Hatchet

# Fix list: Hatchet

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/hatchet, the October 2026 research run, assessed 9 October 2026. Grade B, 68.3 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 Hatchet: 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. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No machine payment protocol (0). Per-unit prices published without a login, $30 per million runs and $5 per million events (20). The pay-as-you-go plan starts at $0 without a card (20). A person signs up in the browser for the cloud. Embedded mode needs no account but is local software for development (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.

## 2. Reliability, 66 out of 100, up to 6.8 more on the total

Why it scored 66: Graded as Hatchet Cloud, with the hosted lines. Status page at status.hatchet.run on Better Stack with three components (20). Its incident list, which reaches back to August 2026, shows one incident, about nine hours of elevated latency and delayed task processing on one shared shard in us-west-2 on 22 August 2026. Other shards and regions were unaffected, so we score it between minor and major (15 of 30). The pricing page gives plan quotas with numbers (2,000 runs a day without a card, worker slots, queue backlog) and the docs give a 4 MB payload limit, but no request rate for the API was found (8 of 15). The docs say which errors each SDK retries, the Go SDK honours `Retry-After` on 429, the OpenAPI file lists 429 on event creation, and idempotency keys are generally available (12 of 15). SLAs are named for the enterprise plan only and no SLA document is published (3 of 10). Event waits carry no beta label, the SDKs are 1.x except Ruby at 0.9.0, and the platform version is still 0.x (8 of 10).

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.

## 3. Security & auth, 62 out of 100, up to 6.7 more on the total

Why it scored 62: API tokens per tenant with a configurable expiry, created and revoked through the dashboard or API. No narrower scopes, and the docs say tokens are not restricted by role or payload visibility (20 of 30). A read-only VIEWER role, a per-member payload flag and deny-by-default profile grants for the MCP server, with no read-only token and no confirmation before `trigger_run` or `replay_run` (11 of 20). Event payloads from webhooks are untrusted content. Hatchet has payload validators, CEL filters and webhook checks by HMAC, API key or basic auth, and no prompt-injection guidance was found (8 of 15). Audit logs with actor, action, IP address and time on Business plans and above, kept 30 days. Event pushes are not among the documented actions (9 of 15). SECURITY.md with private GitHub advisories and security@hatchet.run, advisories published with CVE IDs, SOC 2 Type II, HIPAA and independent testing stated in the docs. No bug bounty (SECURITY.md says so), no security.txt, and the trust centre was not readable (14 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.

## 4. Agent ergonomics, 82 out of 100, up to 2.9 more on the total

Why it scored 82: Seven compact MCP tools, and list endpoints with limit and offset (23 of 25). Events list by key, workflow, status and search text (18 of 20). Errors return a code, field, description and docs link, though no list of codes was found (14 of 20). Idempotency keys by TTL or status, and scope with a lookback window against a race between push and wait. No MCP tool sets `readOnlyHint` or `destructiveHint` (12 of 20). A wait needs only an event key, and there are official SDKs in four languages (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. Transparency & trust, 71 out of 100, up to 2.5 more on the total

Made of editorial 69, provenance 72.

Why it scored 71: MIT (30). The privacy policy of 18 August 2026 gives retention periods (workflow logs up to 90 days, account data 12 months after closure) and the pricing page gives 1, 7 or 90 or more days by plan. No DPA or sub-processor list could be read, because the trust centre is script-drawn (18 of 30). No deprecation policy was found. The changelogs carry Removed sections, there are migration guides, and the cloud agreement gives 15 days' notice of price changes (8 of 20). CLI telemetry is disclosed with an opt-out, server analytics are settings, the privacy policy names PostHog and Google Ads without a full list, and regions are published (13 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: hatchet.run, no registry record we could read (0 of 15)
- Terms of service: read, states 3 of the 7 things a reader expects (6.6 of 10)
- security.txt: not found (0 of 10)

## 6. Schema & documentation, 85 out of 100, up to 2.4 more on the total

Why it scored 85: OpenAPI 3.1 file in the repository with 147 operations, and typed inputs on all seven MCP tools (25). llms.txt on the docs host (10). The docs say when to use a durable task and when a DAG, and warn about determinism. The MCP tool descriptions are one line each and do not say when not to use a tool (14 of 20). Typed SDK signatures with an optional payload validator and 56 enums in the API schemas, but event data and MCP run input are free-form objects (11 of 15). Code samples in four languages. The API file has few examples, and no rendered REST reference was found in the docs index (10 of 15). Dated changelogs for the platform and each SDK, and `/api/v1/stable` paths (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, 90 out of 100, up to 0.9 more on the total

Why it scored 90: Platform v0.110.19 tagged on 8 October 2026 (30). Tags on most working days in September and October, and five dated changelog entries between 25 August and 5 October (20). Issue reply times were not sampled. There is a Discord, a stale-issue workflow and a changelog (15 of 25). Current official SDKs, Python 1.41.1 on 24 September and Ruby 0.9.0 on 5 October 2026. The MCP server is not in the official registry (15). CI workflows for tests, each SDK, lint and an OSV scanner (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.

