diff --git a/backend/app/agents/schemas/skills.py b/backend/app/agents/schemas/skills.py
new file mode 100644
index 0000000..65f96d8
--- /dev/null
+++ b/backend/app/agents/schemas/skills.py
@@ -0,0 +1,38 @@
+from __future__ import annotations
+
+from datetime import datetime, timezone
+from typing import Any, Literal
+
+from pydantic import BaseModel, Field
+
+
+SkillStatus = Literal["candidate", "shadow", "active", "deprecated", "retired"]
+SkillInjectionMode = Literal["metadata_only", "summary", "full"]
+
+
+class SkillShortlistEntry(BaseModel):
+    skill_name: str
+    source: str = "runtime"
+    source_id: str | None = None
+    status: SkillStatus = "active"
+    scope: list[str] = Field(default_factory=list)
+    effectiveness: float | None = None
+    score: float = 0.0
+    rationale: str | None = None
+    summary: str | None = None
+    matched_terms: list[str] = Field(default_factory=list)
+    injection_mode: SkillInjectionMode = "metadata_only"
+    metadata: dict[str, Any] = Field(default_factory=dict)
+
+
+class SkillActivationRecord(BaseModel):
+    skill_name: str
+    source: str = "runtime"
+    source_id: str | None = None
+    status: SkillStatus = "active"
+    injection_mode: SkillInjectionMode = "metadata_only"
+    matched_terms: list[str] = Field(default_factory=list)
+    rationale: str | None = None
+    activated_at: datetime = Field(default_factory=lambda: datetime.now(timezone.utc))
+    outcome: str | None = None
+    metadata: dict[str, Any] = Field(default_factory=dict)
diff --git a/backend/app/agents/skills/__init__.py b/backend/app/agents/skills/__init__.py
index 56b8c07..f436148 100644
--- a/backend/app/agents/skills/__init__.py
+++ b/backend/app/agents/skills/__init__.py
@@ -1,16 +1 @@
-"""Skills 注册表 - Phase 9"""
-
-from app.agents.skills.registry import SkillRegistry, get_skill_registry
-from app.agents.skills.metadata import SkillMetadata
-from app.agents.skills.loaders.local_loader import LocalSkillLoader
-from app.agents.skills.loaders.plugin_loader import PluginSkillLoader
-from app.agents.skills.mcp_builder import MCPSkillBuilder
-
-__all__ = [
-    "SkillRegistry",
-    "SkillMetadata",
-    "LocalSkillLoader",
-    "PluginSkillLoader",
-    "MCPSkillBuilder",
-    "get_skill_registry",
-]
+"""Skill package."""
diff --git a/backend/app/agents/skills/effectiveness.py b/backend/app/agents/skills/effectiveness.py
new file mode 100644
index 0000000..4691cff
--- /dev/null
+++ b/backend/app/agents/skills/effectiveness.py
@@ -0,0 +1,14 @@
+from __future__ import annotations
+
+from app.models.skill import Skill
+
+
+def summarize_skill_effectiveness(skill: Skill) -> dict[str, object]:
+    return {
+        "name": skill.name,
+        "status": skill.status,
+        "effectiveness": skill.effectiveness,
+        "activation_count": skill.activation_count,
+        "candidate_count": getattr(skill, "candidate_count", 0),
+        "last_activated_at": skill.last_activated_at.isoformat() if skill.last_activated_at else None,
+    }
diff --git a/backend/app/agents/skills/evaluator.py b/backend/app/agents/skills/evaluator.py
new file mode 100644
index 0000000..3d2949e
--- /dev/null
+++ b/backend/app/agents/skills/evaluator.py
@@ -0,0 +1,58 @@
+from __future__ import annotations
+
+from datetime import UTC, datetime, timedelta
+
+from app.agents.schemas.learning import SessionRetrospective, SkillCandidate
+from app.agents.skills.models import SkillLifecycleDecision
+from app.services.skill_service import SkillService
+
+
+class SkillPromotionEvaluator:
+    def __init__(self, db):
+        self.db = db
+        self.skill_service = SkillService(db)
+
+    async def sync_retrospective(
+        self,
+        *,
+        user_id: str,
+        retrospective: SessionRetrospective,
+    ) -> list[SkillLifecycleDecision]:
+        decisions: list[SkillLifecycleDecision] = []
+
+        for candidate in retrospective.skill_candidates:
+            decisions.append(
+                await self.skill_service.upsert_learned_candidate(
