Handoff: Codex Orchestrator Integration
Context: The Infrastructure (Codex) and Intelligence (Claude) layers are scaffolded and verified. Your mission is to bridge the gap and make the spec-agentic-source-orchestrator End-to-End Operational by implementing the remaining lifecycle tools.
1. Directives (Integration)
- Index & Embedding (
index_crawled_source):-
- Implement heading-aware chunking (using
02_System/chunker.pyas a reference). - Integrate with the Supabase
source_chunkstable. - Goal: Transform crawled Markdown into vector-ready chunks.
- Implement heading-aware chunking (using
-
- Retrieval Logic (
semantic_search_sources):-
- Implement the
match_documentsRPC call defined in spec-firecrawl-pgvector-pipeline. - Ensure results include metadata (source_url, title, heading) for Claude's provenance blocks.
- Implement the
-
- Integrity Audit (
verify_source_index):-
- Implement automated checks for chunk coherence and provenance link stability.
- Check for T3 (Stale Evidence) if re-indexing an existing source.
-
- Promotion Workflow (
promote_synthesis_candidate):-
- Handle the final transition from
Synthesized->Promoted. - Update the Postgres
source_pagesstatus. - Ensure the
provenanceblock is validated against the DB records before merging to01_Wiki/.
- Handle the final transition from
-
- Test Hardening:
-
- Update
02_System/test_vulture_ingest.pyto cover these new state transitions. - Mock the Supabase/Firecrawl network calls to ensure CI stability.
- Update
-
2. Dependencies
- spec-agentic-source-orchestrator (The Master Spec)
02_System/vulture-ingest/server.py(The tool surface)- synthesis-intelligence-layer (The Mind)
3. Next Seam
Once these four tools are implemented and mocked tests pass, the orchestrator is ready for its Inaugural Ingestion Run with live credentials.