Harness² vs OpenAI Agents API

Harness² 36

0.1.0, first committed on 8 October 2026, with a README fix on 13 October (30). No tags, releases or dated changelog entries (0). 42 pull requests (41 from outside contributors, one from Dependabot) and 6 issues open on 18 October, against one pull request and no issues closed. No outside change is on main, which has three commits by the authors, and reply times couldn't be read (3 of 25). Not on PyPI and no SDK (0). No CI. Dependencies are pinned by version and upstream checkouts by commit, a ruff configuration is in pyproject.toml, and Dependabot's litellm bump of 8 October is still open (3 of 10). 36 in all.

OpenAI Agents API 87

Closed API. Last dated change 16 October 2026, Ultrafast mode and document skills (30). Four dated changelog entries in the 57 days since release, 10 September, 29 September, 9 October and 16 October (20). Changelog and help centre, as the openai-api listing reads them (12 of 15). Official SDKs carry beta.agents (15). SDK CI, per the openai-api listing (10). 30 + 20 + 12 + 15 + 10 = 87.

Harness² 44

Harness reading of the framework checklist, as for the other harnesses. 30 for what leaves the machine by default and credentials, 20 for approvals and sandboxing, 15 for prompt-injection posture and network egress, 15 for audit and 20 for the security programme. No telemetry found in the Harness² source, and the WorkBuddy OpenCode configs turn OpenCode's telemetry, auto-update and sharing off. Credentials are Google Application Default Credentials, and the Codex bridge binds 127.0.0.1 by default with a generated key stored with mode 0600. Against that, executors inherit the whole shell environment (the editor channel drops only five Anthropic variables), and no .gitignore keeps a filled-in .env out of commits (20). Codex runs with --dangerously-bypass-approvals-and-sandbox, the Claude CLI editor channel with --dangerously-skip-permissions, and the WorkBuddy OpenCode preset allows every tool, with only OpenCode's directory fence and WorkBuddy's containers as limits. The README and setup guide say so (4). Executor network access is unrestricted by default. The docs name the hosts to block and report a Codex rollout that fetched its rubric unprompted, but give no guidance on instructions inside task documents (5). Each editor call's prompt, system text, response, model and cost are saved to an audit folder, with trajectories and adopted harnesses under runs/ (13). No SECURITY.md or security.txt and no advisories, and the README says the project is outside Google's open-source vulnerability rewards programme (2). 44 in all.

OpenAI Agents API 87

Project keys with restricted per-endpoint permissions, service accounts and workload identity, per the openai-api listing; we didn't confirm a key can be limited to /v1/agents (27 of 30). Automatic review for approvals, a sandbox network restricted to a host allowlist, vaults for secrets and a sandbox security guide (17 of 20). The agent reads web pages, MCP results and browser content; the sandbox security guide and approval flow are the mitigations, with no prompt-injection guide of its own found (8 of 15). Tracing, observability and usage per session and admin audit logs (15). SOC 2 Type 2, ISO 27001 family, Bugcrowd bounty and a disclosure policy, per the openai-api listing (20). Not ZDR-eligible and US residency only, recorded under transparency. 27 + 17 + 8 + 15 + 20 = 87.

Harness² 20

Local-package reading. Not on PyPI (the release feed for harness2 answers 404). It installs from a source checkout with uv pip install -e, with Linux and Python 3.12 stated (the package allows 3.13), plus Node.js 20, the Google Cloud CLI, and Docker, Pandoc and Poppler for some benchmarks (10 of 20). No CI workflow in the repository. tests/test_workflow.py holds five unit tests over the deterministic demo, which needs no credentials (5 of 25). Six issues and 42 pull requests were open on 18 October, the pull requests mostly fixes by their titles, such as the missing .env.example, report failures, seeds lost between refinement and grading, and resume. None is merged and no issue is closed. Issue text and replies couldn't be read (5 of 25). No tags or changelog (0). Version 0.1.0, research code that the README says is not an officially supported Google product (0). 20 in all.