+                    user_id=user_id,
+                    candidate=candidate,
+                    primary_agent=retrospective.primary_agent,
+                    evidence_refs=candidate.evidence_refs,
+                )
+            )
+
+        outcome_score = self._derive_outcome_score(retrospective)
+        for skill_name in retrospective.used_skill_names:
+            decision = await self.skill_service.record_activation_feedback(
+                user_id=user_id,
+                skill_name=skill_name,
+                outcome_score=outcome_score,
+                evidence_refs=retrospective.evidence_refs,
+            )
+            if decision is not None:
+                decisions.append(decision)
+
+        return decisions
+
+    @staticmethod
+    def _derive_outcome_score(retrospective: SessionRetrospective) -> float:
+        if retrospective.verification_status == "passed":
+            return 0.9
+        if retrospective.verification_status == "skipped":
+            return 0.55
+        if retrospective.verification_status == "failed":
+            return 0.15
+        return 0.7 if retrospective.outcome == "completed" else 0.2
+
+
+def next_review_after(days: int = 7) -> datetime:
+    return datetime.now(UTC) + timedelta(days=days)
diff --git a/backend/app/agents/skills/matcher.py b/backend/app/agents/skills/matcher.py
new file mode 100644
index 0000000..1ebdc38
--- /dev/null
+++ b/backend/app/agents/skills/matcher.py
@@ -0,0 +1,32 @@
+from __future__ import annotations
+
+import re
+
+
+def extract_match_terms(text: str | None) -> list[str]:
+    source = (text or "").lower()
+    terms = [token for token in re.findall(r"[a-z0-9_]+", source) if len(token) >= 3]
+
+    for chunk in re.findall(r"[\u4e00-\u9fff]+", text or ""):
+        if len(chunk) >= 2:
+            terms.append(chunk)
+            if len(chunk) > 4:
+                for index in range(len(chunk) - 1):
+                    terms.append(chunk[index : index + 2])
+
+    return list(dict.fromkeys(terms))
+
+
+def score_text_match(query_text: str, *corpus_parts: str | None) -> tuple[float, list[str]]:
+    query_terms = extract_match_terms(query_text)
+    if not query_terms:
+        return 0.0, []
+
+    corpus = " ".join(part for part in corpus_parts if part).lower()
+    matched_terms = [term for term in query_terms if term and term in corpus]
+    if not matched_terms:
+        return 0.0, []
+
+    coverage = len(matched_terms) / max(len(query_terms), 1)
+    density = min(len(matched_terms), 4) / 4
+    return round(min(1.0, coverage * 0.7 + density * 0.3), 3), matched_terms
diff --git a/backend/app/agents/skills/metadata.py b/backend/app/agents/skills/metadata.py
index 9414a3e..1482dfb 100644
--- a/backend/app/agents/skills/metadata.py
+++ b/backend/app/agents/skills/metadata.py
@@ -20,6 +20,10 @@ class SkillMetadata:
     source_id: str = ""  # 来源 ID
     enabled: bool = True  # 是否启用
     tools: list[str] = field(default_factory=list)  # 关联的工具
+    status: str = "active"  # candidate/shadow/active/deprecated/retired
+    scope: list[str] = field(default_factory=list)
+    effectiveness: float | None = None
+    review_after: str | None = None
 
     def to_dict(self) -> dict[str, Any]:
         return {
@@ -35,6 +39,10 @@ class SkillMetadata:
             "source_id": self.source_id,
             "enabled": self.enabled,
             "tools": self.tools,
+            "status": self.status,
+            "scope": self.scope,
+            "effectiveness": self.effectiveness,
+            "review_after": self.review_after,
         }
 
     @classmethod
diff --git a/backend/app/agents/skills/models.py b/backend/app/agents/skills/models.py
new file mode 100644
index 0000000..79d65df
--- /dev/null
+++ b/backend/app/agents/skills/models.py
@@ -0,0 +1,29 @@
+from __future__ import annotations
+
+from datetime import datetime
+from typing import Literal
+
+from pydantic import BaseModel, Field
+
+
+SkillLifecycleAction = Literal[
+    "created_candidate",
