# Fix list: Docker Agent From Anchor Terminal's listing at https://www.anchorterminal.com/tools/docker-agent, the October 2026 research run, assessed 8 October 2026. Grade BB, 76.5 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 Docker 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. Security & auth, 70 out of 100, up to 5.3 more on the total Why it scored 70: Framework reading. 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. Usage telemetry is on by default, with a notice on first start and `TELEMETRY_ENABLED=false` to stop it. The docs say command events include positional arguments, which for `run` and `exec` can include prompts, and error text (12). Four safety modes with allow, ask and deny rules, a default that asks for anything not read-only, `--exec` rejecting unapproved calls, `--sandbox` for a Docker Sandboxes VM, and HTTP MCP defaulting to loopback and `restricted`. A run's `--listen` control plane has no built-in authentication (19). The docs treat skill files and third-party configs as untrusted and restrict URI schemes against prompt injection, with no dedicated guidance page (10). Hooks for audit logging and approval decisions, opt-in OpenTelemetry traces, NDJSON transcripts and a session database (14). SECURITY.md with a 72-hour acknowledgement, a valid security.txt on docker.com, two advisories published on GitHub with fixed versions, and CodeQL and zizmor in CI. No paid bounty, and we saw no CVE identifiers on the advisories (15). SOC 2 isn't scored for a framework. 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. ## 2. Payments & pricing, 60 out of 100, up to 5 more on the total Why it scored 60: No payment protocol (0). Read with the self-hosted rule. The CLI is free and Apache-2.0 with nothing to buy, so 20, 20 and 20 on the last three lines. It installs without an account and runs local models through Docker Model Runner. Cloud sandboxes need a Docker login and weren't priced in this check. 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. Agent ergonomics, 81 out of 100, up to 3.1 more on the total Why it scored 81: Framework reading, scored on what an agent with MCP needs. The README's agent with one MCP server is eight lines of YAML and no code, with deferred toolsets and a code mode (23). Budgets cap cost, tokens and time, tool results are bounded to 50 KiB since 1.148.0, and context is compacted, but `max_iterations` is unlimited and no budget applies unless declared (16). `--json` writes one NDJSON event per message, tool call and error, and the API documents 404, 409, 413 and 502 responses. No exit-code reference found for `run --exec` (14). Sessions resume from SQLite, the event stream replays from a sequence number, follow-ups accept an `Idempotency-Key`, and read-only and destructive tool annotations drive the safety modes (18). One binary and a model key or a local model are enough to start. The library is Go only (10). 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. ## 4. Reliability, 88 out of 100, up to 2.4 more on the total Why it scored 88: Read with the local-software lines, because the owner runs it. Binaries for macOS, Linux and Windows on amd64 and arm64 on every GitHub release, a Homebrew formula at 1.149.0, a Docker Hub image, and Docker Desktop 4.63 and later bundle it. Go 1.27 is the stated minimum for source builds (20). The ci workflow runs on every push to main with about 1,350 test files and an e2e directory. Nine of the last ten push runs passed, including the newest on 8 October, with one failure that morning (25). 25 open issues and 17 open pull requests, most labelled and answered, with 27 issues closed and 32 opened in the 30 days to 8 October. Open reports include a race in `transfer_task` (#4156), a hand-off dropped silently (#4242) and `edit_file` editing the first of several matches (#3929) (20). The changelog has Breaking Changes sections, but those changes ship in minor releases (1.114.0, 1.133.0 and 1.137.0 since July) and the major version has never moved (8). 1.149.0, past 1.0 (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. ## 5. Transparency & trust, 80 out of 100, up to 1.8 more on the total Made of editorial 75, provenance 84. Why it scored 80: Apache-2.0 (30). The telemetry page lists what's collected and warns that arguments and error text can carry secrets or personal data. It doesn't name the endpoint (api.docker.com, per the source), link Docker's privacy policy or give a retention period (15). Breaking changes are dated in the changelog and one renamed variable kept a fallback for one release, with no written deprecation policy, and the changelog says frozen config schema versions aren't maintained long-term (10). Telemetry is disclosed in the README, the docs and a first-start notice, with an opt-out variable (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): - Terms of