AGENTS.md

Guidance for coding agents working on LMSSEP Vibe MCP Server. This server turns vague prompts into implementation-ready specifications through interactive refinement.

Project overview

The Vibe server analyzes prompt specificity, generates refinement steps, asks contextual questions, presents suggestions, records explicit user selections, tracks progress, and produces a final specification report.

Setup commands

bashcd <project-root>/genesis/LMSSEP/vibe
pip install -r requirements.txt
python3 main.py

Edit configs/vibe.py for enablement, minimum and maximum refinement steps, and specificity thresholds.

Tool workflow

1. Call vibe_refinement_initialize with the initial prompt.
2. Call vibe_refinement_get_next to receive the next question and suggestions.
3. Show all suggestions to the user.
4. Wait for the user's explicit selection.
5. Call vibe_refinement_submit only after the user chooses.
6. Repeat until complete.
7. Call vibe_refinement_report for the final specification.

Non-negotiable interaction rules

Do not auto-select the recommended option. Do not skip user interaction. Do not infer preferences when the server is asking for a real choice. is_recommended is advisory only; user selection is the contract.

Testing and verification

Verify initialize, get-next, status, submit, report, and list behavior with a synthetic prompt. Include one test path where the prompt starts vague and requires multiple steps, and one path where specificity is high and fewer steps are expected.

Security and privacy

Prompts can contain product strategy, credentials, or private requirements. Do not persist or publish raw prompts without user consent and redaction. Keep generated reports local unless they are explicitly meant to become project docs.

Commit and PR rules

Changes to suggestion ranking, step count, or report format must include an example before-and-after flow. Preserve the explicit user-selection rule.