- Published 4 September 2026, -2. GHSA-992g-9cr3-vm5x (CVE-2026-88978), Moderate, CVSS 3.1 AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:N/A:N. The `DurableTask` worker status path looked up durable event log entries by a caller-supplied task UUID with no tenant filter, so a holder of any tenant's token who knew another tenant's task UUID could read its entries. The advisory names Hatchet Cloud as affected and calls exploitation very unlikely because the IDs are random UUIDs. Fixed in 0.106.1. It is the event log this listing is graded on, so 2 points for a fixed path. https://github.com/hatchet-dev/hatchet/security/advisories/GHSA-992g-9cr3-vm5x (read through https://api.osv.dev/v1/vulns/GHSA-992g-9cr3-vm5x)
- Not deducted. GHSA-hf2m-86fv-rjw5, fixed in v0.110.5 on 5 October 2026, enforced payload visibility restrictions on endpoints that had returned payload data. The OSV mirror answered 404 for it, so its severity was not read. https://github.com/hatchet-dev/hatchet/blob/main/CHANGELOG.md

## 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: the trust centre at trust.hatchet.run is script-drawn, so the SOC 2 report, DPA, sub-processor list and penetration test summaries were not read.
- unchecked: GHSA-hf2m-86fv-rjw5. The OSV mirror answered 404 and GitHub's advisory page was not tried, so its severity and scope are known only from the changelog line.
- unchecked: issue and pull request reply times, GitHub stars and package download counts.
- unchecked: the domain's registration date, and whether the MCP server is in the official registry (no server.json is in the repository).
- unchecked: the status page history before August 2026, and whether the incident list is complete for the 90 days.
- The standard terms (Common Paper Cloud Service Agreement 2.1), as our reader summarised them, forbid security or vulnerability tests and allow aggregated usage data and customer content to be used for machine learning. Whether Hatchet's cover page changes either was not established. Recorded as a fact with no deduction, and it matters before any probe is run.
- The audit log page says Business plans and above, while the pricing page lists audit logs only under the enterprise plan and shows no Business plan.
- The pricing page lists 1 million runs a month as included on the plan without a card and also a limit of 2,000 runs a day. How the two interact was not established.
- The lead named an MCP server among the interfaces. It exists, but it is a local operations tool for coding agents and cannot push the event that resumes a wait.
- api.osv.dev answered 404 for robots.txt, so it publishes no rules. status.hatchet.run answered 200 with an empty file, and trust.hatchet.run answered its app page.

## Weaknesses

- No approver inbox, Slack or email channel, routing or reminder. The owner builds the request and the way the answer is pushed
- An event wait has no timeout of its own. A deadline needs a sleep condition in the same or group
- API tokens are scoped to a tenant with an expiry, with no narrower scopes, and the role and payload restrictions do not apply to them
- Audit logs are for Business plans and above, kept 30 days, and the documented actions do not include event pushes
- The seven MCP tools trigger, inspect and replay runs. None pushes an event, and none sets a read-only or destructive annotation
- The standard terms the cloud agreement incorporates forbid security or vulnerability tests of the product. This matters before any probe is run

## 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.

- Put the approval wait in a durable task, and keep API calls, database reads and random values in child tasks, because the code between checkpoints is replayed
- Add a sleep condition in the same or group as the event condition to set a deadline, and treat the sleep firing first as a timeout
- Give the event a scope and the wait a lookback window when the answer can arrive before the wait starts. Both must be set together
- Filter with a CEL expression such as `input.user_id == '1234'` so one approval key cannot resume another request's task
- Push the answer with the SDK's event push or `POST /api/v1/tenants/{tenant}/events`. The MCP server has no tool for it
- Grant the MCP server one profile with `hatchet mcp auth --grant <profile>`. It refuses profiles that were not granted

## 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.