Security pentest creates self-feeding task loop

[fazxes]

Problem

The security agent finds issues → creates tasks → evolve fixes them → next security scan finds more → creates more tasks. This loop never converges because each fix can introduce new patterns the security scanner flags.

Sessions 3-5 were dominated by this loop. The security agent created tasks 0208-0213, evolve fixed some, next session found more.

Fix

1. Security agent should compare findings against previous pentest reports and only flag NEW issues
2. After a security scan, the brain should NOT run another security scan for at least 5 sessions (let fixes stabilize)
3. Security tasks should be batched: one evolve session fixes all security findings, not one per session

Acceptance Criteria

• Security agent deduplicates against prior reports
• Minimum 5-session cooldown between security scans
• Security fix tasks are batched into single evolve sessions

[fazxes]

Converted to task #221