Monitor Working Memory token budget with CT-rich memories

[michael-wojcik]

Context

PR #222 (v3.8.0) added CT-aware memory fields (reasoning_chains, agreements_reached, disagreements_resolved) to pact-memory. These add more content per memory entry.

Issue

CT-rich memories render more fields in Working Memory formatting (_format_memory_entry), which could consume the 800-token budget faster per entry, causing older entries to compress sooner.

Action

Monitor in practice whether CT-heavy memories are pushing the Working Memory budget. If so, consider:

• Adjusting the budget threshold
• Truncating CT field rendering when budget is tight
• Adding a token count warning when CT fields are populated

Source

Review finding from PR #222 (architect reviewer).