Warp (Warp Agent CLI and the Oz Agent API)

# Fix list: Warp (Warp Agent CLI and the Oz Agent API)

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/warp, the October 2026 research run, assessed 18 October 2026. Grade C, 59.8 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 Warp (Warp Agent CLI and the Oz Agent API): 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, 20 out of 100, up to 10 more on the total

Why it scored 20: Harness reading of the published rubric, scored on the Warp plan the agent needs. No payment protocol (0). Plan prices are public, and Warp-provided inference is billed at public API rates on paid plans, but the agent-hour platform rate and compute rate for cloud runs were not on the pricing page (10). Free is $0 and includes Warp Agent CLI access, but no bundled Warp inference, so the agent runs only with purchased usage or your own provider key. Whether sign-up asks for a card was not tested (10). A person signs in through a browser or device code, and API keys are created by hand in the Oz web app or the Warp app (0). 0 + 10 + 10 + 0 = 20.

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, 52 out of 100, up to 9.6 more on the total

Why it scored 52: Local-package reading, as for the other harnesses, although the FAQ says the agent harness runs on Warp's servers. An install script for macOS and Linux and a PowerShell one for Windows, with supported systems stated (macOS, Linux x64 and Arm64, Windows x64 and Arm64), and a `warp-agent-cli` cask in Homebrew core that carries a SHA-256. The script itself checks no checksum (18). The repository's Warp CI workflow runs cargo nextest, doc tests, clippy and release-compilation checks on every pull request to master. The repository has 23 workflow files. Whether CI passes on the default branch could not be read (15). Per our GitHub record (read 18 October 2026), the tracker had 3.9k open issues against 7.1k closed, and 1.5k open pull requests against 3.8k closed. How many open issues are crashes or regressions, and how fast they get replies, could not be read. The weekly changelog records crash fixes in most releases, such as two crash fixes on 7 October 2026 (10). Weekly dated releases with versions such as v0.2026.10.14.08.27 and no breaking-change heading. The binary rename from warp-tui to warp on 23 July 2026 sat under Improvements (4). Version 0.x with date-based numbers. A stable channel is named and the CLI pages carry no beta label (5). Not scored on this reading. status.warp.dev has components for the agent harness and cloud execution, and logged scheduled cloud agents unavailable for about 2.5 hours on 16 October and session sharing degraded for about 4.75 hours on 30 September 2026. The docs say Warp has no formal SLA. 18 + 15 + 10 + 4 + 5 = 52.

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, 52 out of 100, up to 8.4 more on the total

Why it scored 52: Framework reading (telemetry defaults, approvals, guardrails, sandboxing), five lines. The FAQ says the built-in harness runs server-side, so prompts, file contents and command output pass through Warp. The privacy page says AI interactions and console inputs may be collected unless Help improve Warp is off, and the security overview says data collection is on by default for Free and paid teams and off on Business and Enterprise. The opt-out is documented in the desktop app's settings, not for the CLI. Personal and agent keys can expire after 1, 30 or 90 days, and provider keys stay in the device's secure storage (14). Per-action permissions with `always_ask`, `agent_decides` and `always_allow`, edits always shown as a diff under `agent_decides`, a default denylist and child agents asking first. Auto-approve bypasses the user's denylist by default, and no local OS sandbox is documented. Cloud runs use a Docker sandbox as a non-root user with passwordless sudo (11). No prompt-injection guidance was found in the CLI, platform or factory pages read, and secret redaction is off by default (4). Cloud runs record transcripts, tool calls and logs, the API returns timelines and conversations, the app has a Network Log, and the Enterprise Analytics API can filter tool calls for audit (11). SECURITY.md names security@warp.dev and private GitHub advisories, SOC 2 Type II is stated with a trust centre and a subprocessor list. No security.txt and no bug bounty were found, and the GitHub advisory list was not read (12). 14 + 11 + 4 + 11 + 12 = 52.

