feat(skill-learning): merge plugin skill updates

app-instance/backend/beaver/skills/learning/pipeline.py

@@ -35,6 +35,7 @@ class SkillLearningPipelineService:
         publisher: SkillPublisher,
         safety_checker: SkillDraftSafetyChecker | None = None,
         evaluator: SkillDraftEvaluator | None = None,
+        publish_observer: Callable[[SkillDraft, SkillVersion | SkillSpec], None] | None = None,
     ) -> None:
         self.learning_store = learning_store
         self.learning_service = learning_service
@@ -43,6 +44,7 @@ class SkillLearningPipelineService:
         self.publisher = publisher
         self.safety_checker = safety_checker or SkillDraftSafetyChecker()
         self.evaluator = evaluator
+        self.publish_observer = publish_observer
 
     def list_candidates(self, status: str | None = None) -> list[SkillLearningCandidate]:
         return self.learning_store.list_learning_candidates(status=status)
@@ -238,6 +240,16 @@ class SkillLearningPipelineService:
         else:
             result = self.publisher.publish(skill_name, draft_id, publisher=publisher, notes=notes)
         self._mark_candidate_by_draft(skill_name, draft_id, "published", "published")
+        if self.publish_observer is not None:
+            try:
+                self.publish_observer(draft, result)
+            except Exception as exc:  # noqa: BLE001 - observer is best effort after successful publish.
+                candidate = self._candidate_by_draft(skill_name, draft_id)
+                self.learning_store.append_audit_event(
+                    candidate.candidate_id if candidate is not None else f"draft:{draft_id}",
+                    "plugin_publish_ack_failed",
+                    {"error": str(exc), "skill_name": skill_name, "draft_id": draft_id},
+                )
         return result
 
     def rollback(
@@ -391,6 +403,14 @@ class SkillLearningPipelineService:
             preservation = eval_report.preservation_report or {}
             if preservation.get("passed") is False:
                 raise ValueError("Draft preservation check did not pass")
+            if draft.proposal_kind == "plugin_skill_update":
+                if draft.provenance.get("merge_mode") == "three_way" and preservation.get("mode") != "plugin_three_way":
+                    raise ValueError("Plugin update requires a three-way preservation report")
+                if preservation.get("unresolved_conflicts"):
+                    raise ValueError("Plugin update has unresolved merge conflicts")
+                supporting_plan = draft.provenance.get("supporting_file_plan")
+                if isinstance(supporting_plan, dict) and supporting_plan.get("conflicts"):
+                    raise ValueError("Plugin update has unresolved supporting-file conflicts")
 
     def _mark_candidate_by_draft(
         self,

app-instance/backend/beaver/skills/learning/preservation.py

@@ -32,6 +32,30 @@ def check_preservation(*, base_content: str, draft_content: str) -> dict[str, An
     }
 
 
+def check_plugin_merge_preservation(
+    *,
+    local_content: str,
+    upstream_content: str,
+    draft_content: str,
+    merge_decisions: dict[str, Any],
+) -> dict[str, Any]:
+    local = check_preservation(base_content=local_content, draft_content=draft_content)
+    upstream = check_preservation(base_content=upstream_content, draft_content=draft_content)
+    unresolved = [str(item) for item in merge_decisions.get("unresolved_conflicts") or []]
+    safety_sections_missing = _important_sections_missing(upstream, local)
+    passed = bool(local.get("passed")) and bool(upstream.get("passed")) and not unresolved and not safety_sections_missing
+    return {
+        "mode": "plugin_three_way",
+        "passed": passed,
+        "risk_level": "high" if not passed else "low",
+        "local": local,
+        "upstream": upstream,
+        "unresolved_conflicts": unresolved,
+        "safety_sections_missing": safety_sections_missing,
+        "resolved_conflicts": [str(item) for item in merge_decisions.get("resolved_conflicts") or []],
+    }
+
+
 def _sections(content: str) -> dict[str, str]:
     current = "body"
     sections: dict[str, list[str]] = {current: []}
@@ -51,3 +75,13 @@ def _sections(content: str) -> dict[str, str]:
 
