Imported from aperger/teams-meeting-summarizer (AGENTS.md). Install upstream with npx skills add aperger/teams-meeting-summarizer. Copyright stays with the author.

AGENTS.md

Mission and Current State

• This repo targets a Teams live-meeting summarizer pipeline: ACS join -> audio ingest -> Whisper STT -> GDPR filter -> Ollama summary -> Angular UI (README.md, SKILLS.md).
• Current code is a scaffold, not the full pipeline: only app bootstrap + context test exist in app/src/main/java/com/summarizer/teams/TeamsSummarizerApplication.java and app/src/test/java/com/summarizer/teams/TeamsSummarizerApplicationTests.java.
• Treat SKILLS.md + DEVELOPMENT_PHASES.md as the executable roadmap; each phase has a matching prompt in .github/prompts/.

Architecture You Should Preserve

• Intended backend shape is layered and modular: controller -> service -> component, with services split by concern (.github/copilot-instructions.md).
• Planned transport boundaries are explicit: REST join/leave, WebSocket audio ingest, SSE summary fanout.
• Canonical endpoint contracts to keep stable:
  • POST /api/meetings/join, POST /api/meetings/{connectionId}/leave, POST /api/callbacks (SKILLS.md, phase2-acs-integration.prompt.md).
  • /ws/audio-stream for ACS media (phase3-audio-websocket.prompt.md).
  • GET /api/stream/summary/{connectionId} for SSE (phase7-sse-updates.prompt.md).
• GDPR rule is non-negotiable: filter transcript content before LLM summarization (README.md, .github/copilot-instructions.md, SKILLS.md).

Codebase-Specific Conventions (Important)

• Use Spring Boot 4.0.5 + Java 21 as already pinned in pom.xml; avoid Boot 3 APIs.
• Use jakarta.* (never javax.*) per .github/copilot-instructions.md.
• Prefer constructor injection; use records for DTO/value types; return ProblemDetail for API errors.
• Backend is now split into Maven modules:
  • app: Spring Boot entrypoint and web/API layer.
  • infra: external integration adapters.
  • model: shared domain records/DTOs.
• Use com.summarizer.teams as the base package for new code.

Developer Workflows That Matter

• Build and test with Maven wrapper from repo root:
  • ./mvnw clean compile
  • ./mvnw test
  • ./mvnw spring-boot:run
• Current application.yaml only sets app name; expected ACS/Whisper/Ollama/Azure config placeholders are documented but not yet added (DEVELOPMENT_PHASES.md, phase1-scaffold.prompt.md).
• External runtime dependencies assumed by design docs:
  • Whisper HTTP service (default shown as http://localhost:8080/inference).
  • Ollama API (http://localhost:11434).
  • Public callback/WebSocket URL for ACS during local dev (ngrok/dev tunnel in README.md).

How AI Agents Should Execute Work

• Implement by phase, one vertical slice at a time, using .github/prompts/phase*.prompt.md as task contracts.
• Ground decisions in existing docs first, then code reality: do not claim a feature exists unless implemented under src/main.
• Keep integrations mockable (ACS SDK, Whisper, Ollama) and add tests as each service appears (SKILLS.md acceptance criteria).
• Avoid hardcoded secrets; wire env-driven config keys before integration logic (README.md, .github/copilot-instructions.md).