Enhance MCP development features and introduce skills management

- Added configuration options for requiring human approval before applying LLM-generated MCP patches.
- Updated Docker setup to include skills directory.
- Integrated skills management into the backend, allowing for procedural guides and skill matching.
- Refactored database initialization to apply Alembic migrations.
- Enhanced task approval process to handle MCP patch applications with optional approval.
- Introduced new schemas for skills and updated existing APIs to support skills functionality.

This commit lays the groundwork for improved agent capabilities and better management of MCP development processes.

backend/app/agent/graph.py

@@ -37,6 +37,7 @@ from app.services.device_defaults import resolve_slot_defaults
 from app.services.investigation_context import gather_investigation_context
 from app.services.approval_filter import split_proposed_fixes
 from app.services.troubleshooting_rules import rule_guidance_block
+from app.services.skill_registry import match_skills, skill_guidance_block
 from app.services.llm import triage_model, triage_model_info
 from app.services.llm_usage import LlmUsageTracker, merge_run_history
 from app.services.model_config import get_routing_config
@@ -75,6 +76,7 @@ class AgentState(TypedDict, total=False):
     prior_context: dict | None
     investigation_context: dict | None
     matched_rule: dict | None
+    matched_skills: list[dict] | None
     follow_up: str | None
     run_number: int
     prior_runs: list[dict]
@@ -348,6 +350,7 @@ async def triage_node(state: AgentState) -> AgentState:
         f"Onboard devices to check: {json.dumps(device_summary)}"
         f"{_investigation_context_block(state.get('investigation_context'))}"
         f"{rule_guidance_block(state.get('matched_rule'))}"
+        f"{skill_guidance_block(state.get('matched_skills'))}"
         f"{_prior_context_block(state.get('prior_context'), state.get('follow_up'))}"
     )
     triage_info = await triage_model_info()
@@ -449,6 +452,7 @@ async def reason_node(state: AgentState) -> AgentState:
         f"Title: {state['title']}\nIssue: {state['issue']}\nTriage: {json.dumps(triage)}\n\n{diag_text}"
         f"{_investigation_context_block(state.get('investigation_context'))}"
         f"{rule_guidance_block(state.get('matched_rule'))}"
+        f"{skill_guidance_block(state.get('matched_skills'))}"
         f"{_prior_context_block(state.get('prior_context'), state.get('follow_up'))}"
     )
     messages = [SystemMessage(content=prompt), HumanMessage(content=ctx)]
@@ -604,6 +608,10 @@ async def report_node(state: AgentState) -> AgentState:
         }
     if state.get("matched_rule"):
         report["matched_rule"] = state["matched_rule"]
+    if state.get("matched_skills"):
+        report["matched_skills"] = [
+            {k: v for k, v in s.items() if k != "body"} for s in state["matched_skills"]
+        ]
 
     report = merge_run_history(state.get("prior_runs") or [], run_number, report, llm_summary)
 
@@ -786,6 +794,7 @@ async def run_task_agent(task_id: int) -> None:
 
         devices: list[dict] = []
         matched_rule: dict | None = None
+        matched_skills: list[dict] = []
         vessel_name: str | None = None
         vessel_site: str | None = None
         vessel_public_ip: str | None = None
@@ -795,6 +804,18 @@ async def run_task_agent(task_id: int) -> None:
                 all_devices, routing_text, title=task.title
             )
             matched_rule = rule.as_dict() if rule else None
+            rule_id = matched_rule.get("id") if matched_rule else None
+            matched_skills = [
+                {
+                    "id": s.id,
+                    "name": s.name,
+                    "description": s.description,
+                    "body": s.body,
+                    "score": s.score,
+                    "priority": s.priority,
+                }
+                for s in match_skills(task.title, routing_text, matched_rule_id=rule_id)
+            ]
             vres = await db.execute(select(Vessel).where(Vessel.id == task.vessel_id))
             vessel = vres.scalar_one_or_none()
             if vessel:
@@ -807,6 +828,20 @@ async def run_task_agent(task_id: int) -> None:
             if d:
                 devices = [_device_dict(d)]
 
+        if not matched_skills:
+            rule_id = matched_rule.get("id") if matched_rule else None
+            matched_skills = [
+                {
+                    "id": s.id,
+                    "name": s.name,
+                    "description": s.description,
+                    "body": s.body,
+                    "score": s.score,
+                    "priority": s.priority,
+                }
+                for s in match_skills(task.title, routing_text, matched_rule_id=rule_id)
+            ]
+
         task.status = TaskStatus.running
         task.details = details
         await db.commit()
@@ -842,6 +877,14 @@ async def run_task_agent(task_id: int) -> None:
             f"Matched rule: {matched_rule.get('name')} → {', '.join(matched_rule.get('devices') or [])}",
             matched_rule,
         )
+    if matched_skills:
+        names = ", ".join(s["name"] for s in matched_skills)
+        await _emit(
+            task_id,
+            "skill",
+            f"Matched skills: {names}",
+            {"skills": [{k: v for k, v in s.items() if k != "body"} for s in matched_skills]},
+        )
 
     remaining = await balance.latest_remaining_usd()
     from app.config import settings
@@ -873,6 +916,7 @@ async def run_task_agent(task_id: int) -> None:
         "run_number": run_number,
         "prior_runs": prior_runs if continue_mode else [],
         "matched_rule": matched_rule,
+        "matched_skills": matched_skills or None,
     }
 
     try: