# Fix list: AgentGate From Anchor Terminal's listing at https://www.anchorterminal.com/tools/agentgate, the October 2026 research run, assessed 15 October 2026. Grade D, 48.5 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 AgentGate: 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, 44 out of 100, up to 11.2 more on the total Why it scored 44: Read as local software, since the owner runs it. No package on npm, PyPI or a container registry. Install is a git clone and `docker compose up`, with Python 3.11+, Node 22+ and Docker Compose v2 stated (10 of 20). A public CI workflow runs ESLint, Ruff, 25 Vitest tests and the decision service's Pytest file. The last six runs on main passed by 15 October 2026, after two failures on 13 and 14 October. The Docker job only validates the compose file (20 of 25). No outside issues yet. The maintainer filed 20 issues as a self-audit, and six stay open, including the MCP wrapper deadlock and the wildcard-rule bypass, which a commit of 13 October says it fixed (12 of 25). No tags, and CHANGELOG.md holds one Unreleased line (2 of 15). Version 0.1.0, three days old (0 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. Security & auth, 49 out of 100, up to 8.9 more on the total Why it scored 49: Read with the API checklist, as the gateway is itself the security layer. Credential model. Agent endpoints are open unless `AG_AGENT_KEY` is set, and compose publishes the proxy on 0.0.0.0:8080 without it. The dashboard key falls back to `dev-key-123`, both keys are single static secrets compared in constant time, and the dashboard key is accepted as `?token=`, which the API reference documents for the SSE stream (15 less 10, 5 of 30). Approvals are the product's core. Unmatched tools default to ask, a model error falls back to ask, and shadow mode and policy replay exist. The checkpoint revert endpoint, which runs git on the proxy's workspace, has no auth check (15 of 20). Injection posture. Tool results are wrapped in escaped `` sentinels, but the proxy marks every call `trusted` and sends no history unless the caller passes `history_override`, and no detection accuracy is published (9 of 15). Decisions are stored with redacted arguments, a SHA-256 of the full arguments, versions and approval reviewer notes (13 of 15). SECURITY.md with private reporting and a 48-hour acknowledgement target, and security issues filed and fixed in public. Issue 6 was closed as resolved on 15 October 2026 while the default key and query-string token it named remain in the code. No security.txt, bounty or certification (7 of 20). 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. ## 3. Schema & documentation, 47 out of 100, up to 8.6 more on the total Why it scored 47: `docs/openapi.yaml` covers seven paths. It lists `GET /v1/approvals`, which the code does not serve, leaves out the checkpoint and shadow endpoints, and types `verdict` as a free string (12 of 25). No llms.txt. Markdown docs in the repository and the GitHub wiki (5 of 10). The API reference and wiki state each endpoint's purpose and the policy DSL is specified, with no guidance on when not to use it (10 of 20). Requests are validated with zod and required fields are marked. `action.arguments` is free-form by nature, and verdicts and reason codes are not enumerated in the contract (8 of 15). curl, Python and TypeScript examples. Error responses are not documented, and the docs disagree with the code on details such as a `201` for sessions where the code returns 200 and the approval response shape (7 of 15). Paths under `/v1`, with an empty changelog (5 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. ## 4. Agent ergonomics, 53 out of 100, up to 7.6 more on the total Why it scored 53: The assess response is small, a verdict, reason codes, four head probabilities, versions and latency (20 of 25). `/v1/decisions` has `limit`, a keyset `cursor` and a `status` filter, behind the dashboard key (12 of 20). Reason codes such as `untrusted_prompt_influence` and `model_error` tell an agent why. On `ask` the agent gets a `decision_id` but cannot read the outcome without the dashboard key, the MCP wrapper returns `allow_with_checkpoint` as an unhandled-verdict error, and HTTP error bodies vary in shape (8 of 20). No idempotency key. A repeated assess writes a new decision and a new approval request (6 of 20). Two required fields per call and fail-closed defaults. No SDK, only snippets in Python and TypeScript and the MCP wrapper (7 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. ## 5. Maintenance & community, 28 out of 100, up to 6.3 more on the total Why it scored 28: No tagged release or package. The code and the Hugging Face weights were last changed on 15 October 2026, so half credit because nothing