EOS

Review Workflow

This project supports staged editing and auditability:

  • Source (Live): canonical content (previously called “Approved” in the UI)

  • Review: human review draft

  • AI Review: agent-only staging area

  • Revisions: database snapshots for audit/rollback

Where data lives (DB columns)

The posts table stores draft payloads:

  • review_draft (JSON) — human review content

  • ai_review_draft (JSON) — agent staging content

Module data is stored separately:

  • module_instances.props (source/live)

  • module_instances.review_props / module_instances.ai_review_props (staged props for global/local modules)

  • post_modules.overrides (source/live per-post overrides)

  • post_modules.review_overrides / post_modules.ai_review_overrides (staged overrides)

  • Structural flags:

    • post_modules.review_added / review_deleted

    • post_modules.ai_review_added / ai_review_deleted

Key services/actions

  • Revisions: app/services/revision_service.ts

  • Post drafts + approvals: app/controllers/posts/posts_crud_controller.ts

  • Module ops: app/controllers/posts/posts_modules_controller.ts

  • Module actions:

    • app/actions/posts/add_module_to_post.ts

    • app/actions/posts/update_post_module.ts

Promotion flow (high level)

Source → Review

Editors save partial changes into review_draft (and module review_* fields).

AI Review (agents)

Agents must stage changes into ai_review_draft / ai_review_* fields.

Agents should not mutate Source or Review content directly.

AI Review → Review

When agents submit AI Review, the system:

  • merges AI Review into Review (draft + module staging)

  • clears AI Review staging

This ensures humans always approve “Review” content.

Review → Source

When a human approves Review in the admin UI:

  • Review draft content is promoted into source/live fields

  • module staging is promoted

  • taxonomy assignments are applied if provided (see MCP docs for staging behavior)

Revisions

Revisions are stored in post_revisions with a mode (e.g. review, ai-review). They are used for:

  • audit history

  • rollback / inspection

Developer notes

  • Keep “approval” logic in one place (promotion functions), so MCP + UI stay consistent.

  • Prefer deterministic merges (patches) over freeform replaces.