feat(task-023): improve generated context quality

context/agent-guidelines.md

@@ -23,6 +23,40 @@ Agents should favour many small successful changes over large risky changes.
 
 ---
 
+# Objectivity
+
+- Do not praise, flatter, or agree unnecessarily.
+- Do not repeat information the user already knows.
+- State problems directly without hedging or softening.
+- Provide factual assessments; separate observation from opinion.
+
+---
+
+# Context Discipline
+
+- Read each file at most once unless its content changed between reads.
+- Do not re-read context files that were read earlier in the same session unless explicitly asked.
+- Track which files you have already read in your working notes rather than reloading them.
+
+---
+
+# Communication Efficiency
+
+- Do not restate the task or requirements before acting.
+- Skip summarising instructions the user has already given.
+- Provide results, not process descriptions, unless asked for a rationale.
+- Use concise technical language; avoid filler phrases ("sure", "great", "I can help with that").
+
+---
+
+# Execution Efficiency
+
+- Act on the task immediately after understanding it.
+- Avoid unnecessary planning or confirmation steps when the next action is clear.
+- Report only what changed, not everything you considered.
+
+---
+
 # Before Starting Work
 
 Agents should first read: