docs(llms): setup agents documentation

docs/agent-guides/change-assessment.md

@@ -0,0 +1,30 @@
+# Change assessment
+
+Use before non-trivial planning/implementation and during non-trivial review. Let it shape scope and verification; avoid process theatre.
+
+## Rate three dimensions
+
+- **Value** — concrete user, product, operational, or maintenance benefit; include urgency.
+- **Risk** — regression likelihood/impact: data loss, security, cloud incompatibility, disruption, hard rollback.
+- **Complexity** — concepts, dependencies, layers, states, verification surfaces; not line count.
+
+Use `Low`, `Medium`, or `High`. Give one evidence-based sentence each. No pseudo-precise scores.
+
+```markdown
+## Assessment
+
+- Value: High — <concrete benefit>
+- Risk: Medium — <failure modes and reversibility>
+- Complexity: Low — <conceptual and verification burden>
+- Recommendation: Proceed | Reshape | Defer — <why>
+```
+
+One assessment per overall change.
+
+- High value never excuses unmanaged risk/complexity.
+- High risk needs earlier proof, narrower increments, rollback, or stronger checks.
+- High complexity needs clearer boundaries and smaller steps, not automatic abstraction.
+- Low value plus high risk/complexity suggests reshape or defer.
+- Revise after material scope discovery.
+
+Plans: assessment before steps. Reviews: findings first, then assessment. Assessment informs; never replaces user intent or evidence.