[claude] Autonomous research pipeline — orchestrator + SOVEREIGNTY.md (#972) #1274

claude · 2026-03-24T01:40:33Z

claude commented

2026-03-24 01:40:33 +00:00

Fixes #972

What was built

Three deliverables for the governing research sovereignty spec:

`src/timmy/research.py` — ResearchOrchestrator

6-step autonomous pipeline:

Step 0: Semantic cache check (SQLite, instant, zero API cost)
Step 1: Research template loading from skills/research/
Step 2: Query formulation via Ollama slot-fill (5–15 queries)
Step 3: Web search via SerpAPI (graceful degradation if key absent)
Step 4: Full-page fetch via trafilatura (web_fetch)
Step 5: Synthesis cascade — Ollama (local) → Claude API fallback
Step 6: Store to semantic memory + optional disk persist to docs/research/

`tests/timmy/test_research.py`

24 unit tests covering all components. All passing.

`SOVEREIGNTY.md`

Machine-readable manifest at repo root: pipeline status table, cascade tiers, template index, sovereignty metrics targets.

What already existed (no changes needed)

skills/research/ — all 6 templates (tool_evaluation, architecture_spike, game_analysis, integration_guide, state_of_art, competitive_scan)
web_fetch tool — already in src/timmy/tools/system_tools.py with trafilatura

Cascade tiers implemented

Tier	Model	Status
4	SQLite semantic cache	✅ Active
3	Ollama `qwen3:14b`	✅ Active
2	Claude API (haiku)	✅ Active (opt-in, set `ANTHROPIC_API_KEY`)
1	Groq llama-3.3-70b	🔲 Planned (#980)

Refs #973, #974, #975

Fixes #972 ## What was built Three deliverables for the governing research sovereignty spec: ### `src/timmy/research.py` — ResearchOrchestrator 6-step autonomous pipeline: - **Step 0**: Semantic cache check (SQLite, instant, zero API cost) - **Step 1**: Research template loading from `skills/research/` - **Step 2**: Query formulation via Ollama slot-fill (5–15 queries) - **Step 3**: Web search via SerpAPI (graceful degradation if key absent) - **Step 4**: Full-page fetch via trafilatura (`web_fetch`) - **Step 5**: Synthesis cascade — Ollama (local) → Claude API fallback - **Step 6**: Store to semantic memory + optional disk persist to `docs/research/` ### `tests/timmy/test_research.py` 24 unit tests covering all components. All passing. ### `SOVEREIGNTY.md` Machine-readable manifest at repo root: pipeline status table, cascade tiers, template index, sovereignty metrics targets. ## What already existed (no changes needed) - `skills/research/` — all 6 templates (tool_evaluation, architecture_spike, game_analysis, integration_guide, state_of_art, competitive_scan) - `web_fetch` tool — already in `src/timmy/tools/system_tools.py` with trafilatura ## Cascade tiers implemented | Tier | Model | Status | |------|-------|--------| | 4 | SQLite semantic cache | ✅ Active | | 3 | Ollama `qwen3:14b` | ✅ Active | | 2 | Claude API (haiku) | ✅ Active (opt-in, set `ANTHROPIC_API_KEY`) | | 1 | Groq llama-3.3-70b | 🔲 Planned (#980) | Refs #973, #974, #975

claude added 1 commit 2026-03-24 01:40:34 +00:00

feat: implement autonomous research pipeline (#972 )

Tests / lint (pull_request) Failing after 35s

Details

Tests / test (pull_request) Has been skipped

Details

81ee0557d6

Closes three P0 items from the governing research sovereignty spec:

- `src/timmy/research.py` — ResearchOrchestrator (6-step pipeline):
  Step 0 semantic cache check (SQLite, instant, $0 cost)
  Step 1 research template loading from skills/research/
  Step 2 query formulation via Ollama slot-fill
  Step 3 web search via SerpAPI (graceful fallback when key absent)
  Step 4 full-page fetch via trafilatura (web_fetch)
  Step 5 synthesis via cascade (Ollama → Claude API fallback)
  Step 6 store to semantic memory + optional disk persist

- `tests/timmy/test_research.py` — 24 unit tests, all passing

- `SOVEREIGNTY.md` — machine-readable research independence manifest
  with pipeline status, cascade tiers, templates, and metrics targets

Refs #972 (governing spec), #973 (web_fetch), #974 (templates), #975 (orchestrator)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>