FULL Mode Workflow (Snipara + RLM Runtime)

Starting FULL mode for complex development (token budget: ~8-15K).

Best for:

• Multi-day features
• Architectural changes
• 5+ files affected
• Team documentation needed
• Safe code execution required

Workflow:

Phase 1: Context Loading

1. Load team standards: rlm_shared_context()
2. Recall previous context: rlm_recall("$ARGUMENTS progress")
3. Query comprehensive context: rlm_context_query("$ARGUMENTS", max_tokens=8000)

Phase 2: Planning 4. Generate plan: rlm_plan("$ARGUMENTS") 5. Decompose: rlm_decompose("$ARGUMENTS", max_depth=2) 6. Upload spec (if exists): rlm_upload_document(path="docs/features/...", content="...")

Phase 3: Implementation (Chunk-by-Chunk with RLM) For each chunk: 7. Query context: rlm_context_query(query="chunk task", max_tokens=6000) 8. Execute with RLM Runtime:

rlm run --env docker "
    Implement: {chunk_task}
    Context: {snipara_context}
    Steps:
    1. Write implementation
    2. Write tests
    3. Run tests
    4. Return results
"

9. Remember decisions: rlm_remember(type="decision", content="...")

Phase 4: Verification 10. Run full test suite with RLM:

rlm run --env docker "
    Run full test suite:
    - pytest
    - pnpm lint
    - pnpm type-check
    - SKIP_ENV_VALIDATION=true pnpm build
"

Phase 5: Documentation 11. Upload implementation: rlm_upload_document() 12. Store summary: rlm_store_summary()

Phase 6: Session Save 13. Remember progress: rlm_remember(type="context", content="Completed X, next: Y")

View Execution:

• Logs: rlm logs
• Trajectory: rlm visualize

Let me start Phase 1 for: $ARGUMENTS