prompt-builder

Deterministic prompt building executes a script that reads context and token budgets from your database, constructs complete prompts for agents like Developer or QA Expert, validates required markers, and writes the final prompt to a file before spawning. This ensures consistent, reproducible agent behavior across sessions.

Generated prompts include task context from databases, agent-specific specializations, per-model token budgets, required validation markers, and role definitions for Developer, QA, Tech Lead, PM, Investigator, and Requirements Engineer roles. The output includes token count, line count, and status metadata.

Yes. The prompt builder reads per-model token budgets from configuration, applies them during prompt construction, and validates that the final prompt respects those limits before saving and returning the result for spawning.

The prompt builder validates that all required markers are present during construction. If validation fails, the build process reports the status so you can correct the agent definitions or context before spawning.

Call the prompt builder script with your parameters JSON file specifying the agent role, context group, and output path. It returns a JSON result with token count and status; use this result to immediately spawn your agent with the generated prompt.

The builder reads parameters from a JSON file specifying agent role and context group, retrieves specializations and token budgets from your database, loads agent definitions from the filesystem, and combines them to construct the complete prompt.