chore: merge v0.1.11 release prep + strict ClawHub allowlist

CHANGELOG_SINCE_YESTERDAY.md

@@ -0,0 +1,37 @@
+# Changelog Since Yesterday's Submission
+
+Assuming yesterday’s ClawHub submission corresponds to commit `6942cbd` (release prep for v0.1.10).
+
+## Included commits (`6942cbd..4a58db3`)
+
+### 1) CI safety fix
+- `75f71c5` — **fix(ci): narrow low-signal filter to avoid dropping valid training pairs**
+
+### 2) Draft quality improvements
+- `fe80fd7` — **feat(quality): implement sender-type style anchors and exemplar cache**
+  - Adds sender-type style anchoring in draft prompting.
+  - Adds exemplar cache logic for more consistent draft selection.
+
+### 3) Persistence + onboarding defaults
+- `d336c5b` — **feat(YouOS): Implement persistent exemplar cache and quickstart default; fix syntax**
+  - Persists exemplar cache in SQLite (`exemplar_cache` table) across restarts.
+  - Wires cache read/write/clear through API + generation paths.
+  - Sets quickstart-first onboarding defaults.
+  - Fixes syntax issue in generation service.
+
+### 4) Dashboard metrics visibility
+- `4a58db3` — **feat(YouOS): Display edit-reduction metrics in dashboard**
+  - Surfaces edit-distance and high-rating deltas in stats UI.
+
+## Test status (targeted)
+- `.venv/bin/pytest -q tests/test_style_anchor_cache.py tests/test_config.py tests/test_setup_wizard.py tests/test_generation_improvements.py`
+- Result: **35 passed**
+
+## Files touched in this delta
+- `app/api/feedback_routes.py`
+- `app/api/stream_routes.py`
+- `app/db/bootstrap.py`
+- `app/generation/service.py`
+- `scripts/setup_wizard.py`
+- `templates/stats.html`
+- `youos_config.yaml`

CONTRIBUTING.md

@@ -46,7 +46,8 @@ Configuration is in `pyproject.toml` (line length 100, Python 3.11 target).
 3. Add tests for new functionality
 4. Ensure all tests pass: `python -m pytest tests/ -q`
 5. Ensure linting passes: `ruff check .`
-6. Submit a PR with a clear description of the change
+6. Run the release checklist in `docs/RELEASE_GUARDRAILS.md` for any user-facing/docs/submission change
+7. Submit a PR with a clear description of the change
 
 ## Architecture overview
 

PUBLISHING.md

@@ -5,12 +5,18 @@
 - GitHub account authenticated
 - All CI checks passing
 
-## Steps
-1. Bump version in clawhub.json and CHANGELOG.md
+## Steps (default release prep flow)
+1. Bump version in `clawhub.json`, `pyproject.toml`, and `CHANGELOG.md`
 2. Commit: `git commit -m "chore: bump version to X.Y.Z"`
 3. Tag: `git tag vX.Y.Z && git push origin vX.Y.Z`
-4. Publish: `clawhub publish ./`
-5. Verify on https://clawhub.com/skills/youos
+4. Build a **minimal allowlist release folder** (default):
+   - `./scripts/prepare_clawhub_release.sh`
+   - Optional custom output: `./scripts/prepare_clawhub_release.sh ~/Documents/youos-release-X.Y.Z`
+   - This script includes only: `app/`, `clawhub.json`, `configs/`, `PRIVACY.md`, `pyproject.toml`, `README.md`, `scripts/`, `SKILL.md`
+5. Publish from that folder (not repo root):
+   - `cd ~/Documents/youos-release-X.Y.Z`
+   - `clawhub publish ./`
+6. Verify on https://clawhub.com/skills/youos
 
