Skip to content

0011 — Project documents in the database and markdown import

  • Status: Accepted (scope and order). A detailed plan and estimate are written before the step starts. Planned for 0.4.0, after screenshot annotations (0009) and before the runner (0012).
  • Size: to be estimated with the plan; it builds on the document storage and import of 0008.

Goal

The working documents of a project — spec, status, per-task notes and decisions — live in relaybay with versions and can be read and edited by agents through MCP, so a coordinator that today keeps them as markdown files can move over completely.

Scope

  1. Document kinds per project: SPEC (how it is built, with stable section anchors), STATUS (the owner's map: what runs where, branches in flight), per-task document (what was asked, decisions, work log), decisions. Each with versions (author, reason, base version, diff).
  2. Editing via MCP: read by section anchor, doc_update with optimistic locking (0008), links from tasks and reports to sections.
  3. UI: project → Docs & rules: view, history, diff, edit; a task's document in the task dialog.
  4. Import of markdown registries: an existing task registry table (project, task, stage, waiting on, next step, document) → tasks, with column aliases in any language, the leading stage word mapped through an alias map and the rest kept as the stage note; per-task markdown → task document; dry-run diff first, idempotent re-run.
  5. Moving a coordinator over: its instructions point at relaybay instead of files; the markdown folder becomes the export target.

Open before planning

  • Which document sections are structured (status table, decision list) and which stay free markdown.
  • How repository-owned documents (that must stay in a repository) are linked rather than imported.
  • Overlap with the memory step's document tables and import: reuse them, add only what is specific to project documents.

Risks

Risk Mitigation
Two sources of truth during the move Import is idempotent with dry run; export keeps file readers working
Formats of existing markdown vary too much Alias maps, import by heading names, unknown parts kept as free text
Lost edits from concurrent agents Optimistic locking with base version