{"batch":"deep-38","setup":"Cloned Hermes Agent and ran uv sync --locked with Python 3.14 in test/scratch; the Hermes CLI exposed install, validate, and doctor. The supplied tryit.py did not complete in this sandbox: its local scripted-model/isolated-runtime session hung, so catalog validation and plugin tests were used and no model tool-call result is claimed.","plugins":[{"pr":149414,"name":"xray","validate":"passed: manifest, isolated load/capability probe, registrations, security scan, and desktop surface checks passed.","inside":"Hermes catalog capability probe loaded it in isolation and found its CLI commands plus session/tool hooks; tryit empty and planned runs did not finish in the sandbox, so no chat calls are reported.","tests":"passed: 13 passed.","useful":4,"reliable":4,"polish":4,"creative":5,"beyond":5,"total":27,"rank":4,"best":"It gives Hermes a distinctive behavioral supply-chain check with time-warped bait rather than another static prompt review.","worst":"Its detonation is only a Python-level tripwire, not an operating-system sandbox, and the integration runner could not finish here.","blocked":null},{"pr":149272,"name":"no-cap","validate":"passed: manifest, in-process capability probe, registrations, security scan, and desktop surface checks passed.","inside":"Hermes capability probing loaded its reply and tool-result hooks, commands, platform handler, and Desktop surface; tryit could not complete the sandboxed fake-model session, so no chat calls are reported.","tests":"passed: 60 selected claim, evidence, and verification tests passed.","useful":5,"reliable":5,"polish":5,"creative":5,"beyond":5,"total":30,"rank":1,"best":"It makes the agent's success claims answerable to its actual tool receipts, a broad and unusually well-finished improvement to everyday Hermes work.","worst":"It verifies evidence for a claim rather than whether the underlying work is correct, so green receipts can still support a bad outcome.","blocked":null},{"pr":150453,"name":"workgraph","validate":"passed: manifest, isolated load/capability probe, tool/hook/middleware/skill registrations, and security scan passed.","inside":"Hermes capability probe loaded the workgraph tool, hooks, middleware, and skill; the empty and planned tryit sessions could not complete against the sandboxed fake model, so no chat calls are reported.","tests":"failed in this sandbox: 9 failed and 9 passed; failures were checks running bubblewrap, which could not create a NETLINK_ROUTE socket under the sandbox.","useful":4,"reliable":4,"polish":5,"creative":5,"beyond":5,"total":28,"rank":3,"best":"It turns exploratory agent coding into a persistent, inspectable graph of snapshots, alternatives, verification, and landing rather than a chat-only workflow.","worst":"The core verification path depends on bubblewrap, which makes installation and use less portable and could not run under this sandbox.","blocked":null},{"pr":152421,"name":"sheets","validate":"failed: the catalog security scan flagged prompt_injection_ignore in g1abc.json and g2.json; the other manifest, load, registration, collision, and desktop checks passed.","inside":"The catalog capability probe loaded the four sheet tools and its skill/Desktop surface, but tryit could not complete the sandboxed fake-model session, so no chat calls are reported.","tests":"passed: 14 Python tool tests passed; the Node workspace test was not run.","useful":4,"reliable":2,"polish":5,"creative":4,"beyond":4,"total":23,"rank":6,"best":"Its reviewed write-back model makes agent-collected spreadsheet data legible and reversible in a polished Desktop workspace.","worst":"It fails the catalog admission security scan, which caps reliability regardless of the passing unit tests.","blocked":null},{"pr":152214,"name":"passkey-gate","validate":"passed: manifest, isolated load/capability probe, registrations, dependencies, and security scan passed.","inside":"Hermes capability probing loaded gate_challenge, gate_verify, gate_status, CLI commands, and hooks; tryit could not complete the sandboxed fake-model session, so no chat calls are reported.","tests":"environment-limited: 256 passed, 47 failed, and 51 errored when local HTTP fixtures were routed through the sandbox proxy; a rerun without that proxy was blocked by the network policy.","useful":4,"reliable":5,"polish":5,"creative":5,"beyond":5,"total":29,"rank":2,"best":"It binds a human passkey approval to the exact sensitive action and can enforce one-use approval at Hermes's tool boundary.","worst":"It needs phone, HTTPS reachability, and a setup flow, so its strongest protection is not a one-click install for every user.","blocked":null},{"pr":153316,"name":"hermes-lens","validate":"passed with a warning: all catalog checks passed, but validation found private Hermes approval imports that may break on a future release.","inside":"Hermes capability probing loaded the Lens platform, approval hook, and CLI command; tryit could not complete the sandboxed fake-model session, so no chat calls are reported.","tests":"failed: 66 passed and 1 failed; the failing real-local-server CLI test could not reach its HTTP fixture through the sandbox proxy.","useful":3,"reliable":4,"polish":5,"creative":5,"beyond":5,"total":27,"rank":5,"best":"It gives a personal Hermes agent a carefully gated glasses-and-voice channel, with a phone tap rather than speech approving sensitive actions.","worst":"It is tied to Ray-Ban Meta hardware, an iPhone, and Gemini Live, while also relying on private Hermes approval imports.","blocked":null}]}