 ## What clawhub publish does
 - Reads SKILL.md and clawhub.json

README.md

@@ -35,10 +35,12 @@ Gmail (sent mail)          Your feedback
 - Learns your writing style — richer persona: bullet point rate, directness score, sentence length, paragraph style; EWMA-weighted toward recent emails
 - Persona re-analysis is incremental (recent 90 days × 3 weight), with full weekly refresh; confidence intervals (p25/p75) shown in prompts
 - Per-sender-type personas: different voice, length, greeting, and closing for internal, external client, and personal contacts
+- Sender-type style anchors: explicit prompt slot (`[STYLE ANCHOR — internal|client|personal]`) to stabilize first-draft tone by audience
 - Per-account corpus isolation — drafts for work emails draw from work history; personal from personal
 - Greets people by first name, closes in your style — greeting and closing injected from persona config per contact type
 - Classifies multi-intent (meeting + urgent, etc.), boosts matching exemplars; per-intent reply length calibrated from corpus
 - Drafts grounded in score-ranked few-shot exemplars (confidence-annotated, thread-deduplicated); exemplar reply text preserved (600 chars), inbound trimmed (400)
+- Exemplar cache by intent+sender-type (TTL + feedback-triggered invalidation) improves consistency and reduces repeated ranking churn
 - Prompt token budget enforced — exemplars auto-trimmed if prompt exceeds 2000 tokens
 - Confidence thresholds are relative (mean±σ of retrieval scores), not hardcoded
 - Subject line + topic-aware retrieval; FTS queries expanded with email vocabulary synonyms

app/api/feedback_routes.py

@@ -15,7 +15,7 @@
 from app.core.facts_extractor import extract_and_save
 from app.core.rate_limit import RATE_LIMIT_RESPONSE, draft_limiter
 from app.db.bootstrap import resolve_sqlite_path
-from app.generation.service import DraftRequest, generate_draft
+from app.generation.service import DraftRequest, clear_exemplar_cache, generate_draft
 
 logger = logging.getLogger(__name__)
 
@@ -198,7 +198,7 @@ def feedback_generate(body: GenerateBody, request: Request) -> dict:
                 """INSERT INTO draft_history
                    (inbound_text, sender, generated_draft, confidence, model_used, retrieval_method)
                    VALUES (?, ?, ?, ?, ?, ?)""",
-                (body.inbound_text, body.sender, response.draft, response.confidence, response.model_used, response.retrieval_method),
+                (body.inbound_text, body.sender, response.draft, response.confidence, response.model_used, f"{response.retrieval_method}|cache_hit={int(response.exemplar_cache_hit)}|cache_key={response.exemplar_cache_key or ''}"),
             )
             conn.commit()
         finally:
@@ -213,6 +213,8 @@ def feedback_generate(body: GenerateBody, request: Request) -> dict:
         "confidence_reason": response.confidence_reason,
         "confidence_warning": response.confidence == "low",
         "suggested_subject": response.suggested_subject,
+        "exemplar_cache_hit": response.exemplar_cache_hit,
+        "exemplar_cache_key": response.exemplar_cache_key,
     }
 
 
@@ -273,9 +275,12 @@ def feedback_submit(body: SubmitBody, request: Request) -> dict:
                 quality_score = max(0.3, min(1.3, quality_score))
                 conn.execute("UPDATE reply_pairs SET quality_score = ? WHERE id = ?", (round(quality_score, 4), rp_id))
                 conn.commit()
+                # Invalidate exemplar cache on successful quality update
+                clear_exemplar_cache(database_url=request.app.state.settings.database_url)
         except Exception:
-            logger.warning("Failed to update quality_score for reply pair", exc_info=True)
+            logger.warning("Failed to update quality_score for reply pair or clear cache", exc_info=True)
 