 def _normalize(value: str) -> str:
     return re.sub(r"\s+", " ", value or "").strip().lower()
+
+
+def _important_sections_missing(*reports: dict[str, Any]) -> list[str]:
+    important = {"safety", "required tools", "required tool", "tools"}
+    missing: list[str] = []
+    for report in reports:
+        for section in report.get("dropped_sections") or []:
+            if str(section).strip().lower() in important and str(section) not in missing:
+                missing.append(str(section))
+    return missing

app-instance/backend/beaver/skills/learning/service.py

@@ -5,6 +5,7 @@ from __future__ import annotations
 from dataclasses import dataclass, field
 from datetime import datetime, timedelta, timezone
 from itertools import combinations
+from pathlib import Path
 import re
 from typing import Any
 from uuid import uuid4
@@ -14,9 +15,12 @@ from beaver.memory.runs.models import RunRecord, SkillEffectRecord
 from beaver.memory.runs.store import RunMemoryStore
 from beaver.memory.skills.models import SkillLearningCandidate, SkillPerformanceSnapshot
 from beaver.memory.skills.store import SkillLearningStore
+from beaver.plugins.hashing import hash_plugin_skill_tree
+from beaver.plugins.tree_merge import merge_supporting_file_trees
 from beaver.skills.drafts.service import DraftService
 from beaver.skills.learning.evidence import EvidencePacket, EvidenceSelector
 from beaver.skills.learning.synthesizer import SkillDraftSynthesizer
+from beaver.skills.catalog.utils import parse_frontmatter
 from beaver.skills.specs import SkillActivationReceipt
 
 
@@ -179,6 +183,8 @@ class SkillLearningService:
         candidate = candidates.get(candidate_id)
         if candidate is None:
             raise ValueError(f"Unknown learning candidate: {candidate_id}")
+        if candidate.kind == "plugin_skill_update":
+            return await self._synthesize_plugin_update(candidate, provider_bundle)
         if candidate.kind == "retire_skill":
             target_skill = candidate.related_skill_names[0]
             return self.draft_service.create_retire_proposal(
@@ -242,6 +248,85 @@ class SkillLearningService:
             evidence_refs=[{"run_id": item} for item in candidate.source_run_ids],
         )
 
