fix: address 5 CodeRabbit review findings

Aureliolo · Aureliolo · commit de91d028bbf0 · 2026-04-04T09:27:33.000+02:00
- Fix docs/design/engine.md: standard profile Acceptance Criteria
  column 'Bullet list' -&gt; 'Nested list' (simplify_acceptance_criteria=False)
- Add tier 1/2 budget-met invariant to PersonalityTrimInfo validator
  (after_tokens must not exceed max_tokens for tiers 1 and 2; only
  tier 3 is best-effort)
- Include '## Personality' heading in _estimate_personality_tokens so
  budget_met reflects actual rendered section cost
- Add task_id to PROMPT_PERSONALITY_TRIMMED log in agent_engine for
  task correlation
- Propagate trimming_enabled through _trim_sections so re-renders
  honor the original personality_trimming_enabled flag instead of
  defaulting to True
diff --git a/docs/design/engine.md b/docs/design/engine.md
@@ -711,7 +711,7 @@ each model tier.
 | Profile    | Tier   | Personality          | Max Personality Tokens | Org Policies | Acceptance Criteria | Autonomy |
 |------------|--------|----------------------|------------------------|--------------|---------------------|----------|
 | **full**   | large  | Full behavioral enums | 500                   | Included     | Nested list         | Full     |
-| **standard** | medium | Description + style + traits | 200              | Included     | Bullet list         | Summary  |
+| **standard** | medium | Description + style + traits | 200              | Included     | Nested list         | Summary  |
 | **basic**  | small  | Style keyword only   | 80                     | Excluded     | Flat semicolon line | Minimal  |
 
 ### Personality Trimming
diff --git a/src/synthorg/engine/_prompt_helpers.py b/src/synthorg/engine/_prompt_helpers.py
@@ -71,6 +71,12 @@ def _check_cross_field_invariants(self) -> Self:
         if self.after_tokens > self.before_tokens:
             msg = "after_tokens must not exceed before_tokens"
             raise ValueError(msg)
+        if self.trim_tier in {1, 2} and self.after_tokens > self.max_tokens:
+            msg = (
+                "after_tokens must not exceed max_tokens for "
+                f"trim_tier {self.trim_tier} (only tier 3 is best-effort)"
+            )
+            raise ValueError(msg)
         return self
 
     @computed_field  # type: ignore[prop-decorator]
@@ -151,9 +157,9 @@ def _estimate_personality_tokens(
     """Estimate token count of the personality section as the template renders it.
 
     Assembles the text that the Jinja2 template would produce for the
-    given *personality_mode*, including inline markdown formatting
-    (bold labels, list prefixes), and runs it through the estimator.
-    The section heading and template whitespace are not included.
+    given *personality_mode*, including the section heading and inline
+    markdown formatting (bold labels, list prefixes), and runs it
+    through the estimator.
 
     Args:
         ctx: Template context dict with personality fields populated.
@@ -163,7 +169,7 @@ def _estimate_personality_tokens(
     Returns:
         Estimated token count.
     """
-    parts: list[str] = []
+    parts: list[str] = ["## Personality"]
     desc = ctx.get("personality_description", "")
     style = ctx.get("communication_style", "")
 
diff --git a/src/synthorg/engine/agent_engine.py b/src/synthorg/engine/agent_engine.py
@@ -892,6 +892,7 @@ async def _prepare_context(  # noqa: PLR0913
                 PROMPT_PERSONALITY_TRIMMED,
                 agent_id=agent_id,
                 agent_name=identity.name,
+                task_id=task_id,
                 before_tokens=ti.before_tokens,
                 after_tokens=ti.after_tokens,
                 max_tokens=ti.max_tokens,
diff --git a/src/synthorg/engine/prompt.py b/src/synthorg/engine/prompt.py
@@ -482,6 +482,7 @@ def _trim_sections(  # noqa: PLR0913
     context_budget: str | None = None,
     currency: str = DEFAULT_CURRENCY,
     profile: PromptProfile | None = None,
+    trimming_enabled: bool = True,
 ) -> tuple[
     str,
     int,
@@ -510,6 +511,7 @@ def _trim_sections(  # noqa: PLR0913
             context_budget=context_budget,
             currency=currency,
             profile=profile,
+            trimming_enabled=trimming_enabled,
         )
         if estimated <= max_tokens:
             break
@@ -543,6 +545,7 @@ def _trim_sections(  # noqa: PLR0913
             context_budget=context_budget,
             currency=currency,
             profile=profile,
+            trimming_enabled=trimming_enabled,
         )
 
     _log_trim_results(agent, max_tokens, estimated, trimmed_sections)
@@ -623,6 +626,7 @@ def _render_with_trimming(  # noqa: PLR0913
             context_budget=context_budget_indicator,
             currency=currency,
             profile=profile,
+            trimming_enabled=trimming_enabled,
         )
 
     return _build_prompt_result(
