ZeroClaw

# Fix list: ZeroClaw

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/zeroclaw, the October 2026 research run, assessed 12 October 2026. Grade BB, 70.2 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 ZeroClaw: 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, 74 out of 100, up to 5.2 more on the total

Why it scored 74: Read with the local-software lines, as software the owner runs. An install script that fetches a prebuilt binary, Homebrew, a ghcr.io image and source builds on Rust 1.96, with setup pages for Linux, macOS, Windows, FreeBSD, NixOS and Docker (20). The CI workflow runs on every push to master, and the three latest runs, on 11 October 2026, passed (25). 576 open issues and 384 open pull requests. The 30 newest bug reports carry labels and at least one comment within days, and several P1 reports from that week are open, among them a Telegram listener that can hang (#11608) and a memory leak in config paths (#11614) (15 of 25). Release notes carry a Breaking Changes section, but 0.8.4 and 0.8.5 are patch-level versions that removed settings and skill sources (10 of 15). Version 0.8.5, with v1.0 shown as planned on the home page (4 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 (0). Free under MIT or Apache-2.0 with nothing to buy for the assistant, so 20, 20 and 20 on the last three lines. It runs local models with no sign-up. ZeroRouter, the vendor's optional hosted model gateway, was not graded and its prices were not read.

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, 72 out of 100, up to 4.6 more on the total

Why it scored 72: Harness reading. MCP servers reach an agent only through its `mcp_bundles`, skills can load on demand in compact mode, and most built-in integrations are off until configured, though the built-in tool list is long (18 of 25). Daily and monthly cost budgets, history trimming and a 1 MB cap on Google Workspace output (15 of 20). A blocked call returns a tool error that names the policy, and the docs describe each failure path (15 of 20). Provider fallback chains, resumable procedure runs and approval timeouts that deny, with no idempotency guidance for sends (14 of 20). One binary and a guided quickstart, and `zeroclaw agent -a <alias> -m` for one-shot runs, but a working config needs four sections and there is no SDK (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.

## 4. Security & auth, 76 out of 100, up to 4.2 more on the total

Why it scored 76: Harness reading, with 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 docs state no telemetry, we found no vendor host in the source apart from the optional ZeroRouter provider, secrets are encrypted with ChaCha20-Poly1305 beside a 0600 key file on the same disk, and the gateway binds 127.0.0.1 with pairing on. Web search and page fetches run without approval (27). Supervised mode is the default, `always_ask` outranks full autonomy, prompts time out to a denial, and an OS sandbox is auto-detected, but sandboxed tools keep network egress, unattended runs skip the shell prompt and rely on the command policy, and an open report says `firejail_args` is never applied (#11594) (17). A guard scans for injection patterns with a default action of warn, trigger payloads are framed as untrusted, a leak detector redacts credentials in replies by default, and an empty peer set denies all senders. No guidance page on injection through mail or web content was found (10). Approval events and a runtime trace are logged, tool receipts are opt-in and not durable, and the docs state command execution is not audited (8). SECURITY.md with private reporting and response times, one advisory published, CodeQL, cargo-deny, a daily audit and build attestations. No bounty and no security.txt, and SECURITY.md and the README name different contacts (14).

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.

## 5. Transparency & trust, 64 out of 100, up to 3.2 more on the total

Made of editorial 74, provenance 53.

Why it scored 64: MIT or Apache-2.0, with a CLA that leaves copyright with contributors (30). The docs map every state store, say what is sent to a model provider and list what to back up, and the site says there is no ZeroClaw cloud in between. There is no privacy policy or terms on zeroclaw.com, and no data statement was read for the hosted ZeroRouter beta that the `zerorouter` slot defaults to (16 of 30). Release notes name removed and deprecated settings, and SECURITY.md says only the latest minor line gets fixes. No deprecation policy with notice periods was found (10 of 20). The docs state there is no telemetry, and our source search found no vendor host beyond ZeroRouter. We did not audit every outbound call (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: zeroclaw.com, registered 2026-03-02 (under a year) (0 of 15)
- Status page: not found (0 of 10)
- security.txt: not found (0 of 10)

## 6. Schema & documentation, 82 out of 100, up to 2.9 more on the total

Why it scored 82: Harness reading of the six lines, as the Goose dossier used. A running gateway serves an OpenAPI document at `/api/openapi.json` for what the docs call the currently documented subset, and `zeroclaw config schema` prints the full config schema, from which the published config reference is generated. We read neither from a running copy (18 of 25). An llms.txt on www.zeroclaw.com, and the docs are Markdown in the repository (10). The autonomy, sandboxing and YOLO pages say what each setting allows, when to use it and when not to (16 of 20). Typed TOML with enumerated levels, validation warnings and rejection of unknown values such as `read_only` (12 of 15). Worked config examples on most pages and a troubleshooting page, with no single error reference (12 of 15). Dated release notes, a blog post a release and a versioned config schema with automatic migration (14 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.

