenAI-compatible endpoint, while the default endpoint is a LAN address and the narrative can overstate causality.","standout":true,"blocked":null},{"pr":149611,"name":"hermes-security","does":"Runs evidence-oriented repository security reviews with coverage tracking, optional controlled validation, secret redaction, reports, and a Desktop findings page.","useful":5,"reliable":5,"polish":5,"creative":4,"beyond":5,"total":29,"best":"It is a deeply integrated, tested security workflow whose evidence model, bounded validation, sealed reports, and coverage gaps address real agent risks.","worst":"The main review is model-mediated and dependency checks/validation add substantial runtime and operational complexity.","standout":true,"blocked":null},{"pr":152979,"name":"lydite","does":"Launches the external Lydite engine to split repository work among isolated workers, replay proofs, and report pull-request receipts, with an approval lock.","useful":4,"reliable":4,"polish":4,"creative":5,"beyond":5,"total":27,"best":"It gives Hermes a genuinely new verified multi-worker delivery loop rather than another prompt wrapper, and refuses launches until the human approves the exact plan.","worst":"The plugin ships no engine and requires Linux, a separately attested binary, profiles, Claude Code, GitHub CLI, and sandbox tooling.","standout":true,"blocked":null},{"pr":149414,"name":"xray","does":"Detonates installed plugins in child processes with bait and time warps, traces runtime access, compares versions, and can route flagged calls to approval.","useful":4,"reliable":4,"polish":4,"creative":5,"beyond":5,"total":27,"best":"It is a novel supply-chain auditing layer for Hermes that combines pre-install behavior checks with runtime attribution.","worst":"The disclosed child-process harness is not an OS sandbox and native code can bypass its tripwires, limiting the strength of its security conclusion.","standout":true,"blocked":null},{"pr":152596,"name":"coach","does":"Uses a second model for on-demand advice, automatic turn reviews, pre-completion diff review, and typed decisions with budgets and secret-scope handling.","useful":5,"reliable":5,"polish":5,"creative":4,"beyond":5,"total":29,"best":"Its hooks, review gate, spend controls, model separation, Desktop settings, and extensive regression tests make second-opinion review a usable Hermes capability.","worst":"It adds latency and model cost, and its safety still depends on the second model's judgment rather than a deterministic verifier.","standout":true,"blocked":null},{"pr":154003,"name":"crowdcode","does":"Gives cloud agents Base wallets, funds and runs paid tasks, supports x402/MPP services, and withdraws earnings behind wallet and spend controls.","useful":4,"reliable":3,"polish":4,"creative":5,"beyond":5,"total":26,"best":"It opens a genuinely new agent economy capability with wallet signatures, task execution, receipts, and an explicit Hermes approval threshold.","worst":"It is intrinsically high-risk financial code: the default is not ask-before-every-call, transfers/withdrawals have special accounting, and correctness depends on a remote service and wallet key.","standout":true,"blocked":null},{"pr":153225,"name":"agent-pool","does":"Coordinates isolated contributor agents that fetch repositories, claim and push work, exchange encrypted task artifacts, and track budgets and audit state.","useful":5,"reliable":4,"polish":4,"creative":5,"beyond":5,"total":28,"best":"It adds a substantial multi-agent contribution workflow with repository registration, claims, transport privacy, crypto, quotas, and many targeted tests.","worst":"Its large surface area spans Git, subprocesses, Tor, crypto, and remote coordination, so the trust model and failure modes remain difficult to validate completely.","standout":true,"blocked":null},{"pr":153960,"name":"on-air","does":"Stages a separate Hermes installation for live shows with setup/readiness checks, bounded process lifecycle, approvals, secret isolation, and stop truth.","useful":4,"reliable":5,"polish":4,"creative":5,"beyond":5,"total":28,"best":"It safely makes Hermes usable as a controlled live-show system while explicitly isolating homes, credentials, installs, and process shutdown.","worst":"The workflow is specialized and operationally heavy, requiring a prepared second Hermes stage and careful owner-side setup.","standout":true,"blocked":null},{"pr":151341,"name":"permission-canvas","does":"Adds a fail-closed computer-use backend with element-level allow, observe, approve, and hide policies plus screenshot/accessibility redaction and a human Canvas UI.","useful":5,"reliable":5,"polish":5,"creative":5,"beyond":5,"total":30,"best":"It changes Hermes desktop control at the backend boundary with exact action approvals, element identity, redaction, expiry, recovery rules, and unusually broad tests and demos.","worst":"It protects only traffic through this backend and still trusts the driver, OS accessibility layer, policy files, and a cooperating local account.","standout":true,"blocked":null}]}