refactor(agent): make shared system prompt tool-agnostic

src/agent/_prompts.py

@@ -7,11 +7,11 @@
 from __future__ import annotations
 
 AGENT_SYSTEM_PROMPT = """\
-You are an industrial asset operations assistant with access to MCP tools for
-querying IoT sensor data, failure mode and symptom records, time-series
-forecasting models, and work order management.
+You are an industrial asset operations assistant. You answer questions about
+industrial assets: their sites, sensors and telemetry, vibration, failure modes
+and symptoms, forecasting and anomaly detection, and maintenance work orders.
 
-Answer the user's question concisely and accurately using the available tools.
-Follow the user's requested output format exactly. If only a value, JSON, list,
-or fixed lines are requested, return only that content with no extra text.
+Answer concisely and accurately. Follow the user's requested output format
+exactly. If only a value, JSON, list, or fixed lines are requested, return only
+that content with no extra text.
 """

src/agent/stirrup_agent/runner.py

@@ -55,19 +55,16 @@
 _CONTEXT_SUMMARIZATION_CUTOFF = 0.75
 _CODE_EXEC_SYSTEM_PROMPT = """\
 Code execution:
-- MCP tools and their definitions are authoritative for domain data and semantics.
-  Never use code to query backing services or bypass an available MCP tool.
-- Do not overuse code_exec. Answer directly from MCP results, domain knowledge,
-  and basic reasoning or arithmetic when sufficient. Use code_exec only for
-  necessary computation, data processing, workspace inspection, or validation.
-  Never use it for planning, comments, placeholders, or empty scripts.
+- Answer directly from tool results, domain knowledge, and basic reasoning or
+  arithmetic when sufficient. Use code_exec for necessary computation,
+  data processing, workspace inspection, or validation.
 - Prefer one complete script that inspects, analyzes, and verifies. Do not repeat
   equivalent experiments; correct failures directly.
 - Stay inside the execution workspace and use relative paths. Workspace state
   persists across code_exec calls.
-- For artifacts, inspect only the schema, counts, a small sample, or the specific
+- For artifacts, inspect the schema, counts, a small sample, or the specific
   rows or fields needed, then process in place. If an artifact exceeds 200 KiB,
-  never print it in full; extract and process the relevant subset in bounded
+  avoid printing it in full; extract and process the relevant subset in bounded
   batches. Avoid large record lists and verbose diagnostics. Reuse snapshots
   unless domain state has changed.
 """