+    "promoted_to_shadow",
+    "promoted_to_active",
+    "degraded_to_deprecated",
+    "retired",
+    "reactivated",
+    "feedback_recorded",
+    "no_change",
+]
+
+
+class SkillLifecycleDecision(BaseModel):
+    skill_name: str
+    action: SkillLifecycleAction
+    previous_status: str | None = None
+    new_status: str
+    reason: str
+    evidence_refs: list[dict[str, object]] = Field(default_factory=list)
+    confidence: float | None = None
+    review_after: datetime | None = None
diff --git a/backend/app/agents/skills/policy.py b/backend/app/agents/skills/policy.py
new file mode 100644
index 0000000..c93a806
--- /dev/null
+++ b/backend/app/agents/skills/policy.py
@@ -0,0 +1,27 @@
+from __future__ import annotations
+
+from app.agents.schemas.skills import SkillInjectionMode, SkillShortlistEntry
+
+MAX_SUMMARY_CHARS = 120
+
+
+def choose_injection_mode(score: float, summary_available: bool) -> SkillInjectionMode:
+    if score >= 0.75 and summary_available:
+        return "summary"
+    return "metadata_only"
+
+
+def render_skill_shortlist_context(entries: list[SkillShortlistEntry]) -> str:
+    if not entries:
+        return ""
+
+    lines = ["[Task-Scoped Skills]"]
+    for entry in entries[:3]:
+        detail = entry.summary or "Relevant to the current request."
+        detail = detail[:MAX_SUMMARY_CHARS]
+        lines.append(f"- {entry.skill_name} | mode={entry.injection_mode} | score={entry.score:.2f}")
+        lines.append(f"  {detail}")
+        if entry.matched_terms:
+            lines.append(f"  matched_terms={', '.join(entry.matched_terms[:6])}")
+
+    return "\n".join(lines)
diff --git a/backend/app/agents/skills/retriever.py b/backend/app/agents/skills/retriever.py
new file mode 100644
index 0000000..7f52760
--- /dev/null
+++ b/backend/app/agents/skills/retriever.py
@@ -0,0 +1,153 @@
+from __future__ import annotations
+
+from collections import OrderedDict
+
+from app.agents.schemas.skills import SkillShortlistEntry
+from app.agents.skills.matcher import score_text_match
+from app.agents.skills.policy import choose_injection_mode, render_skill_shortlist_context
+from app.agents.skills.registry import get_skill_registry
+from app.services.skill_service import SkillService
+
+
+class RuntimeSkillRetriever:
+    def __init__(self, db):
+        self.db = db
+
+    async def shortlist(
+        self,
+        *,
+        user_id: str,
+        query_text: str,
+        memory_context: str | None = None,
+        retrospectives: list[dict] | None = None,
+        include_learned: bool = True,
+        limit: int = 3,
+    ) -> list[SkillShortlistEntry]:
+        deduped: "OrderedDict[str, SkillShortlistEntry]" = OrderedDict()
+        retrospective_text = "\n".join(
+            (item.get("summary") or item.get("summary_text") or "")
+            for item in (retrospectives or [])
+            if isinstance(item, dict)
+        )
+
+        service = SkillService(self.db)
+        for skill in await service.list_runtime_candidates(user_id, include_learned=include_learned):
+            score, matched_terms = score_text_match(
+                query_text,
+                skill.name,
+                skill.description,
+                skill.instructions,
+                retrospective_text,
+                memory_context,
+            )
+            if score <= 0:
+                continue
+            entry = SkillShortlistEntry(
+                skill_name=skill.name,
+                source="database",
+                source_id=skill.id,
+                scope=[skill.agent_type, skill.visibility],
+                status=skill.status,
+                effectiveness=skill.effectiveness,
+                score=score,
+                matched_terms=matched_terms,
+                rationale=(
+                    "Shadow skill matched current request; keep metadata-only injection."
+                    if skill.status == "shadow"
+                    else "Matched against DB skill metadata and instructions."
