About the project
The spark
Most hackathon projects sprint from idea to spec, then stall at the missing WHY. Stakeholders want market sizing, ROI, risks, and a crisp executive narrative—fast. Vibe PM Agent fills that gap as Kiro’s why-layer, turning raw ideas into evidence-backed business cases and board-ready one-pagers inside the same workflow where specs and code already live.
What it does
- Runs as a Kiro-native MCP server.
- Analyzes opportunities (market, competition, risks, assumptions) with citations.
- Generates business cases with multi-scenario ROI and sensitivity notes.
- Produces executive outputs (one-pager / PR-FAQ) that are ready for review.
- Saves artifacts to predictable folders (
reports/,.kiro/steering/,examples/spec-augmentation/) for easy auditing.
How we built it
- Architecture: Node/TypeScript MCP server exposing focused tools (
analyze_business_opportunity,generate_business_case,create_stakeholder_communication, etc.). Stdio transport for easy Kiro integration. - Kiro integration:
- Emits steering files to keep conversations in “PM mode.”
- Produces spec-augmentation snippets Kiro can ingest in spec-to-code flows.
- Structures outputs to be hook-friendly (e.g., trigger on PR labels).
- Emits steering files to keep conversations in “PM mode.”
- Evidence & outputs: Each major claim includes citations, a lightweight credibility tag, and a confidence score. Outputs are JSON/Markdown for reproducibility and diffs.
- DX:
npm ci && npm run build && npm testfor sanity; demo scripts reproduce the full flow in under three minutes.
What we learned
- Steering > prompts: Reusable steering reduced drift and made outputs reviewable by non-authors.
- Rationale is first-class: Pairing specs with explicit assumptions/risks/trade-offs improves both engineering clarity and the business case.
- Citations change behavior: Requiring sources cut vague statements and increased stakeholder trust.
- Reproducibility wins demos: Structured artifacts committed to the repo turn subjective “looks good” into objective diffs.
Challenges we faced
- Grounding & hallucination control: Designed a simple credibility rubric and limited citation age to reduce noise.
- Latency vs. depth: Split market scans into smaller tools and streamed partials to keep UX responsive.
- Spec alignment: Iterated to ensure augmentation helped engineering (scope/acceptance criteria) rather than adding fluff.
- Deterministic runs: Standardized outputs, added basic tests, and fixed key parameters so judges can re-run and see comparable results.
Why it matters
Great teams don’t just ship features—they make better bets. By embedding market/ROI reasoning alongside specs—directly in Kiro—Vibe PM Agent shortens the path from “promising idea” to “greenlight,” leaving auditable artifacts leadership can trust.
