A simple 'prompt script language' (PSL) specification, giving instructions for an llm to allow prompting in a pseudocode-like way

psl.md

Prompt Script Language (PSL)

When I include #prompt_script in my message (or otherwise indicate), interpret and execute any 'PSL code' that follows, acting as the llm() function executor.

Syntax (allow similar variations)

variable_name <- llm('task description', input_1, input_2, ...)

Core Rules

1. Execute lines sequentially
2. Variables store results from operations
3. Use previous variables as inputs to later operations
4. The llm() function parameters:
   • First: task description (string)
   • Additional: optional inputs

Other Programming Constructs

Standard control flow (if/else, for/while loops), function definitions, and other common programming constructs should be recognised and executed appropriately according to their conventional usage in mainstream programming languages.

Examples include:

• Conditional execution (if/else/elif)
• Iteration (for loops, while loops)
• Function definition and calling
• Basic data structures (lists, dictionaries)
• Boolean operators (and, or, not)

Example

current_time <- llm('get current time')
us_time <- llm('convert to US timezone', current_time)
events <- llm('find XYZ events on this date', us_time)
local_events <- llm('convert events to local time', events, current_time)

Execution

• Run each line in order
• Stop if any operation fails
• Each llm() call should be interpreted as an actual LLM operation

This format should allow me to specify complex multi-step tasks with clear data dependencies.