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, 66 out of 100, up to 5.5 more on the total

Why it scored 66: Harness reading, adapted to a harness driven by a pipeline. MCP servers are attached per run with `--mcp` or the `mcp_servers` config, and profiles gate MCP calls, but no per-tool disabling or deferred tool loading was found (12). The API lists runs with `limit`, `cursor`, `sort_by`, `state`, `task_status` and `name`, and returns a normalised conversation and a timeline per run. The `warp` CLI has no print mode or JSON output and the Oz CLI's run output is described only as printed tool calls (12). RFC 7807 errors with codes such as `insufficient_credits` and `budget_exceeded`, a `retryable` flag, and since 7 October 2026 unattended CLI runs end as blocked instead of hanging when the agent asks for input. No CLI exit codes are documented (16). Follow-ups, interrupts and cancels by run or conversation ID, `conversation_id` to continue a run, and `warp --resume`. No idempotency key on run creation (14). Python and TypeScript SDKs with retries and timeouts, but a cloud run needs an environment ID and a Warp account with usage, and the `warp` CLI itself can't be driven headless (12). 12 + 12 + 16 + 14 + 12 = 66.

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, 72 out of 100, up to 2.5 more on the total

Made of editorial 57, provenance 86.

Why it scored 72: The client, including the `warp` and `oz` binaries, is AGPL-3.0, an OSI licence, with the UI crates under MIT. The built-in agent harness, the server and Oz orchestration are proprietary, so the part that runs the agent loop is closed. Judgement call (20). The privacy page says Business and Enterprise are covered by zero data retention, the security overview says ZDR applies across all plans and elsewhere mentions it for Enterprise teams, and the legal security page says it is available on the Enterprise plan. The privacy notice gives criteria rather than retention periods. A subprocessor list and a DPA are linked (15). The Oz CLI carries a legacy notice with a promise of guidance before any support change, and the changelog dates renames such as warp-tui to warp, but there is no written deprecation policy or notice period (8). An exhaustive telemetry event table, Rudderstack and Sentry named, and an opt-out in the app's privacy settings. The CLI pages don't describe what the CLI sends or how to switch it off (14). 20 + 15 + 8 + 14 = 57.

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: warp.dev, registered 2020-08-24 (6 years) (11 of 15)
- security.txt: not found (0 of 10)

## 6. Schema & documentation, 89 out of 100, up to 1.8 more on the total

Why it scored 89: API reading for the Warp Platform API, with the CLI docs read as a framework. docs.warp.dev/openapi.json is OpenAPI 3.0.0 with 293 component schemas and bearer authentication (25). llms.txt with per-section Markdown bundles, including one for the Warp Agent CLI and one for the Automation Platform (10). The run request schema says when each field applies (prompt unless a skill is given, `on_behalf_of` only for agent keys on team runs), and the CLI permissions page states each default and when auto-approve bypasses the denylist (15). Run modes, states and permission values are enumerated, and request bodies are typed objects (13). Curl and SDK examples throughout, and an error catalogue of RFC 7807 bodies with a machine-readable code, `retryable` flag and trace ID per error (14). The API sits under /api/v1 with info version 1.0.0, and the weekly changelog has an Oz updates section, but there is no API-only changelog (12). 25 + 10 + 15 + 13 + 14 + 12 = 89.

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, 84 out of 100, up to 1.4 more on the total

Why it scored 84: Release 2026.10.15 (v0.2026.10.14.08.27), also the version the installer served on 18 October 2026 (30). 15 dated releases between 23 July and 15 October 2026 (20). Issues and pull requests are public on GitHub, with 3.9k open and 7.1k closed issues, 1.5k open and 3.8k closed pull requests, and 925 commits in the 90 days to 18 October 2026, per our GitHub record. The FAQ describes an agent first review then a staff review, with @oss-maintainers for escalation and a Slack community, and the changelog credits outside contributors each week. Maintainer reply times are still unread (12). Official Python and TypeScript SDKs, released 6 October 2026 as 1.0.0 and 2.0.0 (15). CI on every pull request and a checksummed Homebrew cask, but the install script verifies nothing (7). 30 + 20 + 12 + 15 + 7 = 84.

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.