service: read, states 7 of the 7 things a reader expects, and has 2 clauses that cost points (6 of 10) - Status page: not found (0 of 10) ## 6. Schema & documentation, 90 out of 100, up to 1.6 more on the total Why it scored 90: Framework reading. agent-schema.json in the repository is a JSON Schema for the whole config, and the chat server serves its OpenAPI document at /openapi.json. We found no OpenAPI document for the native /api server and didn't check the Go package reference (22). llms.txt on docker.github.io/docker-agent and on docs.docker.com (10). The docs say when to use the API server and when the chat server, compare six ways to pass large input, and state that permissions aren't a security boundary (17). Configs are parsed strictly, unknown fields fail to load with a hint, and safety fields accept four named modes only (13). An examples directory, a troubleshooting page of common errors and documented HTTP status codes for the API. We found no exit-code reference for `run --exec` (13). A dated CHANGELOG.md and a numbered config schema, now version 16, with migration on load (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. ## 7. Maintenance & community, 96 out of 100, up to 0.4 more on the total Why it scored 96: v1.149.0 published on 2026-10-07 (30). 45 tagged releases in the 90 days to 8 October, v1.111.0 to v1.149.0 (20). 25 open issues, most labelled by area and answered, 27 closed in the last 30 days and 17 open pull requests. We didn't measure reply times (21). The Homebrew formula was at 1.149.0 and the Docker Hub image was updated on 8 October, and the Go module ships from the same tags (15). Dependabot, CodeQL, zizmor and actions pinned to commit SHAs, with CI passing on main (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. - 2026-09-01. GHSA-8f6v-26r8-3r4h (high). `read_skill` was treated as read-only and could be auto-approved, yet it ran shell commands embedded in local skill files, so opening a repository that ships skills could run commands without a prompt. Affects 1.32.5 up to 1.130.0. Fixed in 1.130.0 and published, inside six months, -2. https://github.com/docker/docker-agent/security/advisories/GHSA-8f6v-26r8-3r4h - 2026-08-20. GHSA-f2g7-cxrq-x54m (high). `docker agent serve a2a` created unattended sessions with tool approval disabled, so a client that could reach the endpoint could run tools without approval. The listener binds to loopback by default. Affects 1.8.2 up to 1.126.0. Fixed in 1.126.0 and published, inside six months, -2. https://github.com/docker/docker-agent/security/advisories/GHSA-f2g7-cxrq-x54m ## 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. - Unchecked: CVSS scores and CVE identifiers for the two advisories. The GitHub API rate limit stopped the detail requests, and the advisory pages we loaded showed no CVE identifier - Unchecked: reply times on issues and pull requests. Only counts, labels and comment numbers were read - Unchecked: the Go package reference on pkg.go.dev - Unchecked: pricing and terms for cloud sandboxes (`--cloud`), which need a Docker login - The first release date isn't established. The clone was shallow, and the oldest tags date from July 2025 under the cagent name - Neither the README nor the telemetry page says which Docker terms govern the open-source CLI, or how long telemetry is kept - The category is arguable. Docker Agent is a finished CLI with a terminal UI as well as a Go library, so agent-harnesses would also fit. We kept the scout's frameworks - No OpenAPI document was found for the native /api server, and no exit-code reference for `run --exec` ## Weaknesses - Usage telemetry is on by default and command events can include prompts passed as arguments - Two high-severity advisories in 2026, each letting tools or shell commands run without approval, both fixed - Breaking changes ship in minor releases, three of them between July and September 2026 - `max_iterations` is unlimited and no budget is set unless the config declares one - The `--listen` control plane has no built-in authentication, and A2A support is described as early ## 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 `TELEMETRY_ENABLED=false` before `run` or `exec` with a prompt on the command line. Telemetry is on by default and sends positional arguments - For unattended runs pass `--safety restricted` with an allow-list, or `--sandbox`. Without a policy, `--exec` rejects every tool call that needs approval - Set `max_iterations` and a `budget` block in the config. Both are unlimited by default - Run 1.130.0 or later. Earlier versions ran shell commands embedded in local skills, and A2A sessions before 1.126.0 skipped tool approval - Review `runtime.safety` in any config pulled from a registry or URL. An author default of `autonomous` runs every tool call unprompted ## 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.