StatCraft Assistant Skill (Autonomous Code Mode)

This skill enables AI agents to control and inspect the active StatCraft desktop application session using an autonomous Code Mode (SPSS-syntax REPL).

1. Code Mode REPL Architecture

Instead of issuing dozens of slow, fragmented micro-tools, agents use batch SPSS syntax (statcraft_run_syntax) or direct tabular loading (statcraft_load_dataset).

• Instant Output Return: statcraft_run_syntax directly returns the ASCII/Unicode tabular output (Pivot Tables) of all generated procedures (ANOVA, Descriptives, Frequencies, T-Tests, Regression, Reliability) in the tool response! The agent immediately sees means, degrees of freedom, test statistics, and exact $p$-values in its context.
• Self-Correction & Dynamic Self-Documentation: The compiled StatCraft binary is the single source of truth:
  • Run HELP. to see the entire directory of supported categories and commands.
  • Run HELP <COMMAND>. (e.g. HELP ONEWAY., HELP T-TEST.) to get the exact syntax template and available subcommands.
  • Or inspect the dynamic MCP resource statcraft://syntax-reference.
  • If a command is misspelled, the syntax parser returns a "Did you mean ...?" suggestion.

2. Tools Reference

1. statcraft_run_syntax (Primary Batch Execution Engine)

Executes a batch SPSS syntax script (.sps) in StatCraft and returns formatted result tables.

• Parameters:
  • syntax (string, required): Full SPSS syntax script. Each command MUST end with a period (.).
  • timeout_seconds (integer, optional): Timeout in seconds (default: 45, max: 300).
• Universal Code Mode Directive:
  • Any core command can be directly invoked in syntax via: RUN COMMAND "<command_id>" { <json_params> }. Example: RUN COMMAND "transform.shift_values" { "source_var": "salary", "target_var": "salary_lag", "direction": "lag", "step": 1 }.
• Core Procedures:
  • Self-Documentation: HELP. or HELP <COMMAND>. (e.g. HELP ONEWAY.).
  • Variable Properties:
    • VARIABLE LEVEL varlist (SCALE | ORDINAL | NOMINAL).
    • VARIABLE LABELS var 'Descriptive Label'.
    • VALUE LABELS var 1 'Male' 2 'Female'.
    • MISSING VALUES var (99, 999).
  • Transformations: COMPUTE target = expr., SORT CASES BY var (A)., SPLIT FILE BY var., FILTER BY var., WEIGHT BY var.
  • Statistical Procedures: DESCRIPTIVES, FREQUENCIES, CROSSTABS, MEANS, T-TEST, ONEWAY, CORRELATIONS, REGRESSION, RELIABILITY, FACTOR.

2. statcraft_list_commands (Dynamic Introspection)

Inspects all available statistical procedures, transformations, and utilities compiled into StatCraft.

• Returns command IDs, menu hierarchy, syntax keywords, and complete JSON parameter schemas.
• Supports optional "category" filter (e.g. "Statistics", "Transform", "Data").
• Dynamic resource equivalent: read statcraft://commands for full Markdown catalog.

3. statcraft_execute_command (Universal Direct Command Execution)

Directly executes any StatCraft engine command by its command_id and JSON arguments.

• Automatically validates parameters according to the command's JSON schema.
• Records output into Output Viewer and returns structured JSON plus formatted output text.
• Can invoke any current or future procedure without code changes in the bridge!

4. statcraft_load_dataset

Directly loads raw tabular text (CSV, TSV) into the active session memory without writing temporary files to disk.

• Parameters: csv_content (string, required), name (string, optional).

5. statcraft_list_variables

Inspects the variable dictionary of the active dataset: names, types, measures, labels, value labels, and missing value rules.

6. statcraft_get_data_preview

Fetches the first $N$ rows (default 10) to inspect value distributions before running analysis.

• Parameters: limit (integer, optional).

7. statcraft_insert_output_note

Inserts an analytical card or APA-style conclusion into the StatCraft Output Viewer tree.

• Parameters: title (string, required), content (string, required).

8. statcraft_set_active_tab

Switches active view tab in StatCraft ("data", "variable", "output", "syntax").

3. Recommended Workflow

1. Inspect Data: Call statcraft_list_variables or statcraft_get_data_preview.
2. Consult Syntax if Needed: Run HELP <COMMAND>. via statcraft_run_syntax to verify available subcommands.
3. Execute Analysis in Batch:

   VARIABLE LEVEL salary (SCALE) /jobcat (NOMINAL).
   ONEWAY salary BY jobcat /STATISTICS DESCRIPTIVES HOMOGENEITY.

4. Interpret Output: Read the returned ASCII ANOVA and Descriptives tables directly from the tool response.
5. Summarize Findings: Insert an analytical note into the report tree via statcraft_insert_output_note.