When implementing complex features (5+ files, multiple steps):
Phase 1: Upload & Plan
- •Upload feature document:
rlm_upload_document(path="docs/features/...", content="...") - •Generate plan:
rlm_plan(query="Implement X", max_tokens=16000) - •Decompose:
rlm_decompose(query="Implement X", max_depth=2)
Phase 2: Chunk-by-Chunk Implementation For each chunk:
- •Query context:
rlm_context_query(query="chunk task", max_tokens=6000) - •Implement using RLM Runtime:
python
from rlm import RLM
rlm = RLM(backend="anthropic", environment="docker")
result = rlm.completion(f"""
Implement: {chunk_task}
Context: {snipara_context}
Steps:
1. Write implementation
2. Write tests
3. Run tests with pytest
4. Return results
""")
- •Remember decisions:
rlm_remember(type="decision", content="...")
Phase 3: Verification
python
rlm.completion("""
Run full test suite:
- pytest
- pnpm lint
- pnpm type-check
- pnpm build
""")
Benefits:
- •Context-aware: Each chunk gets relevant docs from Snipara
- •Safe execution: RLM Runtime uses Docker isolation
- •Iterative: Test each chunk independently
- •Traceable: Full trajectory logs for debugging
Example:
code
User: Implement OAuth 2.0 integration
Phase 1: Planning
- Upload spec → Plan → Decompose into 6 chunks
Phase 2: Implement chunks
Chunk 1: Database schema
→ rlm_context_query("OAuth database schema")
→ RLM: Write migration + tests + run
→ rlm_remember("decision", "Used JWT tokens table")
Chunk 2: OAuth provider config
→ rlm_context_query("OAuth configuration patterns")
→ RLM: Write config + env vars + tests
→ rlm_remember("decision", "Support Google + GitHub providers")
... repeat for remaining chunks ...
Phase 3: Integration test
→ RLM: Run full test suite