Harness² vs Verdent

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.

Verdent 61

Closed service. v2.18.0 on 29 October 2026, eight days before this check (30). More than three dated releases in the 90 days from 8 August, among them 2.10.0 to 2.18.0 (20). A dated public changelog, a support address, Discord and a subreddit, with no public tracker to read reply times (8). No official SDK or MCP server to keep current (0). No public packages or CI to judge (3). 30+20+8+0+3 = 61.

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.

Verdent 35

Harness reading. MCP servers per project or globally, subagents, skills and rules, but no per-tool disabling or deferred loading found (10). No output-size, turn-limit or budget controls for a run were found beyond credit balances (5). No exit codes or error format, since there is no headless mode (3). Parallel workers in git worktrees and a review column make runs reviewable and reversible, with no resume by ID documented (10). No CLI, API or SDK. A pipeline can only reach Manager through a paired Telegram, Slack or Feishu bot (7). 10+5+3+10+7 = 35.

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.

Verdent 43

Local-package reading, as for the other closed-source harnesses. The desktop app installs from verdent.ai with macOS 11 or later and 64-bit Windows 10 and 11 stated, and the extension from the editor marketplaces, but not from a package registry (15). No public CI or test suite, since the source isn't published (0). No public issue tracker, every task runs through Verdent's servers, and no status page was found, so there is no readable record (5). A dated changelog by version, with the end of Free Mode called out in v2.18.0, but no breaking-change sections (8). 2.18.0, generally available (15). 15+0+5+8+15 = 43.

Harness² 40

No payment protocol (0). Read with the self-hosted rule, as for the other harnesses. Free under Apache-2.0 with nothing to buy (20). The software needs no card, but every model-backed run in the docs bills a Google Cloud project on Vertex AI, and only the local demo runs without credentials (10). An agent can clone, install and run the demo and harness2-check alone, but a person has to set up the Vertex project and sign in with gcloud in a browser or supply a service account key (10). 40 in all.

Verdent 20

Harness reading of the published rubric, scored on the Verdent subscription the agent needs. No payment protocol (0). Plan prices from $5 a month, top-up prices and the credit value of about $0.059 are public without a login (20). Free Mode and the free trial ended on 29 October 2026, so there is no free tier (0). A person signs up in a browser (0). 0+20+0+0 = 20.

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.

Verdent 33

Framework reading (telemetry defaults, approvals, guardrails, sandboxing). Conversation history and tool results go to Verdent's US servers and on to Azure, Bedrock or Vertex, with zero data retention stated for Vertex and Bedrock only. Event logging to log.verdent.ai has no documented opt-out. BYOA keys stay on the machine (10). Plan Mode is read-only, Agent Mode asks before destructive commands and edits outside scope, and workers are isolated by git worktree, but there is no sandbox or network control and the approval list isn't enumerated (9). No prompt-injection guidance found, while web search and fetch results enter the context (2). Usage history in the dashboard, no audit log found (5). A security page with a disclosure address and a five-day acknowledgement, SOC 2 and ISO/IEC 42001 being pursued rather than held, no security.txt and no bug bounty (7). 10+9+2+5+7 = 33.

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.

Verdent 55

Proprietary, under terms from Verdent AI, Inc. that are clear on use and ownership (15). The privacy policy and security page agree that data stays in the US and that input isn't trained on without consent, and the security page lists subprocessors. The privacy policy gives no retention periods, names MiniMax as a provider the security list leaves out, and deletion is by email request (14). The credit bonus carries a promise of 15 days' notice, but Free Mode was retired on the day v2.18.0 shipped and no deprecation policy was found (6). Subprocessors and their locations are listed, and the domains the app calls are named, with event logging undescribed (12). 15+14+6+12 = 47.