+                ),
+                summary=skill.description or (skill.instructions[:160] if skill.instructions else None),
+                injection_mode=(
+                    "metadata_only"
+                    if skill.status == "shadow"
+                    else choose_injection_mode(score, bool(skill.description or skill.instructions))
+                ),
+            )
+            self._upsert(deduped, entry)
+
+        registry = get_skill_registry()
+        if not registry.list_all():
+            try:
+                registry.load_all()
+            except Exception:
+                pass
+
+        for skill in registry.list_all():
+            score, matched_terms = score_text_match(
+                query_text,
+                skill.name,
+                skill.description,
+                " ".join(skill.tags),
+                " ".join(skill.triggers),
+                skill.content[:400],
+                retrospective_text,
+                memory_context,
+            )
+            if score <= 0:
+                continue
+            entry = SkillShortlistEntry(
+                skill_name=skill.name,
+                source=skill.source,
+                source_id=skill.source_id or skill.id,
+                scope=skill.scope or list(skill.tags),
+                status=skill.status,
+                effectiveness=skill.effectiveness,
+                score=score,
+                matched_terms=matched_terms,
+                rationale="Matched against local or external skill metadata.",
+                summary=skill.description or skill.content[:160],
+                injection_mode=choose_injection_mode(
+                    score,
+                    bool(skill.description or skill.content),
+                ),
+            )
+            self._upsert(deduped, entry)
+
+        return sorted(deduped.values(), key=lambda item: item.score, reverse=True)[:limit]
+
+    @staticmethod
+    def _upsert(
+        deduped: "OrderedDict[str, SkillShortlistEntry]",
+        entry: SkillShortlistEntry,
+    ) -> None:
+        existing = deduped.get(entry.skill_name)
+        if existing is None or existing.score < entry.score:
+            deduped[entry.skill_name] = entry
+
+
+def build_shortlisted_skill_context(
+    shortlist: list[dict] | list[SkillShortlistEntry] | None,
+    *,
+    agent_type: str | None = None,
+) -> str:
+    if not shortlist:
+        return ""
+
+    entries: list[SkillShortlistEntry] = []
+    for item in shortlist:
+        entry = item if isinstance(item, SkillShortlistEntry) else SkillShortlistEntry.model_validate(item)
+        if agent_type and entry.scope and agent_type not in entry.scope:
+            continue
+        entries.append(entry)
+
+    return render_skill_shortlist_context(entries)
+
+
+async def shortlist_skills_for_request(
+    db,
+    *,
+    user_id: str,
+    user_query: str,
+    memory_context: str | None = None,
+    retrospectives: list[dict] | None = None,
+    include_learned: bool = True,
+    limit: int = 3,
+) -> list[SkillShortlistEntry]:
+    return await RuntimeSkillRetriever(db).shortlist(
+        user_id=user_id,
+        query_text=user_query,
+        memory_context=memory_context,
+        retrospectives=retrospectives,
+        include_learned=include_learned,
+        limit=limit,
+    )
diff --git a/backend/app/schemas/skill.py b/backend/app/schemas/skill.py
index 0697afc..3431025 100644
--- a/backend/app/schemas/skill.py
+++ b/backend/app/schemas/skill.py
@@ -14,6 +14,10 @@ class SkillCreate(BaseModel):
     visibility: str = "private"
     team_id: Optional[str] = None
     is_active: bool = True
+    status: str = "active"
+    scope: list[str] = []
+    effectiveness: Optional[float] = None
+    review_after: Optional[datetime] = None
 
 
 class SkillUpdate(BaseModel):
@@ -28,6 +32,10 @@ class SkillUpdate(BaseModel):
     visibility: Optional[str] = None
     team_id: Optional[str] = None
     is_active: Optional[bool] = None
+    status: Optional[str] = None
+    scope: Optional[list[str]] = None
+    effectiveness: Optional[float] = None
+    review_after: Optional[datetime] = None
 
 
 class SkillOut(BaseModel):
@@ -43,6 +51,12 @@ class SkillOut(BaseModel):
     is_builtin: bool
     team_id: Optional[str]
     is_active: bool
+    status: str
+    scope: list[str]
+    effectiveness: Optional[float]
+    review_after: Optional[datetime]
+    activation_count: int
+    last_activated_at: Optional[datetime]
     owner_id: str
     created_at: datetime
     updated_at: datetime