Repository navigation
Conversation
|
Outcome: changes-requested Policy assessment @xqliu The complete first-submission diff supports agent-native / orchestrators / open-source-framework. This is not just a CLI-agent launcher: the actual Python runner persists run/base/head identity across issue claim, isolated implementation, independent exact-head review, tests/coverage, same-run remediation and guarded merge/blocked states. Official workflow documentation and the concrete issue 1367 -> merged PR 1370 delivery substantiate that outcome. Local/remote-runner interfaces, operator/model prerequisites and the bounded verification level are supportable. No published or open-PR duplicate was found. Please address both remaining items together:
Required fields, taxonomy and direct schema/provenance inspection otherwise pass. The contradiction pass covered website, docs, runner source/README, package/release metadata, AGPL/SUL and separate managed-cloud surfaces. Core self-hosted open-source pricing is not a promise of free model usage or free managed-cloud service. Evidence checked
Evidence requiring correction or removal
Please correct or remove each evidence item above. Checks run Exact final head 5efbffd merged without committing with current main d5a6bbd in an isolated temporary clone:
|
|
Thanks for the review, @foo-bender. Both items are addressed in 0dd5a55:
I also merged current |
|
Outcome: meets-policy Policy assessment Incremental re-review: both previously requested changes are resolved. websiteUrl is the canonical https://orbi.build/; the LICENSE evidence now claims only the AGPL text actually present. Settled inclusion/category analysis remains valid: agent-native / orchestrators / software-application. This is a substantive runner with isolated execution, recorded test evidence, independent exact-head review and guarded merge states, not a CLI launcher. Current core/managed-cloud, licence and release surfaces introduce no unresolved material contradiction. Operator/model prerequisites and documented macOS limitations remain disclosed. Product fit
Evidence checked
All submitted evidence items were checked at their exact URLs without substituting redirected/canonical sources. The independent contradiction pass covered the official product site, docs, repository/README and implementation, release/package channels, and applicable pricing/licensing/linked first-party surfaces. Verification remains documentation-reviewed; no live agent integration, security benchmark or production execution was exercised. Reviewer-owned SEO metadata tools/orbi.md
Reviewer commit: 2e3ee60 Checks run Final head 2e3ee60 merged without committing with current main 6fe956f in an isolated detached worktree:
@bradvin This meets the directory policy and is ready for your inclusion decision. |
Adds
tools/orbi.md(categoryorchestrators, classificationagent-native, pricingopen-source) and the submitter entry intool-submitters.jsongenerated withnpm run sync:tool-submitters.Orbi takes an Issue labeled
ai-ready, runs a coding agent in a worktree created from a frozen base SHA, opens a PR, has a separate review session check and fix it, and merges only through a merge gate (reviewed head, latest base, concluded CI, mergeable). For the coding-orchestrator boundary, the evidence covers:All evidence is first-party (repo, docs, Issue, PR, release, LICENSE), accessed 2026-10-21. I left out
reviewedBy/reviewedAt/publishedAt/contentModifiedAt.Checks run locally on Node 24:
npm test(139 pass, 0 fail, 1 skipped) andnpm run validate:content -- --require-submitters(passed: 16 categories, 66 tools).Disclosure: I am the maintainer of Orbi.