is versioned as a release (15 of 30). No releases or dated changelog entries (0 of 20). All 20 issues are the maintainer's own, and 14 were closed within two days, each with one comment. The one we read, on issue 6, names the fixing commit. No outside issue or pull request has been answered yet, and the project has one fork and no stars (8 of 25). Not in the official MCP registry and no SDK package (0 of 15). CI passes on main, with older major versions of Fastify 4, Vitest 0.34 and ESLint 8 in the proxy (5 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. ## 6. Payments & pricing, 60 out of 100, up to 5 more on the total Why it scored 60: Self-hosted rule. Free under MIT with nothing to buy (20). No card or account (20). Clone and run with no signup, and the model weights on Hugging Face are not gated (20). No payment protocol (0). 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. ## 7. Transparency & trust, 56 out of 100, up to 3.9 more on the total Made of editorial 59, provenance 53. Why it scored 56: MIT for the code and for the model weights (30). Self-hosted, so data stays in the owner's Postgres. The docs describe argument redaction and hashing, and arguments are redacted before the classifier unless a tool sets `send_raw_to_model`. No page sets out retention, and the README's "$0, self-hosted" claim agrees with the code (15 of 30). SECURITY.md lists 0.1.x as supported. No deprecation policy (2 of 20). No telemetry code found in AgentGate's source, and no statement either way. The dashboard loads Google Fonts and the decision service downloads weights from Hugging Face on start, which the docs mention only as model loading (12 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: , no registry record we could read (0 of 15) - Status page: not found (0 of 10) - security.txt: not found (0 of 10) ## 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. - Judgement call on category. Listed in guardrails, not agent-auth. Agent-auth covers OAuth and delegated access to third-party apps, and AgentGate grants no access. It scores each tool call for injection and blast radius and applies a policy, which is the guardrails category's injection and policy work. - The lead matched. The repository's second commit on 12 October 2026 is "feat: initial open-source release of AgentGate security checkpoint", and the wiki's Client Integration Guide covers MCP and REST. Several unrelated GitHub projects created in October 2026 share the name, and the Linux Foundation's agentgateway is a different product. - Possible deduction left open. Issue 6 (insecure default secret and query-string token) was closed on 15 October 2026 as "Resolved in commit f84960d", but `dev-key-123` and `?token=` remain in `proxy/src/api/index.ts` at commit 104ac0b. The README's verification output ends "System is secure". We did not deduct, since the issue is the maintainer's own and partly fixed (constant-time comparison was added). - `firstReleased` is the commit date, 12 October 2026, as there is no tag, release or package. `lastRelease` is null for the same reason. - We did not install or run the stack, the model or the demo Space. Behaviour claims come from the source at commit 104ac0b, the docs and the model card. - `.env.example` sets `AGENTGATE_AUTH_TOKEN`, which the code does not read. The variables the code reads are `AG_AGENT_KEY` and `AG_DASHBOARD_KEY`. - unchecked: whether Unsloth or the Hugging Face libraries send usage data when the decision service starts. We read AgentGate's own source only. ## Weaknesses - Agent endpoints need no key unless `AG_AGENT_KEY` is set, the proxy binds 0.0.0.0:8080, and the dashboard key defaults to `dev-key-123` - Checkpoint revert, which runs git on the proxy's workspace, has no auth check in the source of 15 October 2026 - No tagged release, an empty changelog and version 0.1.0, three days after the first commit - The MCP wrapper treats `allow_with_checkpoint` as an error and does not send the agent key to `/v1/assess` - No published accuracy or false-positive rate for the classifier, trained in 60 steps per its model card ## 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. - Create a session with the user's real goal first. The goal-alignment head scores every later call against it - Pass earlier tool results in `history_override` with `untrusted: true` for web or email content, or the injection head sees only the action - On `ask`, stop and tell the user the `decision_id`. Polling `/v1/decisions/:id` needs the dashboard key - Treat `allow_with_checkpoint` as allow and keep the `checkpoint_id` for an undo - Expect HTTP 404 with `not_found` for an unknown session and 401 when the owner has set an agent key ## 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.