## 7. Maintenance & community, 80 out of 100, up to 1.8 more on the total

Why it scored 80: v0.8.5 was released on 5 September 2026, 37 days before this check (20 of 30). Three releases in the 90 days to 12 October, v0.8.3 on 16 July, v0.8.4 on 2 August and v0.8.5 on 5 September (20). Master had 200 commits between 25 September and 11 October, and bug reports from the last week carry priority labels, comments and linked fix pull requests, against 576 open issues and 384 open pull requests (20 of 25). Not an MCP server and no SDK. Release workflows publish the binary, a ghcr.io image and crates, AUR and Scoop packages, which we read as workflow files and did not confirm on the registries (11 of 15). cargo-deny, a daily advisory audit, a monthly outdated-dependency job, CodeQL, Trivy and SBOM workflows, with CI passing (9 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.

- 2026-09-05. GHSA-93f6-34w8-5g98 (high), an unvalidated `wasm_path` let `zeroclaw plugin install` write an attacker-controlled file outside the plugins directory in 0.8.4 and earlier. Fixed in 0.8.5 and published the same day, inside six months, -2. https://github.com/zeroclaw-labs/zeroclaw/security/advisories/GHSA-93f6-34w8-5g98
- 2026-04-01. The project lead wrote that a maintainer's AI automation merged pull requests that should not have gone through and that GitHub took enforcement action on the organisation. Disclosed, the maintainer removed and merge access tightened, over six months old, -1. https://www.zeroclaw.com/blog/community-and-stability

## 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`. ZeroClaw is a self-hosted agent program, and harness is the closest kind the briefs define. Reliability was read with the local-software lines and Payments with the self-hosted rule.
- The lead gave the licence as Apache-2.0. The repository is MIT or Apache-2.0 at the user's choice, and GitHub's API reports only Apache-2.0.
- unchecked: nothing was installed or run, so the approval prompts, the sandbox, the OpenAPI document and the quickstart defaults were read from docs and source only.
- unchecked: zerorouter.ai, the vendor's hosted beta model gateway, was not opened. Its prices, terms and data handling are unknown, as is whether quickstart preselects it.
- unchecked: the crates.io, Homebrew, AUR and Scoop packages were not looked up. Their existence rests on the docs and the release workflow files.
- unchecked: the advisories page on github.com answered 504 twice. The one published advisory was read from the GitHub API, and the full advisory text was not read.
- unchecked: the telemetry statement was checked by a search of the source for vendor hosts, not by an audit of every outbound call.
- The 1 April 2026 merge incident took 1 point. Whether any unreviewed merge reached a tagged release was not established.
- The site heading says shell access you can audit, while the docs say command execution is not audited and receipts are opt-in. Recorded as a weakness with no deduction.
- No company form or address for ZeroClaw Labs was found, and the site publishes no terms or privacy policy.

## Weaknesses

- The docs state that command execution is not audited. `[security.audit]` records certificate events only, and tool receipts are opt-in and not durable
- No privacy policy, terms or security.txt on zeroclaw.com, and the home page links none
- `web_fetch`, `web_search_tool` and `browser_open` run without approval by default, and sandboxed tools keep full network egress
- GHSA-93f6-34w8-5g98 (high) let a plugin install write outside the plugins directory in 0.8.4 and earlier, fixed in 0.8.5 on 5 September 2026
- Version 0.8.5 is pre-1.0, patch releases carry breaking changes, and 576 issues and 384 pull requests were open on 12 October 2026

## 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.

- Keep `[risk_profiles.<alias>] level = "supervised"`. `full` runs every tool not listed in `always_ask`, and the YOLO preset removes the sandbox and path rules too
- Add sending and account tools such as `send_via`, `google_workspace` and `http_request` to `always_ask` so a session-level Always grant cannot skip the prompt
- List senders in `[peer_groups.<name>] external_peers`. `["*"]` accepts anyone on that channel
- Set `[web_fetch].allowed_domains` and `[http_request].allowed_domains`. Fetches run unprompted and the OS sandbox does not block egress
- Run one-shot tasks with `zeroclaw agent -a <alias> -m "<task>"`. A cron or webhook turn has no prompt, so shell calls fall to the command policy

## 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.