docs(core): add anti-slop guardrails

- Add anti-slop guardrails section for reviews
- Explicitly call out AI slop in code generation non-goals

Codifies “write less, say more” guidance to reduce generic boilerplate and keep changes anchored to evidence and scope.

Author: d-padmanabhan

rules/100-core.mdc

@@ -126,6 +126,31 @@ When providing feedback, organize by impact:
 4.  **Suggest incremental refactoring** over big-bang rewrites.
 5.  **Preserve existing code** - no placeholders like `// ... (existing code)` or incomplete sections.
 
+### Anti-slop guardrails (write less, say more)
+
+**Default behavior:** Prefer the smallest change that solves the stated problem. If a change, paragraph, or abstraction doesn’t clearly earn its keep, don’t add it.
+
+**Banned patterns (AI slop):**
+- Generic filler (e.g., “Great question”, “Hope this helps”, long restatements of the prompt)
+- Un-grounded “best practices” lists not tied to this repo/task
+- Speculative features, refactors, or “nice-to-haves” not requested
+- Placeholder code, TODO-only stubs, or incomplete sections
+- Repeating the same point in multiple sections
+
+**Required behaviors:**
+- **Be anchored:** Tie recommendations to concrete evidence (file/line references, errors, command output, or a specific risk)
+- **Be minimal:** Provide the smallest diff that fixes the issue; avoid “big-bang rewrites”
+- **Be specific:** Replace vague advice with an actionable change (“Change X to Y because Z”)
+- **Be honest about uncertainty:** Mark assumptions explicitly; do not invent details
+- **Prefer deletion over addition** when it improves clarity without breaking behavior
+
+**Slop check (run before replying):**
+- Can I delete 30% of this response and keep the value?
+- Did I propose any unrequested work? If yes, move it to “Next ideas” or drop it
+- Did I provide at least one concrete action (diff, command, file/line)?
+- Did I avoid generic “best practices” unless they directly apply here?
+- Did I keep examples to the minimum needed to be runnable?
+
 ### Advanced Review Patterns
 
 **Security-First Review:**
@@ -453,6 +478,7 @@ def process_order(order_id: int) -> Order:
   * Extensive test suites (provide 1–2 examples instead).
   * Complex frameworks when stdlib suffices.
   * Features not in requirements.
+  * Generic boilerplate text or “best practices” not tied to the task (“AI slop”).
 
 ### Assumption Format