4.0 KiB
4.0 KiB
description
| description |
|---|
| Converts conversational brain dumps into structured, metric-driven Markdown tickets in the ./tickets directory. |
WORKFLOW: PRAGMATIC ARCHITECT TICKET GENERATOR
1. High-Level Goal
Transform informal user "brain dumps" into high-precision, metric-driven engineering tickets stored as Markdown files in the ./tickets/ directory. The workflow enforces a quality gate via targeted inquiry before any file persistence occurs, ensuring all tasks are observable, measurable, and actionable.
2. Assumptions & Clarifications
- Assumptions: The agent has write access to the
./tickets/and/temp/directories. The current date is accessible for naming conventions. "Metrics" refer to quantifiable constraints (latency, line counts, status codes). - Ambiguities: If the user provides a second brain dump while a ticket is in progress, the agent will prioritize the current workflow until completion or explicit cancellation.
3. Stage Breakdown
Stage 1: Discovery & Quality Gate
- Stage Name: Requirement Analysis
- Purpose: Analyze input for vagueness and enforce the "Quality Gate" by extracting metrics.
- Inputs: Raw user brain dump (text).
- Actions: 1. Identify "Known Unknowns" (vague terms like "fast," "better," "clean"). 2. Formulate exactly three (3) targeted questions to convert vague goals into comparable metrics. 3. Check for logical inconsistencies in the request.
- Outputs: Three questions presented to the user.
- Persistence Strategy: Save the original brain dump and the three questions to
/temp/pending_ticket_state.json.
Stage 2: Drafting & Refinement
- Stage Name: Ticket Drafting
- Purpose: Synthesize the original dump and user answers into a structured Markdown draft.
- Inputs: User responses to the three questions;
/temp/pending_ticket_state.json. - Actions: 1. Construct a Markdown draft using the provided template.
2. Generate a slug-based filename:
YYYYMMDD-slug.md. 3. Present the draft and filename to the user for review. - Outputs: Formatted Markdown text and suggested filename displayed in the chat.
- Persistence Strategy: Update
/temp/pending_ticket_state.jsonwith the full Markdown content and the proposed filename.
Stage 3: Execution & Persistence
- Stage Name: Finalization
- Purpose: Commit the approved ticket to the permanent
./tickets/directory. - Inputs: User confirmation (e.g., "Go," "Approved");
/temp/pending_ticket_state.json. - Actions: 1. Write the finalized Markdown content to
./tickets/[filename]. 2. Delete the temporary state file in/temp/. - Outputs: Confirmation message containing the relative path to the new file.
- Persistence Strategy: Permanent write to
./tickets/.
4. Data & File Contracts
- State File:
/temp/pending_ticket_state.json- Schema:
{ "original_input": string, "questions": string[], "answers": string[], "draft_content": string, "filename": string, "step": integer }
- Schema:
- Output File:
./tickets/YYYYMMDD-[slug].md- Format: Markdown
- Sections:
# Title,## Context,## Acceptance Criteria,## Suggested Affected Files,## Technical Constraints.
5. Failure & Recovery Handling
- Incomplete Inputs: If the user fails to answer the 3 questions, the agent must politely restate that metrics are required for high-precision engineering and repeat the questions.
- Inconsistencies: If the user’s answers contradict the original dump, the agent must flag the contradiction and ask for a tie-break before drafting.
- Missing Directory: If
./tickets/does not exist during Stage 3, the agent must attempt to create it before writing the file.
6. Final Deliverable Specification
- Format: A valid Markdown file in the
./tickets/folder. - Quality Bar:
- Zero fluff in the Context section.
- All Acceptance Criteria must be binary (pass/fail) or metric-based.
- Filename must strictly follow
YYYYMMDD-slug.md(e.g.,20240520-auth-refactor.md). - No "Status" or "Priority" fields.