Context

PromptProfile.max_personality_tokens was introduced in #805 (capability-aware prompt profiles) as a validated field (gt=0) with per-tier values (large=500, medium=200, small=80). Currently it serves only as an ordering proxy -- no code reads the field to actually trim personality content.

The field is documented as "reserved for future token-based trimming" but the infrastructure is now in place to activate it.

Scope

Token-based personality trimming (engine/_prompt_helpers.py or engine/prompt.py)

• In build_core_context(), after assembling personality fields, estimate the token count of the personality section
• If the estimate exceeds profile.max_personality_tokens, progressively trim:
  1. Drop behavioral enum fields (risk_tolerance, creativity, verbosity, decision_making, collaboration, conflict_approach) -- these are already excluded by personality_mode != "full" but trimming adds a token-budget safety net
  2. Truncate personality_description to fit within the remaining budget
  3. If still over, fall back to communication_style only (minimal mode)
• Log the trimming action with before/after token counts

Integration with existing profile modes

• Token trimming should be a secondary control after personality_mode -- the mode selects which fields are included, the token limit enforces a hard cap on the combined result
• For the full profile (500 tokens), trimming should rarely activate -- it is a safety net for unusually verbose personality descriptions
• For basic profile (80 tokens), the minimal mode already excludes most fields, so the cap reinforces the structural reduction

Estimator reuse

• Reuse the existing PromptTokenEstimator (char/4 heuristic) or accept a configurable estimator
• The personality section is a small fraction of the total prompt, so precision matters less than consistency

Deliverables

• Token estimation for personality section content
• Progressive trimming logic respecting max_personality_tokens
• Logging of trim actions (event constants in events.prompt)
• Unit tests for trimming at each tier's limit
• Update PromptProfile.max_personality_tokens docstring to remove "reserved for future" caveat

References

• Parent: feat: capability-aware prompt profiles for model tier adaptation #805 (capability-aware prompt profiles)
• Field definition: src/synthorg/engine/prompt_profiles.py
• Personality rendering: src/synthorg/engine/_prompt_helpers.py:build_core_context()
• Token estimation: src/synthorg/engine/token_estimation.py