# Fix list: Hermes Agent From Anchor Terminal's listing at https://www.anchorterminal.com/tools/hermes-agent, the October 2026 research run, assessed 12 October 2026. Grade B, 67.9 out of 100. This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public. For a coding agent working on Hermes Agent: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published. ## 1. Reliability, 57 out of 100, up to 8.6 more on the total Why it scored 57: Read with the local-software lines, as the Goose and Khoj dossiers are. An install script, desktop installers for macOS and Windows, a Docker image (`nousresearch/hermes-agent`) and a signed Termux repository, with Python 3.14 stated as the only supported version, but the docs say no supported wheel or source distribution is published (18 of 20). `ci.yaml` runs on every pull request and push to main and the repository holds about 5,700 test files, and a stable release must pass the whole suite. The Actions page we loaded showed only pull-request runs, so the state of main is unconfirmed (18 of 25). 15,102 open issues and more than 5,000 open pull requests on 12 October 2026. New reports get priority and component labels the same day, but the first page held P1 bugs on session state and context compression with no comment (10 of 25). The project moved from date versions to SemVer at v0.21.6 on 8 October 2026, and the v0.21.6 notes, which cover about 2,106 pull requests, have no breaking-change section (6 of 15). Version 0.21.7, before 1.0, on a channel the project calls stable (5 of 15). The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability): Hosted APIs, MCP servers, models and platforms. - 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own). - 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so. - 15, rate limits documented with numbers. - 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved. - 10, an SLA published for any paid tier. - 10, the surface agents use is generally available, not beta or preview. Local packages, SDKs, frameworks and stdio MCP servers. - 20, installs from an official package with supported runtimes stated. - 25, a public CI and test suite, passing on the default branch. - 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered). - 15, semver discipline and breaking changes called out in a changelog. - 15, version 1.0 or later, or declared stable. Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors. ## 2. Payments & pricing, 60 out of 100, up to 5 more on the total Why it scored 60: Read with the self-hosted rule. No payment protocol on Hermes Agent's own interfaces (0). The optional MPP skill makes the agent a payer, which this line does not count. The software is free under MIT, needs no account and runs with a local model, so 20, 20 and 20 on the last three lines. Nous sells two optional services we did not score here, Nous Portal plans at $0 to $200 a month and Hermes Cloud at $0.51 a day running on the Standard size, both with public prices. The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments): The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/). - 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which. - 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login. - 20, a free tier or trial that doesn't need a card. - 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API). Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied. Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol. ## 3. Security & auth, 75 out of 100, up to 4.4 more on the total Why it scored 75: Harness reading used for Goose. 30 for what leaves the machine by default, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. The FAQ says no telemetry, usage data or analytics is collected and our search of the source found no analytics client in the core, with passive update checks the owner can disable (28). Three approval modes, a fixed blocklist, owner deny rules, `deny` defaults for cron, one-shot and webhook sessions, card fills that always ask, and a gateway that denies all senders by default. Against that, the default backend runs on the host, the default `smart` mode lets an auxiliary model approve commands, container backends skip the check, memory writes are ungated by default, and sending mail through the Google Workspace skill is held back only by an instruction to the model (14). Context files and cron prompts are scanned for injection patterns, URL tools block private addresses, and SECURITY.md says plainly that these are heuristics, that prompt injection alone is out of scope and that untrusted input calls for whole-process sandboxing, which is not the default (10). Logs under `~/.hermes/logs/`, every session in a local database, an approval history that can be mined and an install audit log (12). SECURITY.md with a scope, a 90-day disclosure window, a security address and GitHub private reporting, plus OSV and supply-chain audit workflows and exact dependency pins. No bug bounty by its own statement, no security.txt (404), and no published advisory although v0.21.6 lists six security fixes (11). The checklist (https://www.anchorterminal.com/benchmark/#checklist-security): - 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option. - 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions. - 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10. - 0 to 15, audit logs or per-call visibility for the operator. - 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public. Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing. ## 4. Schema & documentation, 80 out of 100, up to 3.3 more on the total Why it scored 80: Harness reading of the framework line, as for Goose. Reference pages for configuration, environment variables, CLI commands and MCP config, and the API server documents its routes, but no OpenAPI file was found (17 of 25). `llms.txt`, a per-section docs index and a full-text file are served (10). The security page says what each guard catches and misses, and pages such as heartbeat against cron say when to use which (16 of 20). Config keys are listed with defaults and allowed values, such as the three approval modes (12 of 15). An FAQ explains error messages and their fixes, and each feature page carries examples (13 of 15). Dated release notes on GitHub with no changelog file, and the version scheme changed in October 2026 (12 of 15). The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema): APIs and MCP servers. - 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool). - 10, llms.txt or Markdown docs served for agents. - 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference. - 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs. - 0 to 15, examples and documented error responses. - 15, versioning and a public changelog. Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference. ## 5. Agent ergonomics, 80 out of 100, up to 3.3 more on the total Why it scored 80: Harness reading, scored on what a person or pipeline driving it must supply. A tool search index, toolsets and per-server MCP include and exclude filters keep definitions out of context (22 of 25). `--max-turns` with a default of 500, automatic context compression and a usage report file, though 500 turns is high for an unattended run (15 of 20). `hermes chat -q` with `--format stream-json`, and the FAQ maps errors to fixes (16 of 20). The runs API takes an `Idempotency-Key`, sessions resume, and cron tracks failure streaks and delivery failures apart from run failures (17 of 20). One install command and a setup wizard, but a model provider must be configured first, and the Python library has no published package (10 of 15). The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics): - 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries). - 20, pagination, filtering and output-size controls. - 20, actionable, documented error responses, codes and messages an agent can recover from. - 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations. - 15, sensible defaults, few required parameters, and official SDKs in at least two languages. Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs. ## 6. Transparency & trust, 63 out of 100, up to 3.2 more on the total Made of editorial 73, provenance 53. Why it scored 63: MIT (30). The FAQ says nothing is collected and data stays in `~/.hermes/`, and the docs name where memory, sessions, tokens and the vault live. There is no privacy statement for the software itself, the FAQ's claim that API calls go only to the model provider leaves out the web search, speech and browser backends the docs describe elsewhere, and the docs do not mention that the Portal terms allow training on customer data (17 of 30). The CLI reference marks deprecated commands and `hermes migrate` rewrites retired settings, but no deprecation policy or dated notices were found (8 of 20). No telemetry to opt out of per the docs, and update checks can be turned off (18 of 20). The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency): - 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms. - 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors). - 0 to 20, a deprecation policy or notices with dates. - 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted). The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two. Provenance checks not met in full (half of this category, computed from checked facts): - Domain age: nousresearch.com, no registry record we could read (0 of 15) - Status page: not found (0 of 10) - security.txt: not found (0 of 10) ## 7. Maintenance & community, 84 out of 100, up to 1.4 more on the total Why it scored 84: v0.21.7 was published on 11 October 2026, the day before this check (30). Ten releases appear between 19 August and 11 October 2026 (20). Issues are labelled by priority and component the day they arrive and v0.21.7 merged about 204 pull requests in three days, but 15,102 issues and more than 5,000 pull requests are open and most of the day's reports had no comment (13 of 25). Not an MCP server, so read as for Goose on builds. The Docker image, desktop bundles, installers and Termux packages are promoted from one release receipt (13 of 15). Exact dependency pins with a lock file, an OSV scanner, a supply-chain audit and Dependabot for Actions, with CI on main unconfirmed (8 of 10). The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance): - 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older. - 20, at least three releases or dated changelog entries in the last 90 days. - 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15. - 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models). - 10, package health, current dependencies and CI. Models are read for deprecation notice periods and model churn rather than release counts. ## Deductions Each comes off the total. A fixed and documented problem counts for less at the next check. - 2026-10-08. The v0.21.6 release notes list six security fixes, among them a native sign-in flaw that could send login codes to a non-loopback redirect and a quoted display name in the `From` header that affected the email gateway's sender check. SECURITY.md counts unauthorised access to an external surface as in scope, yet the repository shows no published advisory, so owners on older versions have no advisory to act on. Fixed and described in the notes, -3. https://github.com/NousResearch/hermes-agent/releases/tag/v0.21.6; https://github.com/NousResearch/hermes-agent/security/advisories ## What we couldn't check What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it. - Kind is `harness`. Hermes Agent is an agent program the owner runs, which is the closest kind the briefs define, and Reliability and Payments are read as local, self-hosted software - The Nous Portal terms and privacy policy are not linked as `provenance.terms` and `provenance.privacy`, because they govern the optional hosted services and not the software under the MIT licence. The licence stands in - Hermes Cloud, the hosted version, was not graded. It has its own prices, runs on Nous infrastructure and would be read as a hosted service - Unchecked: whether CI passes on main today. The Actions page showed only pull-request runs - Unchecked: the full text of the six security fixes in v0.21.6. Our reader returned the first line of each, and the affected versions were not stated - Unchecked: release dates before 19 August 2026 and the first release date. GitHub's robots.txt closes the tags and commits pages to Anthropic agents, and the clone held three days of history - Unchecked: whether an OpenAPI file exists for the API server. None was found in the docs pages read - Unchecked: whether the Docker image name in the docs resolves on Docker Hub, and the domain registration date - Whether the Nous Portal free plan needs a card was not established - The star count is as GitHub's page showed it on 12 October 2026, rounded to the thousand - github.com's robots.txt answered 200 and closes commits, tags, tree and raw paths to ClaudeBot and anthropic-ai. We read the source from a shallow clone and the releases, advisories, issues and Actions pages, none of which it closes. hermes-agent.nousresearch.com and portal.nousresearch.com answered 200, and nousresearch.com's robots.txt disallows its terms and privacy paths, which we did not read ## Weaknesses - The default terminal backend runs commands on the host, and the default `smart` mode lets an auxiliary model approve commands it rates low-risk - The Google Workspace skill requests full Gmail, Calendar and Drive scopes, and its confirm-before-sending rule is an instruction in SKILL.md, not enforced in code - Memory writes are saved without approval by default (`memory.write_approval: false`), including from a background review after each turn - Release v0.21.6 of 8 October 2026 lists six security fixes and the repository has no published security advisory - 15,102 open issues and more than 5,000 open pull requests on 12 October 2026 - The Nous Portal terms grant Nous a licence to train models on customer data when the Portal is the model provider ## What costs an agent a turn today The notes we give agents before they call it. Each one is a workaround an agent shouldn't need. - Set `approvals.mode: manual` before giving it real accounts. The default `smart` lets an auxiliary model approve commands - Run the gateway with `terminal.backend: docker` or the project's Docker image when it reads mail, web pages or group chats - Set `memory.write_approval: true` and `skills.write_approval: true` to review what it saves about the owner - Leave `approvals.cron_mode` and `approvals.unattended_mode` at `deny` for scheduled and webhook runs - Set `TELEGRAM_ALLOWED_USERS` or the matching allowlist per channel. Never set `GATEWAY_ALLOW_ALL_USERS=true` ## When it's done Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.