+    async def _synthesize_plugin_update(self, candidate: SkillLearningCandidate, provider_bundle: ProviderBundle) -> Any:
+        evidence = dict(candidate.evidence)
+        skill_name = str(evidence.get("skill_name") or (candidate.related_skill_names[0] if candidate.related_skill_names else ""))
+        plugin_id = str(evidence.get("plugin_id") or "")
+        new_upstream_tree_hash = str(evidence.get("new_upstream_tree_hash") or "")
+        local_version = str(evidence.get("local_version") or "")
+        merge_mode = str(evidence.get("merge_mode") or "")
+        if not skill_name or not plugin_id or not new_upstream_tree_hash or not local_version:
+            raise ValueError("Plugin update candidate is missing required evidence references")
+        new_upstream = self.draft_service.store.read_upstream_snapshot(
+            skill_name,
+            plugin_id,
+            new_upstream_tree_hash,
+        )
+        if new_upstream is None:
+            raise ValueError("Plugin update references a missing upstream snapshot")
+        frontmatter, body = parse_frontmatter(new_upstream.content)
+        if merge_mode == "fast_forward":
+            return self.draft_service.create_plugin_update_draft(
+                skill_name=skill_name,
+                base_version=local_version,
+                proposed_content=body.strip(),
+                proposed_frontmatter=frontmatter,
+                created_by="learning-loop",
+                reason=candidate.reason,
+                provenance={
+                    **evidence,
+                    "proposal_kind": "plugin_skill_update",
+                },
+                evidence_refs=[],
+            )
+        base_upstream_tree_hash = str(evidence.get("base_upstream_tree_hash") or "")
+        old_upstream = self.draft_service.store.read_upstream_snapshot(skill_name, plugin_id, base_upstream_tree_hash)
+        current_local = self.draft_service.store.read_published_skill(skill_name, local_version)
+        if old_upstream is None:
+            raise ValueError("Plugin update references a missing base upstream snapshot")
+        if current_local is None:
+            raise ValueError("Plugin update references a missing local skill version")
+        packet = self.evidence_selector.build_evidence_packet(candidate.source_run_ids, candidate.source_session_ids)
+        provider = provider_bundle.auxiliary_provider or provider_bundle.main_provider
+        model = (
+            provider_bundle.auxiliary_runtime.model
+            if provider_bundle.auxiliary_runtime is not None
+            else provider_bundle.main_runtime.model
+        )
+        local_root = self.draft_service.store.root / skill_name / "versions" / local_version
+        file_plan = merge_supporting_file_trees(
+            base=_digest_map(old_upstream.root),
+            local=_digest_map(local_root),
+            upstream=_digest_map(new_upstream.root),
+        )
+        payload = await self.synthesizer.synthesize_plugin_update(
+            candidate,
+            packet,
+            provider,
+            model,
+            old_upstream={"content": old_upstream.content, "frontmatter": old_upstream.snapshot.frontmatter},
+            current_local={"content": current_local.content, "frontmatter": current_local.version.frontmatter},
+            new_upstream={"content": new_upstream.content, "frontmatter": frontmatter},
+        )
+        return self.draft_service.create_plugin_update_draft(
+            skill_name=skill_name,
+            base_version=local_version,
+            proposed_content=payload["content"],
+            proposed_frontmatter=payload["frontmatter"],
+            created_by="learning-loop",
+            reason=payload["change_reason"] or candidate.reason,
+            provenance={
+                **evidence,
+                "proposal_kind": "plugin_skill_update",
+                "preserved_local_sections": payload.get("preserved_local_sections", []),
+                "adopted_upstream_sections": payload.get("adopted_upstream_sections", []),
+                "resolved_conflicts": payload.get("resolved_conflicts", []),
+                "dropped_sections": payload.get("dropped_sections", []),
+                "supporting_file_plan": file_plan.to_dict(),
+            },
+            evidence_refs=[],
+        )
+
     def _base_skill_snapshot(self, skill_name: str, version: str | None) -> dict[str, Any] | None:
         loaded = self.draft_service.store.read_published_skill(skill_name, version)
         if loaded is None:
@@ -515,3 +600,16 @@ class SkillLearningService:
         if parsed.tzinfo is None:
             return parsed.replace(tzinfo=timezone.utc)
         return parsed.astimezone(timezone.utc)
+
+
+def _digest_map(root: Path) -> dict[str, dict[str, Any]]:
+    digest = hash_plugin_skill_tree(root)
+    return {
+        item.path: {
+            "content_hash": item.content_hash,
+            "executable": item.executable,
+            "size": item.size,
+        }
+        for item in digest.files
+        if item.path not in {"SKILL.md", "version.json", "upstream.json"}
+    }

app-instance/backend/beaver/skills/learning/synthesizer.py

@@ -41,6 +41,55 @@ class SkillDraftSynthesizer:
     ) -> dict[str, Any]:
         return await self._synthesize(candidate, evidence_packet, provider, model, "merge", base_skill=base_skill)
 
