prompt-template-authoring

Reusable prompt templates are created as markdown files with YAML frontmatter containing a name-aligned description, which automatically register as slash commands for a Pi extension. This turns hand-written instructions into shareable slash-command workflows.

Yes, model selection in prompt workflows supports explicit model choices, fallback order for reliability, and rotation to cycle through model choices during execution. This model-aware execution ensures flexible LLM handling.

Deterministic pre-LLM steps run commands or scripts before the LLM turn using configurable handoff rules. This enables deterministic execution and flow control within loops and chains before generating prompts.

Best-of-N compare workflows cycle through model choices to generate multiple outputs for iterative or compare-based prompting tasks. This workflow orchestration pattern enables direct comparison of model responses for quality selection.

Argument substitution uses $@, $1, $2, and ${@:1}-style placeholders to generate prompts from runtime inputs. These placeholders allow dynamic prompt generation when executing slash commands.

Template validation rejects certain runtime flags for deterministic prompts to maintain execution reliability. This constraint ensures deterministic pre-steps and flow control behave predictably without unexpected runtime overrides.