- 2026-05-06. The release of 6 May 2026 fixed a command injection vulnerability when opening files in external editors on Linux. It is recorded as one changelog line with no advisory, CVE or affected versions that we found, and api.osv.dev listed no advisories for the project's packages on 18 October 2026 (our GitHub record). Fixed, so 1 point. Judgement call. Dependency updates for GHSA-5pmw-9j92-3c4c (15 April) and CVE-2026-50185 (31 July) were not counted (https://docs.warp.dev/changelog/2026/)

## 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: CI run results and maintainer response times for warpdotdev/warp. Issue and pull request counts, commits and OSV advisories (none found) come from our GitHub record, which holds no issue text, so the share of open crash reports is unknown
- unchecked: GitHub stars (the badge answered invalid), so `githubStars` is null. The TypeScript SDK had 413 npm downloads in the week to 16 October 2026, and the warp-agent-cli cask 92 Homebrew installs in 30 days
- The newest GitHub release is v0.2026.06.03.09.49.stable_00 from June 2026, and the six tags in the last 90 days are backup and sync tags, while the changelog and the installer show weekly releases up to 15 October 2026. Release dates here come from the changelog
- unchecked: the trust centre at trust.warp.dev and the full subprocessor list and DPA
- The agent-hour and compute rates for cloud runs were not found on the pricing page, which the platform usage page points to
- Whether the Free plan sign-up asks for a card was not tested
- The docs disagree on where zero data retention applies (all plans, Business and Enterprise, or Enterprise only)
- What the `warp` CLI sends as telemetry, and how to switch it off without the desktop app, is not documented
- Whether the `oz agent run` output has a machine-readable format or documented exit codes was not found
- No first-hand date for the first Warp Agent CLI release was found. The changelog first names the TUI on 23 July 2026, and `firstReleased` is the Oz launch of 10 February 2026, which brought the CLI and API for cloud agents
- The terms forbid automated access to the Services by robot or scraper. Recorded as a fact with no deduction. It matters before any probe is run
- The lead was right on the three surfaces, the licence split, the legacy notice and the latest release. The OpenAPI file is titled Warp Platform API, not Oz Agent API, and the `warp` CLI has no headless mode
- robots.txt answers. docs.warp.dev, www.warp.dev, app.warp.dev and status.warp.dev allow every path read. status.warp.dev disallows /api/, so the incident JSON was not used and the Atom feed was read instead. pypi.org disallows /pypi/*/json, so the RSS feed was used. formulae.brew.sh lists only a sitemap

## Weaknesses

- The `warp` CLI has no print or headless mode. Scripts go through the legacy Oz CLI or the API, both of which need a Warp account
- Auto-approve bypasses the command denylist by default, and no local OS sandbox is documented for the CLI
- AI interactions and console inputs are collected by default on Free and on paid plans below Business, and the documented opt-out is in the desktop app
- The agent harness and Oz orchestration are proprietary and run server-side. Only the client is open source
- The install script checks no checksum, and no security.txt, bug bounty or SLA was found
- The terms forbid accessing the Services by robot, scraper or other automated means and using them to build a competitive product

## 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.

- For unattended work, call `POST https://app.warp.dev/api/v1/agent/runs` with `Authorization: Bearer $WARP_API_KEY`. The `warp` CLI is interactive only
- Use an agent key, scoped to one cloud agent and given an expiry, for CI rather than a personal key. Keys are created in the Oz web app or the Warp app
- Set `auto_approve_bypasses_command_denylist = false` under `[agents.warp_agent.other]` before starting `warp --auto-approve`
- Read `code` and `retryable` in error bodies. A 429 `resource_unavailable` means capacity is full and the platform retries the task itself
- Pin a version with `WARP_TUI_VERSION` at install, or set `general.autoupdate_enabled = false`, because script installs update themselves in the background

## What the review panel asked for

- breaking-change heading
- dated Oz CLI support window
- publish the agent-hour rate
- one name for the API
- documented Oz CLI output
- denylist holds under auto-approve
- secret redaction by default

## 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.