+    async def synthesize_plugin_update(
+        self,
+        candidate: SkillLearningCandidate,
+        evidence_packet: EvidencePacket,
+        provider: LLMProvider,
+        model: str,
+        *,
+        old_upstream: dict[str, Any],
+        current_local: dict[str, Any],
+        new_upstream: dict[str, Any],
+    ) -> dict[str, Any]:
+        prompt = self._build_plugin_update_prompt(
+            candidate,
+            evidence_packet,
+            old_upstream=old_upstream,
+            current_local=current_local,
+            new_upstream=new_upstream,
+        )
+        response = await provider.chat(
+            messages=[
+                {
+                    "role": "system",
+                    "content": (
+                        "You merge Beaver plugin skill updates. Return JSON only with keys: "
+                        "frontmatter, content, change_reason, preserved_local_sections, "
+                        "adopted_upstream_sections, resolved_conflicts, dropped_sections. "
+                        "Preserve valid local learning, adopt upstream fixes and safety changes, "
+                        "do not concatenate duplicate sections, and list every intentional drop."
+                    ),
+                },
+                {"role": "user", "content": prompt},
+            ],
+            tools=None,
+            model=model,
+            max_tokens=4096,
+            temperature=0,
+        )
+        payload = self._parse_plugin_update_payload(response.content or "")
+        if payload:
+            return payload
+        fallback = self._fallback_payload(candidate, evidence_packet, "plugin_update")
+        return {
+            **fallback,
+            "preserved_local_sections": [],
+            "adopted_upstream_sections": [],
+            "resolved_conflicts": [],
+            "dropped_sections": [],
+        }
+
     async def _synthesize(
         self,
         candidate: SkillLearningCandidate,
@@ -119,6 +168,28 @@ class SkillDraftSynthesizer:
             + "\nThe JSON may include preserved_sections, changed_sections, and dropped_sections arrays."
         )
 
+    @staticmethod
+    def _build_plugin_update_prompt(
+        candidate: SkillLearningCandidate,
+        evidence_packet: EvidencePacket,
+        *,
+        old_upstream: dict[str, Any],
+        current_local: dict[str, Any],
+        new_upstream: dict[str, Any],
+    ) -> str:
+        return (
+            f"Candidate kind: {candidate.kind}\n"
+            f"Reason: {candidate.reason}\n"
+            f"Task summaries:\n- " + "\n- ".join(evidence_packet.task_summaries or ["No historical run evidence."])
+            + "\n\nOLD UPSTREAM (merge base B):\n"
+            + str(old_upstream.get("content") or "")
+            + "\n\nCURRENT LOCAL (Beaver learned version L):\n"
+            + str(current_local.get("content") or "")
+            + "\n\nNEW UPSTREAM (plugin update U):\n"
+            + str(new_upstream.get("content") or "")
+            + "\n\nReturn JSON only. Preserve useful CURRENT LOCAL learning and adopt important NEW UPSTREAM changes."
+        )
+
     @staticmethod
     def _parse_payload(content: str) -> dict[str, Any]:
         cleaned = content.strip()
@@ -145,6 +216,33 @@ class SkillDraftSynthesizer:
             "dropped_sections": _coerce_string_list(payload.get("dropped_sections")),
         }
 
+    @staticmethod
+    def _parse_plugin_update_payload(content: str) -> dict[str, Any]:
+        cleaned = content.strip()
+        if cleaned.startswith("```"):
+            lines = cleaned.splitlines()
+            if len(lines) >= 3 and lines[0].startswith("```") and lines[-1].startswith("```"):
+                cleaned = "\n".join(lines[1:-1]).strip()
+        try:
+            payload = json.loads(cleaned)
+        except json.JSONDecodeError:
+            return {}
+        if not isinstance(payload, dict):
+            return {}
+        frontmatter = payload.get("frontmatter")
+        content_value = payload.get("content")
+        if not isinstance(frontmatter, dict) or not isinstance(content_value, str):
+            return {}
+        return {
+            "frontmatter": frontmatter,
+            "content": content_value.strip(),
+            "change_reason": str(payload.get("change_reason") or ""),
+            "preserved_local_sections": _coerce_string_list(payload.get("preserved_local_sections")),
+            "adopted_upstream_sections": _coerce_string_list(payload.get("adopted_upstream_sections")),
+            "resolved_conflicts": _coerce_string_list(payload.get("resolved_conflicts")),
+            "dropped_sections": _coerce_string_list(payload.get("dropped_sections")),
+        }
+
     @staticmethod
     def _normalize_payload(payload: dict[str, Any], evidence_packet: EvidencePacket) -> dict[str, Any]:
         frontmatter = normalize_skill_frontmatter(