CLI
Script review loops that create drafts, fetch comments, upload revisions, and report completion from local tooling.
CLI workflow docsCreate human review workflows through the Commentary CLI, HTTP API, MCP tools, coding-agent skills, Forms APIs, Knowledge Brain APIs, and the Live Preview SDK.
Use explicit review ids, files, comments, form answers, permissions, Brain health context, evaluations, and structured output so automation can participate in human-owned review loops.
Use canonical HTTP, MCP and Agent SDK operations for optional Projects, typed Trackers, immutable review packages, exact Sessions, source and evidence. CLI 0.2.0, Agent SDK 0.2.0 and Tracker SDK 0.1.0 are tested candidates; public package availability and CP-039/040 release qualification remain pending.
Keep execution reports, independently observed evidence and human Decisions separate. Agents request review and revise artifacts; eligible humans decide the exact scope. Durable events support bounded reconnection and a current-state resync after a gap.
Check current server discovery and consented scopes before enabling a client. Project links never grant original source access. Custom renderers use an isolated SDK broker without app credentials; native fallback stays available.
Pick the surface that fits the tool you are wiring into Commentary, then link deeper reference detail from docs, OpenAPI, or the MCP endpoint.
Script review loops that create drafts, fetch comments, upload revisions, and report completion from local tooling.
CLI workflow docsUse OpenAPI-backed routes for review Resources, organized work, durable human requests, exact receipts and attributed fulfillment. Agent access never grants human Decision authority.
OpenAPI JSONConnect agents to authenticated tools for reviews, native comments, Interactions, Projects, Trackers, packages, exact Sessions, source and evidence. Check current discovery and scoped consent.
MCP endpointValidate form contracts, create draft review form sources, submit answers, list permitted result collections, read embedded answers, and inspect canonical git result sync.
Forms overviewList Brain reviews, changed files, comments, health findings, requested revisions, ready-for-review replies, and evaluation ledgers.
Brain overviewInstrument opted-in preview apps so UI element comments carry route, selector, and viewport context.
Review SDK overviewBuild framework-neutral custom views through an isolated capability broker. The authoritative Tracker stays on the server; bundles receive no app credentials and native fallback remains available.
Tracker candidate overviewForms APIs and MCP tools keep structured answers tied to review source, permissions, and result ownership.
Developer access is scoped to Commentary review automation, and provider-owned content still depends on the viewer's connected provider permissions.
Automation works best when Commentary passes explicit review records and receives structured output that people can verify.
Human reviewers leave comments on rendered files, plan sections, or selected UI elements.
API or MCP clients read explicit review ids, files, comments, permissions, and state.
Agents return structured replies, revision uploads, or status updates for people to verify.
Agents should receive review ids, files, comments, permitted form answers, permissions, and allowed actions directly from Commentary instead of scraping a rendered page.
Create developer credentials, read the docs, and connect the review surface that matches your workflow.