Skip to content

v0.20.0

Choose a tag to compare

@github-actions github-actions released this 11 Nov 11:06
8225ef1

Added

  • bca layers: the layered documentation, in sync with the source. The
    system explained from the whole down to each procedure: one page per app,
    component, namespace, and object, each with a picture and a sentence that
    says what it is; the procedures, triggers, and events of an object with
    what each reads, writes, and calls; and lenses, stories, and findings
    across them. Every page links one layer up and one layer down. See
    Layered documentation.

    • In sync. The same source gives the same files, byte for byte, so a
      run writes only what changed and removes what it no longer writes.
      --check writes nothing and fails when the pages are behind the
      source. layers is a task of bca run, with layers.outputDir,
      pictures, history, and view in .bca.json.
    • The source as it was. --at <commit, tag, or release> records the
      same pages for that state below history/, read from a temporary
      checkout; --releases records every release not recorded yet.
      history/README.md lists the states and what changed from each to the
      next: objects added, removed, and changed, relations, and findings new
      and fixed. A recorded state also holds view.html, the BC Atlas view
      of it to open in a browser.
  • BC Atlas AI, an optional built-in plugin: narrative documentation
    written by a language model, from facts.
    BC Atlas itself still calls no
    model. The plugin ai is the one place one is called, through one
    OpenAI-compatible endpoint, so a gateway routes to any model of any
    provider. See BC Atlas AI.

    • bca ai context prints the grounding of a subject (the system, or a
      process of the battle test): steps with what each needs, who performs
      or is blocked, requirements, what the recording of Business Central
      shows, findings with fixes. No source code. It calls no model.
    • bca ai write writes one page per subject next to its grounding, and
      again only when the grounding, the model, or the language changed, so a
      run without changes calls no model. --dry-run says what would be
      written.
    • Settings: ai.endpoint, ai.model, ai.outputDir, ai.language in
      .bca.json, or BCA_AI_ENDPOINT and BCA_AI_MODEL. The key comes
      from BCA_AI_API_KEY only, never from a file.
    • The ai task of bca run, and the MCP tools bc_atlas_ai_context and
      bc_atlas_ai_write.
    • Plugins get context.services.atlas(): the whole graph.
  • Define the simulation, from the view, the terminal, and an agent. The
    company of the battle test is changed by operations instead of by hand:
    add or change an agent, a role, a task group, or a requirement, remove
    one, ground a process test on a recording. Each change runs the simulation
    at once, and what is not valid never reaches .bca.json. See
    Define the simulation.

    • Simulate, a menu next to Story in the toolbar of the view: the
      company, its agents and roles, and a form for each; Define… on the
      card of a person, a role, or a requirement. In VS Code a change is
      written and the graph shows the new result; an exported file saves the
      changes as simulation-changes.json.
    • bca simulate define: --agent, --role, --task-group,
      --requirement with their fields, --remove, --ground, --init,
      --from, --dry-run. Without options it shows the simulation.
    • bc_atlas_simulation_define, an MCP tool that takes the operations
      and returns the definition and the result of the run.
  • The Simulation canvas. A new lens shows the company as one picture:
    people, their roles, the process tests each role does or must not do, who
    is blocked, the requirements the tests prove, and the recordings they
    stand on. Roles are nodes of the graph.

  • Grounding. A process test stands on the recording of the same process
    in Business Central (by title, or simulation.grounding). Where the test
    and the recording go through different objects is the finding
    grounding-differs; a named recording that is not there is
    grounding-missing.

Fixed

  • The 3D form is in space. It looked like the 2D graph seen from the
    side. Now the eye stands closer, so what is in front is clearly larger;
    nodes are balls with light on them; relations behind are fainter and
    those in front stronger; the graph stands over a floor of rings and
    spokes and turns slowly until you turn it, point at it, or select a
    node. Pictures of bca view --form 3d show the same.