OpenAI Agents API 60

Read as a hosted API. status.openai.com has an Agents component under APIs (20). Since release on 10 September (57 days, not 90), the feed names the Agents API in a degraded-performance incident on 14 September, the elevated errors across ChatGPT, Codex and the API on 29 September (the openai-api listing reads that one as a major, 17:52 to 23:14 UTC), computer use issues and degraded workspace agents on 6 and 7 October, degraded performance across the APIs for 22 minutes on 21 October, and overbilling for hosted containers on 19 September. One major, the rest minor (10). Rate limits come from the shared guide, published per tier and model with headers, and the errors page adds a Files API allowance on session creation (15). 429 and 503 guidance with Retry-After, capped backoff and a check-then-retry procedure for turns (15). The Scale Tier SLA in the openai-api listing isn't shown to cover a beta API (0). Public beta behind the agents=v1 header (0). 20 + 10 + 15 + 15 + 0 + 0 = 60.

Harness² 51

Framework reading. Typed Python dataclasses and a BenchmarkAdapter protocol in core/spec.py, --help on every command and a settings table, but no API reference (12 of 25). Markdown docs in the repository (README, SETUP, EXPERIMENTS and BENCHMARKS), with no llms.txt (5 of 10). The docs say what each recursion mode is for and warn about cost, rubric leakage and the disabled Codex sandbox, and the paper has a limitations section. Nothing says how to use it outside the three benchmarks (13 of 20). --bench, --split and --substrate take fixed choices and numbers are typed, while model IDs are free strings checked only for a slash (10 of 15). Every command has a worked example and ConfigError messages name the fix, but the quick start's cp .env.example .env fails because the file isn't shipped (8 of 15). Version 0.1.0 with no changelog (3 of 15). 51 in all.

OpenAI Agents API 88

The reference is linked from the overview and the SDKs carry beta.agents types; we didn't confirm /agents/sessions in OpenAI's public OpenAPI file, so 15 of 25. llms.txt with every Agents API page and Markdown twins (10). The overview, architecture page and the agents guide that compares the Agents API, Agents SDK and Responses API say when to use it (20). Typed configuration with enumerated container sizes, network modes and environment types (13 of 15). Examples in six languages and an errors page with codes per HTTP status, turn and session (15). Dated changelog entries and a versioned beta header (15). 15 + 10 + 20 + 13 + 15 + 15 = 88.

Harness² 49

Apache-2.0 under Google LLC, with each bundled runner's licence and provenance file and a hash list. The WorkBuddy runners carry Tencent's custom licence, which has an EU territorial limitation (26 of 30). The docs list every host setup contacts and say credentials, generated configuration and run artefacts are not release assets. No privacy statement covers the package, and task data goes to the configured model provider (12 of 30). No deprecation policy or notices for the 0.1.0 research release (0). No telemetry found in the source, and the docs don't say either way. OpenCode's telemetry is turned off only in the WorkBuddy configs (10 of 20). 48 in all.

OpenAI Agents API 85

Closed service with clear terms, Apache-2.0 SDKs (15). Session state retained until deleted, deletable sessions and artifacts, not ZDR-eligible, US residency only, and the data-controls guide agrees; no retention period for artifacts after a sandbox expires (25 of 30). Dated deprecation notices with minimum periods for GA models; a beta API can change faster (15 of 20). Subprocessors and processing countries listed, per the openai-api listing (20). 15 + 25 + 15 + 20 = 75.

Harness² 59

Framework reading, adapted to a harness that a pipeline drives. Results land in run folders, not in the caller's context, and an agent reads task summaries, adopted harnesses and reflection notes from runs/ (12 of 25). --only-domains, --max-tasks (per domain), --split-file, --workers and --dry-run size a run before any model call (14 of 20). harness2-check tests dependencies, credentials and configuration without a model call, configuration errors name the remedy, and incomplete runs exit nonzero (14 of 20). Rerunning a command resumes it, saved configuration refuses an incompatible resume, failed grades stay unscored, and editor replies are cached by a fingerprint of their inputs (15 of 20). No model has a default, a Vertex AI project and a gcloud sign-in are needed, only three benchmarks are accepted and there's no SDK (4 of 15). 59 in all.

OpenAI Agents API 80

Read as an API an agent drives. Context is compacted automatically, and session budgets and streamed events keep output sized (20 of 25). Sessions can be listed, continued and steered, events streamed or sent by webhook, artifacts fetched (15 of 20). Actionable error codes with recovery steps (20). No idempotency key; a documented check-then-retry procedure (10 of 20). One session call with defaults (medium sandbox, network enabled), SDKs in five languages (15). 20 + 15 + 20 + 10 + 15 = 80.