Improve LLM prompt templates for pact creation (bug / task / feature) #54
Description
Brief Description
What: Improve the three LLM prompt templates used to create pacts (bug, task, feature) so that generated pacts are consistently structured, include required metadata (priority, estimate, reproducible steps, acceptance criteria, labels, related files/paths, and suggested assignees), and are phrased in concise, actionable language suitable for both human review and automated downstream flows.
Why: Higher-quality, standardized pact outputs reduce manual cleanup, speed triage, and enable reliable automation (e.g., creating GitHub issues, linking conversations to work items, and meaningful search/retrieval). Better prompts will decrease false-positives/ambiguous pacts and improve cross-role handoffs between developers, PMs, and QA.
Context
Current state: The application already creates pacts from conversational context via server-side LLM flows (actions/ and orchestration services) and surfaces them in the project left-panel chatbot and pact lists. These prompts currently produce valid pacts but vary in detail, structure, and metadata completeness.
Gap: Prompts lack consistent templates and explicit instruction to populate fields that downstream systems (issue creation, linking, RAG retrievers) rely on. This causes extra manual edits, inconsistent labeling, and weaker traceability between chat discussions and tracked work.
Dependencies & constraints: Changes will touch the server-side LLM prompt templates and therefore affect the pact creation pipeline and any consumers (chat UI, persistence schema, GitHub-issue integration). Prompt updates should preserve existing pact schema compatibility or clearly document any required mapping changes.
Resources
Refer to the server-side pact creation flows and LLM prompt modules used in actions/ and the orchestration components that call them (server API/actions, background LLM worker).