+        clear_exemplar_cache(database_url=request.app.state.settings.database_url)
         total = conn.execute("SELECT COUNT(*) FROM feedback_pairs").fetchone()[0]
     finally:
         conn.close()

app/api/stream_routes.py

@@ -19,6 +19,10 @@
     _precedent_summary,
     _score_confidence,
     assemble_prompt,
+    _apply_cached_order,
+    _get_cached_exemplar_ids,
+    _top_exemplar_source_ids,
+    _update_exemplar_cache,
     generate_draft,
     lookup_sender_profile,
 )
@@ -59,6 +63,18 @@ def _stream_generate(body: StreamBody, settings):
     )
 
     reply_pairs = retrieval_response.reply_pairs
+
+    # Apply exemplar caching (read + write)
+    from app.core.intent import classify_intents_multi
+    intents = classify_intents_multi(clean_inbound)
+    detected_intent = intents[0]
+
+    cached_ids, exemplar_cache_hit, exemplar_cache_key = _get_cached_exemplar_ids(detected_intent, sender_type_hint, database_url=settings.database_url)
+    reply_pairs = _apply_cached_order(reply_pairs, cached_ids)
+
+    selected_ids = _top_exemplar_source_ids(reply_pairs)
+    _update_exemplar_cache(detected_intent, sender_type_hint, selected_ids, database_url=settings.database_url)
+
     confidence, _ = _score_confidence(reply_pairs)
     precedent_used = [_precedent_summary(rp) for rp in reply_pairs]
     detected_mode = retrieval_response.detected_mode
@@ -80,6 +96,7 @@ def _stream_generate(body: StreamBody, settings):
         detected_mode=detected_mode,
         tone_hint=body.tone_hint,
         sender_context=sender_context,
+        sender_type=sender_type_hint,
         user_prompt=body.user_prompt,
     )
 
@@ -117,7 +134,7 @@ def _stream_generate(body: StreamBody, settings):
         except Exception as exc:
             yield f"data: {json.dumps({'token': f'[generation failed: {exc}]'})}\n\n"
 
-    yield f"data: {json.dumps({'done': True, 'confidence': confidence, 'precedent_used': precedent_used})}\n\n"
+    yield f"data: {json.dumps({'done': True, 'confidence': confidence, 'precedent_used': precedent_used, 'exemplar_cache_hit': exemplar_cache_hit, 'exemplar_cache_key': exemplar_cache_key})}\n\n"
 
 
 @router.post("/stream")

app/core/config.py

@@ -130,6 +130,17 @@ def get_autoresearch_iterations(config: dict[str, Any] | None = None) -> int:
     return int(cfg.get("autoresearch", {}).get("iterations", 80))
 
 
+def get_persona_mode_config(sender_type: str, config: dict[str, Any] | None = None) -> dict[str, Any]:
+    cfg = config or load_config()
+    return cfg.get("persona", {}).get("modes", {}).get(sender_type, {})
+
+
+def get_persona_style_anchor(sender_type: str, config: dict[str, Any] | None = None) -> str | None:
+    cfg = config or load_config()
+    mode_config = get_persona_mode_config(sender_type, config)
+    return mode_config.get("style_anchor")
+
+
 def get_ollama_config(config: dict[str, Any] | None = None) -> dict[str, Any]:
     cfg = config or load_config()
     return cfg.get("model", {}).get("ollama", {})

app/db/bootstrap.py

@@ -27,6 +27,7 @@ def bootstrap_database() -> Path:
         _migrate_sender_profiles(connection)
         _migrate_memory(connection)
         _migrate_review_streaks(connection)
+        _migrate_exemplar_cache(connection)
         _populate_fts(connection)
         connection.commit()
     finally:
@@ -147,3 +148,15 @@ def _populate_fts(connection: sqlite3.Connection) -> None:
             )
         except Exception:
             pass
+
+
+def _migrate_exemplar_cache(connection: sqlite3.Connection) -> None:
+    """Create persistent exemplar cache table if it doesn't exist."""
+    connection.execute("""
+        CREATE TABLE IF NOT EXISTS exemplar_cache (
+            cache_key TEXT PRIMARY KEY,
+            source_ids_json TEXT NOT NULL,
+            updated_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP
+        )
+    """)
+    connection.execute("CREATE INDEX IF NOT EXISTS idx_exemplar_cache_updated ON exemplar_cache(updated_at)")