Skip to content

AI_Interface_Reference.md

Latest commit

 

History

History
120 lines (106 loc) · 6.81 KB

AI_Interface_Reference.md

File metadata and controls

120 lines (106 loc) · 6.81 KB

MDB AI Interface - Reference

Embed this in the agent's context. It is the operating manual for entering and editing column data in an MDB database. You work mostly with cell data; the only structural change you can make is adding a data column (addcol). You cannot delete columns, rename them, or create subtable/script columns - those stay in the app.

How you call it

Single in-process method: string AiCommandInterface.Execute(string command). You send one command line; you get one text result. Start every session with open.

Mental model: a folder tree of tables

  • The database is a set of top-level tables. Each table has columns and rows.
  • Some columns are subtables (a whole table nested inside each cell). You enter a subtable cell to see its own columns/rows - like opening a folder.
  • You are always "at" one level, shown by pwd as a dir string:
    • top level: TableName
    • inside a subtable cell: TableName/<row>,<column> (repeating for deeper levels), e.g. Ent Passive Effect/9,Attached Conditions And Properties/0,Passive Effect Properties.
  • You only ever see the columns/rows at your current level. Use row <i> to dump one row in full. Descend with cd, ascend with up, jump anywhere with goto.

Column flags (shown by cols)

  • PK - primary key (unique per table).
  • subtable - enter it with cd; do not set it.
  • script - an Auto Table Constructor Script column (holds a script; validated on set).
  • disabler-source - belongs to a mutually-exclusive group; filling one disables its partners. Header tags: READ-ONLY (cannot write), script-constructed (structure comes from a script), single-row (only one row allowed).

Influence notes (the (...) after a column in cols)

cols exposes "what affects what" so you can act in the right order:

  • references <Table> - a Foreign Key Refrence column; its valid values are that table's PKs (pks <col>).
  • structure from FK: <col> - a receiver subtable whose columns come from the script linked via that foreign-key column. Set the FK first, then the receiver's structure exists to fill.
  • exclusive with: <cols> - a disabler group. While any listed column in the row holds data, this cell is locked (and vice-versa). Clear the partner to enable this one.

row <i> marks any currently-locked cell as [DISABLED by: <col>], and a blocked set returns denied: cell is disabled by '<col>' (clear it first).

Commands

Navigate / read

  • open <path.mdb> - load a file (also load). Do this first.
  • tables - list tables.
  • usetable <name> - switch table.
  • pwd - current dir + structural key.
  • cols (ls) - columns here, with flags.
  • rows [a..b] - rows here (optional index range); shows a preview + filled-cell count.
  • row <i> - every column value of one row (use this instead of scanning all rows).
  • cell <r>,<col> - one cell value, untruncated.
  • find_r <pkval> - index of the row with that primary key (or "no pk column"). Cheaper than rows.
  • find <col>=<value> - indices of rows whose column contains value (case-insensitive).
  • finddeep <text> - recursively search the WHOLE current table (every subtable depth) for a string. Returns an indented directory tree: each row is <idx> [<pk>]/ (pk shown when the level has one), each subtable column is <name>/, and the matched column is |name| with its hits listed below as - [<rowidx>: <pk>] <value> (the : <pk> is omitted when the matched table has no primary key). Use to find everywhere a value appears.
  • col <name> - every value of one column across all rows here (a vertical slice).
  • pks <col> - the valid values for a Foreign Key Refrence column (PKs of its target table).
  • cd <r>,<col> - descend into a subtable cell.
  • goto <dir> - jump to a full dir path in one step (e.g. goto Table/9,Col/0,Col).
  • up - ascend one level. root - back to top of current table.

Edit (data only)

  • set <r>,<col> <value> - validated write.
  • script <r>,<col> <text> - set a script column (validated).
  • clear <r>,<col> - null a cell.
  • setrow <r> c=v; c=v; ... - write several cells in one row.
  • addrow - append an empty row here (respects row limits).
  • duperow <r> - append a deep copy of row r (subtables included). Any primary-key cell in the copy is cleared so you can give it a new unique value. Use for "same as row X but change Y".
  • addcol <name> <type> [refTable] - add a data column to the current (static) level and initialise it in every existing row. Types: t Text, n Numerical, i Integer, b Bool, pk Primary Key (only if the table has none), f <refTable> Foreign Key Refrence (give the table whose PKs it links to). Subtable/script columns cannot be added here - the user makes those in the app. Cannot be used on a script-constructed level.
  • delrow <r> (rmrow) - delete one row here; later rows shift down. (No bulk wipe - delete rows one by one.)

Column deletion is not available. There is no remove-column command (for safety). If the user asks to delete a column, tell them they must do it themselves in the app (right-click the column header). delcol/removecol just return that reminder.

Commit

  • diff - list pending changes.
  • save (commit) - write <name>-AI_edited.mdb. If the loaded file is already an -AI_edited file, it overwrites in place (compounding). Your original is never touched automatically; the user reviews the edited file and overwrites it themselves.
  • revert - discard changes (reload).

What gets rejected on write

Read-only tables; disabled cells (a partner in the disabler group has data); duplicate primary keys; foreign keys that are not a real PK of the target table (you get a "did you mean 'X'?" hint); non-numeric input to Numerical/Integer; bad scripts; rows beyond a row limit. A rejected write returns denied: <reason> and changes nothing.

Value rules

  • Numerical/Integer: digits only (Integer truncates). Bool: true/false (also yes/no/1/0).
  • Foreign Key Refrence: must equal an existing PK in the target table - call pks <col> first.
  • Empty value clears the cell (stores null).
  • You cannot set a subtable/receiver column - cd into it and edit its rows.

Efficient workflow (save context)

  1. open, then usetable.
  2. Locate a row with find_r <pk> or find <col>=<value> instead of dumping rows.
  3. Jump deep with one goto <dir> instead of several cds.
  4. Inspect with row <i> / cell.
  5. For a foreign key, pks <col> then set.
  6. diff to review, then save.

Example (real path)

open D:\...\DATABASE.mdb
usetable Ent Passive Effect
find_r Blindness                 -> row 9
goto Ent Passive Effect/9,Attached Conditions And Properties/0,Passive Effect Properties/0,Static Params
cell 0,status_value_key_directory
set 0,static_value 999
save                             -> DATABASE-AI_edited.mdb